Founder & CEO, Valiotti Analytics | Better Data Analytics for Data-Driven Companies

Founders don’t lack data.
They lack confidence to make decisions with it.

I work as a Fractional Chief Data Officer for scaling tech companies, when analytics becomes business-critical — but hiring a full-time CDO is too early or too risky.

I’m the founder of Valiotti Analytics, a data consultancy trusted by 50+ teams globally. I work with companies across the US and Europe.

I have a PhD in data science, which helps when problems are complex — but my work is always grounded in business outcomes.

What I typically change:
• Data & analytics strategy aligned with business goals
• Clear metrics and decision frameworks
• Analytics org design (people, tools, processes)
• Systems that teams actually use, not just maintain

Before consulting, I was an in-house operator. I’ve lived through messy data, slow reports, and dashboards nobody trusts — and I know why “just build another report” never works.

Since 2019, I’ve helped marketing, sales, and product teams:
→ Scale revenue up to 25x
→ Eliminate manual reporting and data chaos
→ Turn analytics into a decision-making system, not a cost center

I usually work in a fractional format (3–12 months), directly with CEOs, CPOs, and CMOs — with optional execution support from my team.

Nikolay Valiotti

Data engineering is never “done.”

If you think it is, you’re already behind.

Data stacks don’t age gracefully.
Pipelines break. Schemas drift. APIs get “versioned” into oblivion. One quiet column rename, and your executive dashboard goes dark five minutes before a board meeting.

That’s not incompetence. That’s actually normal. The mistake is treating data engineering like a one-off build instead of a living system.
The job isn’t to design something flawless and walk away. It is to stay close enough to catch problems before the business does.

Strong data teams understand this:
→ Failure is inevitable — speed of detection is the real metric
→ Models will drift — adaptability matters more than elegance
→ Requirements will change — velocity beats perfection
This is why “set it and forget it” always turns into “why does no one trust the data?”

If you want data you can rely on, you need three things:
1 — Monitoring that screams the moment something breaks
2 — Engineers who can trace, debug, and patch quickly
3 — A habit of revisiting data contracts when sources evolve

Data is infrastructure. And no serious business builds infrastructure once and never inspects it again.
—
♻️ Repost if you’ve ever been paged because a silent schema change wiped out a revenue dashboard.
Follow me Nick for more unfiltered takes on data strategy, engineering, and keeping systems alive long after launch.

Nick Valiotti

Fractional CDO | Helping Scaling Tech founders turn data into faster decisions | Founder @ Valiotti Analytics

One thing I have to say out loud:
Excel is incredible. And this is exactly the problem with it.

The better Excel works, the longer companies cling to it. The longer they cling, the harder it becomes to outgrow.

Every team hits the same phase:
→ one spreadsheet
→ turns into twelve
→ turns into forty
→ turns into a folder full of excels

At that point, no one knows what’s real anymore.
This isn’t a tooling issue. It’s infrastructure debt.
And it compounds quietly — until decisions start disagreeing with each other.

In my book, Your Fractional CDO, I break down why spreadsheets become dangerous at scale — and what to fix before dashboards, KPIs, and leadership conversations drift out of sync.

Because the answer isn’t “ditch Excel.”
It’s: stop using it as your entire data strategy.

What actually works instead:

1 — Centralize your data.
Stop exporting CSVs like confetti. One shared source—Sheets, a small warehouse, anything—is better than ten “perfect” reports.

2 — Assign ownership.
If everyone owns the data, no one does. Quality only scales when accountability exists.

3 — Automate the pipes, not the panic.
Connect systems directly. Excel is fine as a front end—if the plumbing behind it isn’t chaos.

4 — Measure less. Decide more.
Fifty KPIs don’t make you data-driven. A handful of metrics that guide real decisions do.

Excel belongs in modern data stacks. Just not at the center of them.
Tools don’t create clarity. Frameworks do.

Fix the foundation before it fails you. And grab Your Fractional CDO — unless you genuinely enjoy living one broken formula away from disaster → https://lnkd.in/d--GjMpr

undefined undefined

ClickHouse Developer

Tableau Desktop Certified Associate

A Course for Executives

Digital Transformation Academy

Founder @ Valiotti Analytics | Marketing & Sales Dashboards for Growth-Ready Companies | PhD & MA in Data Analytics

🚀 Founded valiotti.com — the go-to data agency for fixing messy marketing analytics.
Top-rated on Upwork & Clutch. Trusted by growth-stage teams to build dashboards, clean up data chaos, and drive faster decisions. Partnered with AWS, Google Cloud, and Salesforce.

Ex-Head of Consumer Marketing at the largest eastern retail company.
I’ve been the one chasing targets with thousand-row tables, slow reports, and siloed data — I know the pain.

Since 2019, I’ve helped marketing and sales teams:
→ Scale revenue up to 25x
→ Save hundreds of hours on reporting
→ Get insights that actually drive action

Let’s talk: https://calendly.com/nick-valiotti

Let's discuss your project and grow you business with the power of data

Join Friday Data Snack for weekly data, leadership, and business insight

A data strategy that looks perfect on paper

usually dies in real life.

I’ve seen this pattern across companies of every size.
There’s always a polished deck, five-year vision, clean architecture, immaculate governance, dashboards that could win design awards.

And then you look at how decisions are actually made.
→ Sales still lives in spreadsheets
→ Marketing trusts the ads manager more than internal reports
→ Finance keeps a “shadow model” just in case
And that's not because people hate data, but simply because the strategy never met reality.

From a Fractional CDO standpoint, this is the core mistake:
teams confuse designing a system with changing behavior.

The thing is, data strategy is an operating model. So if adoption isn’t happening, the problem is rarely tooling. It’s usually one of these:
1. No clear decision owner
2. No agreed definition of “good enough” data
3. No incentive to switch from existing workflows
4. No tolerance for messiness during transition

A strategy that only works when everything is clean, aligned, and fully implemented is just a concept. Real data leadership starts smaller and rougher:

– Which decisions actually matter this quarter
– Who owns them
– What data is “good enough” to act
– Where the workflow already lives
If the answer doesn’t fit into how people already work, it won’t be used, no matter how elegant it looks.

Strategy isn’t supposed to be beautiful.
It’s supposed to survive contact with reality.

P.S. Had this convo with Marcel Dybalski a while ago, thought more people need to hear it!
-------

👋 Hi, I’m Nick — Fractional CDO. I help teams turn “data vision” into decisions people actually use. If your dashboards look great but nobody trusts them, that’s usually where I start.

AI & Data Platform Engineer | GCP & Fabric Certified

“Data-driven” — easy to say, brutal to explain.

I turned it into my elevator pitch.

Step 1 — Get all your data in one place
Your numbers live everywhere — website, ads, CRM, spreadsheets.
First, you just need to collect them.
→ Fivetran: connects the tools for you (free up to 500K rows).
→ Python: for people who like to build things themselves.
Goal: one single place where all your data lives.

Step 2 — Store it safely
Once you’ve collected it, you need to keep it somewhere reliable.
That’s BigQuery.
Think of it as a big digital storage room for all your numbers.
It’s cheap, fast, and handles anything you throw at it.

Step 3 — Make it visual
Turn all that data into something you can actually use.
That’s your dashboard.
It doesn’t really matter which tool you go for. Looker Studio, Tableau, whatever.
A good dashboard shows what matters — sales, traffic, growth —
without making your eyes hurt.

Bonus — real-time?
Want to see what users do on your site right now?
Use Segment.
Just don’t send every click straight into BigQuery — your CFO won’t like the bill.

That’s it.
Three steps. Zero $10K toys.
Just a simple stack that works.

♻️ Repost if this has been the most insightful elevator ride this quarter.
Follow me Nick to learn how to explain it to your grandma next.

First day at work in 2026. Fresh priorities. Same first request:
“I need a dashboard.”

Every analyst hears this. Regularly. Confidently. With zero additional context. Because 9 times out of 10, when a stakeholder says “I need a dashboard,” what they actually mean is:

“Something feels off and I don’t know why.”
The dashboard is just the physical form their confusion takes.

The moment you accept the request as is, you’ve already lost. You’re no longer solving a problem — you’re redecorating anxiety.

It usually goes like this: “Can we get a dashboard to track campaign performance?”

Sure. You build it. Then come the sequels:
→ “Can we slice it by region?”
→ “Actually, over time?”
→ “What if we try a different chart?”
→ “Can we add just one more KPI?” (famous last words)

Congrats. You’ve entered the Dashboard Cinematic Universe. This happens when analysts become BUILDERS instead of PARTNERS.

A good analyst doesn’t just execute. They investigate. They pause and ask the most dangerous question in analytics:

“What decision will this help you make?”

That one sentence ends most bad dashboards on the spot. Because sometimes the answer isn’t a dashboard at all. It’s:
– a short memo
– a mini-report
– or a five-minute conversation that saves 6 weeks of work

If you want to be seen as a partner, not a chart factory, this is the shift that matters.

And this is exactly why I built a simple dashboard discovery framework, the Dashboard Canvas.
No endless meetings. No vague requests. No dashboards that get ignored.
If you build dashboards for a living, it’ll save you hours → https://lnkd.in/djYMn3Fp

Repost it so others can use it too. Use it on your next project.

You hired a CDO.

Nothing changed. Why?
Because one person can’t fix an 8-person problem.

Most data disasters don’t come from bad talent.
They come from asking one human to be:
a strategist, an architect, an engineer, an analyst, a BI developer, an AI owner,a translator to execs, a firefighter on Slack. 

That’s not leadership. That’s organizational confusion.
So what happens? Your “data scientist” is building pipelines.
Your analyst suddenly owns AI because “they’re good with numbers.”
Your BI dev is trying to architect a warehouse.

Every team has dashboards. None of them agree.
Everyone is busy. Nothing is reliable.

This is what should be happening instead:
→ Data Engineers move data — clean, automated, invisible.
→ Data Architects design how everything fits and scales.
→ BI Developers turn tables into usable interfaces.
→ Analysts explain why numbers move and what to do next.
→ ML / AI roles only exist after the foundations are solid.

It’s a relay, not a solo marathon. Each role carries the baton at the right moment.

Now here’s the uncomfortable truth:
A full-time CDO is expensive. And most companies don’t actually need one sitting on payroll.

What they need is the CDO thinking:
→ role clarity
→ sequencing
→ ownership
→ alignment to revenue and growth

That’s why I wrote Your Fractional CDO. Not a leadership memoir. Not AI hype.
A practical playbook for building a data system that works — without hiring an army or burning budget. Available on Amazon → https://taap.it/Yt0nMRT

A common question I hear:
“What does a CDO actually do? And why can’t our senior data analyst just do that?”

It’s a fair question. The thing is:
A data analyst’s job is to answer questions.
A CDO’s job is to decide which questions matter — and which ones shouldn’t be asked at all.

Analysts work inside the system.
CDOs are responsible for the system itself.

That means:
→ deciding what data the company should trust
→ setting priorities when everything feels important
→ choosing what to build, buy, or ignore
→ clarifying ownership so decisions don’t stall in debate
None of this is about SQL, dashboards, or tools. It’s about judgment.

At Series A–C, this gap becomes visible. Data exists. Dashboards exist. Talent exists. But nobody is accountable for direction.

That’s usually when founders start feeling the drag: slow decisions, duplicated work, quiet confusion.

This is where a fractional CDO helps — not by replacing analysts, but by giving them direction.

I wrote a short guide explaining how this actually works in practice: what a CDO does, when the fractional model makes sense, and what typically changes in the first 90 days.

Feel free to grab it in my Featured or via the link → https://lnkd.in/d8TDg3DJ

----

👋 I’m Nick, Fractional CDO. If you’re looking for this kind of data leadership, DM me.

As a CEO and founder — this dashboard is just great. Simply because it respects how execs actually think.
They don’t open a dashboard to analyze. They open it to confirm or challenge a belief.

This one starts with the only question that matters:
Did we hit quota?
101.3%. Answered. No interpretation required.

Immediately after, it answers the second executive question:
Is this short-term noise or a real trend?
MTD, YTD, last month, last year — side by side, same visual language. 
The user doesn’t have to remember numbers. They just register direction.
Green means “we’re fine.” Red means “ask why.” That’s executive UX.

What’s also strong here is sequencing. Cumulative → period → comparison → distribution over time. That mirrors how leadership conversations actually happen in a room.

The total sales number isn’t there to impress anyone. It’s there to ground the conversation. Execs anchor on scale before they talk variance. This dashboard knows that.

The monthly chart at the bottom is doing quiet but serious work. It shows where execution deviated from expectation and how often that happens.
That’s a performance management view, not an analytics one.

Most importantly:
There is no moment where an exec would have to ask, “So what should I do with this?”
They can already tell:
– whether to push the team
– whether to stay hands-off
– whether to zoom into a specific month
– whether this is a sales issue or a planning issue

That’s why this is an executive dashboard.
Not because it has fewer charts. But because it collapses complexity into decisions, instead of pushing complexity onto the reader.

That’s the difference most teams miss. Bravo, Sebastine Amede!

----

👋 Hi, I’m Nick — fractional CDO. I break down what actually makes dashboards useful to leadership. Follow for reviews like this!

Tableau Developer | Business Intelligence Developer | Data Analyst |Tableau Ambassador’25

My feed and comment section keep panicking over “the death of the data analyst.” 

Which tells me most people still don’t understand what the job was in the first place.

A data analyst was never meant to be a SQL code generator. Not before. Not now. SQL was the easiest part of the job — so we turned it into the job.

For years, we hired analysts to write queries, ship exports, and keep the dashboard lights on. But that role always had an expiration date. AI just made it obvious.

Fast-forward to 2026.
→ LLMs write SQL in seconds.
→ Warehouses scale automatically.
→ “Can you pull this data?” is no longer a skill.

But here’s what didn’t change:
CMOs still don’t trust the numbers. CEOs still ask, “Wait — what’s included here?” Meetings still end with, “Let’s double-check that.” Because the business didn’t suddenly become analytical!!

No one is validating assumptions. No one is pressure-testing metric logic. No one is responsible when AI output sounds right but is wrong.
That responsibility moved to the analyst.

A strong analyst today:
— Lives inside the business, not the warehouse.
— Translates messy questions into precise metrics.
— Knows where models hallucinate and why.
— Owns conclusions, not queries.

SQL is cheap now. Dashboards are everywhere.
Judgment is still rare. That’s the job.

If your data work mostly happens with IT, you’re probably not doing strategy. You’re doing infrastructure.

I always say that and my recent conversation with Mike Drake was another proof to its rlevance across teams. We were talking about fractional CDOs, when they actually add value, and why so much data work quietly underdelivers. Here’s the pattern we kept coming back to:

Strategy doesn’t live where tickets live, it lives where outcomes live.
CEO. CFO. CMO. Revenue. Risk. Growth. Allocation.

When data is treated as an IT function, a few things happen almost automatically:
→ The work optimizes for systems, not decisions
→ Dashboards pile up, but ownership stays fuzzy
→ Teams argue about numbers instead of acting on them

That’s not a tooling, but a positioning problem.
The real job of a fractional CDO isn’t “better dashboards” or “one more model.” It’s turning data into a shared business asset:
– Owned by leadership
– Aligned to decisions
– Designed to compound over time

A few things Mike and I aligned on quickly:
→ Execution is getting commoditized. AI and cost pressure made sure of that.
→ SQL is cheap now. Thinking isn’t.
→ Dashboards should be disposable. Foundations shouldn’t.
And if you don’t historize from day one, you don’t get a second chance later.

The path I keep seeing work looks like this:
IT-centric data → shared business capability → revenue leverage. That’s where strategy actually lives. And that’s the only place it gets paid for.

If you’re a business leader trying to understand where your data work is helping — and where it’s quietly holding you back — comment CDO below.

I’ll send you a link to a quick quiz I built that helps identify data bottlenecks, growth zones, and gives you a concrete 7-day action plan to move things forward.

And Mike, thanks again. Always a good sign when a call reorganizes a few neurons instead of just filling a calendar.

Fractional Data Leader | From Chaos to Clarity | Uncomfortable Conversationalist | Liberator of Hostage Data

Nothing fuels overengineering like a decision nobody wants to make.
I’ve seen this pattern more times than I can count.

When a team can’t agree on what actually matters, the solution quietly becomes: “Build everything.”

So data engineering turns into an insurance policy.
Extra pipelines “just in case.” Abstracted models “for future flexibility.” Another orchestration layer because maybe we’ll need it later. A schema that can support ten businesses when you barely run one.

It looks sophisticated. Feels responsible. And it’s almost always a symptom of avoidance.

Because here’s the uncomfortable truth:
Most data complexity doesn’t come from scale.
It comes from indecision.

When no one wants to commit to:
→ which metric actually drives the business
→ which questions deserve fast answers
→ which use cases can wait or die
…the platform absorbs the uncertainty.

And the platform always pays interest.

Overengineered systems don’t fail loudly.
They rot slowly:
– longer lead times
– fragile dependencies
– higher cognitive load
– teams afraid to touch anything
All in service of optionality no one ever exercises.

Good data engineering starts much earlier than tools and architecture.
It starts when someone is willing to say: “This is the decision we care about now.”

So if your data stack feels heavier every quarter, don’t ask “what tool are we missing?” Ask: “What decision are we still avoiding?”

👋 Hi, I’m Nick, fractional CDO. I help teams replace “future-proof” complexity with clarity, ownership, and decisions. If your data stack feels heavier every quarter, let’s talk.
Not ready to talk but still need help? This book might be a place to start → https://lnkd.in/dxS6W92D

Most dashboards don’t fail because of bad data or bad charts.

They fail because they were built too early.

Someone asks for “a dashboard.”
The team opens the BI tool.
SQL gets written. Charts get polished.

And only at the end does anyone ask the uncomfortable question:
“Wait… what decision is this supposed to support?”

That’s why prototyping is non-negotiable if you want dashboards that actually get used.

Prototyping forces the hard thinking before the build:
→ What questions are we really trying to answer?
→ Who is this for, and what do they need to decide?
→ Which metrics matter, and which ones are just noise?
→ What level of detail is useful vs overwhelming?

A simple low-fidelity prototype (even boxes and labels) does two critical things:
1 — It exposes vague thinking immediately.
2 — It lets stakeholders react to structure and logic, not colors and charts.
By the time you open your BI tool, 80% of the decisions should already be made.

That’s exactly why I put together this practical, no-nonsense cheatsheet for dashboard discovery and prototyping.

It’s a repeatable framework to gather clean, complete dashboard requirements — without endless meetings, vague requests, or dashboards that get ignored.

The guide walks you through:
– Running focused user interviews
– Defining business questions first
– Aligning on metrics and breakdowns
– Mapping data sources
– Translating all of that into a clear, stakeholder-friendly layout

If your dashboards aren’t getting used, don’t add more charts.
Prototype better → https://lnkd.in/dxyFYdUy

Read this book before you start your next analytics project.
Because it teaches you one lesson that will save you hundreds of hours and hundreds of thousands of dollars:

Your first question matters more than your first dataset.

Most failed analytics projects start long before the dashboards do.
They start with vague, unanswerable questions:
“How do we grow?”
“What’s working in marketing?”
“Can we be more data-driven?”
These aren’t questions.

They’re weather forecasts with a KPI attached.

A real analytics question is specific, falsifiable, and tied to a decision:
→ “Which acquisition channel produces customers with the highest 90-day LTV?”
→ “Which step in the funnel creates the biggest revenue leak?”
→ “Which product segment is worth prioritizing next quarter?”

Ask better questions → get better insights → make better decisions.
Every transformation starts there.

This principle is so fundamental that it’s the backbone of my new book, Your Fractional CDO.
Not tools, not architectures, not AI dashboards with glowing neon charts.
Just decision-making clarity.

In the book, I break down how to rewrite messy business questions into clean, analytics-ready ones — the exact process I use with founders, CMOs, and product teams.

Once you do this, everything else gets easier:
You know what data you actually need, you stop building dashboards “just in case”, you avoid 3-month detours into perfect ETLs that solve nothing, and you finally get insights that drive action instead of confusion.

Check out Your Fractional CDO on Amazon → https://lnkd.in/dU-kiEUf

If you think AI will “solve analytics” in 2026, you’re focusing on the wrong problems. Here are my predictions. 3 things I’m confident about.

1 — People trust people. Not models.
This hasn’t changed. It didn’t change before AI, it won’t change after.

When you sent a report to a director, they trusted it because you owned the numbers. You could recheck them. You were responsible.
No one wants to blindly trust LLM output. There is always a human who validates, verifies, and stands behind the result.

That’s why a new role emerging this year: AI-enabled analyst.
Not “AI replaces analyst”. Analyst who uses AI and owns the outcome.

2 — Data is still the foundation.
No matter how good LLMs get, everything still depends on data quality: clean structure, clear logic, well-designed models.

Yes, LLMs can help design schemas and even suggest warehouse structures. But you can’t just say:
“Hey AI, fix my data.”

You still have to explain:
→ what the business actually does
→ how metrics should behave
→ which decisions depend on them
Good prompting still requires a brain, context, real business understanding. AI doesn’t remove thinking. It exposes the lack of it.

3 — The chat interface is already annoying.
Let’s be honest: business leaders don’t want to chat with data. One question leads to ten more. No CXO is spending their day interrogating a chat window about metrics.

Chat is great for exploration, not for decisions.
What’s coming instead:
→ You see a chart.
→ A metric drops.
→ You click it.
→ AI runs the analysis in the background.
→ Returns 2–3 hypotheses actually worth checking.

And then — the crucial part — someone still owns the conclusion.
Same role as point #1.

The boring truth: AI accelerates analytics. Responsibility doesn’t scale automatically.

What changes do you think we’ll see?

--------

👋 Hi, I’m Nick — Fractional CDO. I write about building analytics teams that actually work, not just demo well. Follow for practical takes on data, AI, and decision-making.

Nothing says “data-driven company” like KPIs updated by hand every Friday.
But sure, let’s talk about machine learning next quarter.

And yet, every quarter (if you’re lucky, if not — every month), someone asks for a “cool dashboard.”
As if design can fix dysfunction. Surpise-surprise, it won’t.

Here’s the quick sanity checklist every BI analyst to run before saying yes:

1 — Data foundation
If your inputs aren’t automated, every chart is a ticking time bomb. Dashboards built on manual data rot faster than milk in the sun.

2 — Metric alignment
If “Revenue” means one thing to Marketing, another to Finance, and a third to the CEO — you’re not visualizing insights, you’re hosting a family argument.

3 — Decision logic
Every chart should answer a question someone actually asks. If it doesn’t change behavior, it’s just decoration.

So next time someone says, “We just need a dashboard,” maybe ask: “Do we have data… or do we have Excel?”

Because there’s a big difference between building analytics and making PowerPoint interactive.

----------

I’m Nick, and that’s exactly what we fix at Valiotti Analytics — building systems where data flows, definitions align, and dashboards actually tell the truth. If that sounds like your next step, let’s talk!

Most dashboards fail for a simple reason:

They answer questions nobody asked.

We jump straight into charts, colors, and layouts… but skip the one step that actually makes a dashboard useful:
Understanding how people will use it.

Here’s the part so many teams miss:
A useful dashboard isn’t built around metrics. It’s built around decisions.
Before opening Figma, Tableau, Power BI, or anything else — ask your users:

→ What decision do you need to make faster?
→ What is the pain with your current process?
→ What do you consider a “good” or “bad” week?
→ Which slices, filters, or comparisons do you need every time?
→ Which metrics do you actually trust?

If you can get clear answers to these five things, the dashboard basically designs itself.
No more 12-widget monsters. No more “Can we add one more chart?” No more reports that look great… and get ignored.

Because the truth is simple:
Dashboards fail when they’re built for the tool, not the user. Dashboards succeed when they’re built for the decision.

If you want a shortcut:
I put together a Dashboard Interview Cheatsheet — the exact framework I use with clients to turn vague requests into dashboards that actually get used.

Happy to share it: https://lnkd.in/dNSPbAzx
Let’s make dashboards people actually use.

A lot of companies think they’re “behind” on data.

In reality, they’re just out of sequence.

I wrote Your Fractional CDO for leaders who feel pressure to “level up” analytics, but aren’t sure what level they’re actually on. The book walks through a simple but uncomfortable truth:

Most teams don’t fail because they lack data. They fail because they try to operate like a Stage 5 company while still living in Stage 2 or 3.

That leads to:
→ overbuilt dashboards
→ premature tooling
→ stalled initiatives
→ and a constant sense of “why isn’t this working yet?”

The goal isn’t to be “data-driven.” It’s to build the next layer that actually fits your business right now.

Normally, Your Fractional CDO is available on Amazon. But I’m giving free copies to a small number of founders / execs who feel stuck between “we know data matters” and “we don’t know what to do next.”

If that’s you, comment “CDO” and tell me what stage you think you’re in — or where things feel forced. If the book can help you move forward cleanly, I’ll send you a copy.

If I had to give just one piece of data advice, it would be this:
Build the Boston Public Library.

This idea clicked for me over the holiday break, spending a few quiet hours there. No rush. No noise. Just space designed for thinking.

A good library does a few things quietly and exceptionally well.
Books are easy to store — because there’s a system. Knowledge is easy to access — because things are labeled. Learning feels natural — because the space removes friction.
The room itself helps you think.

Now look at most data environments.
Files stacked like books on the floor. Queries written once and forgotten. Sheets that “kind of” match, but never fully agree. Dashboards that exist, but no one sits down with them.
That’s what a library looks like without shelves.

In a real library:
– every book has a place
– similar topics live near each other
– you know where to look before you start searching
– history is preserved, not overwritten

In good analytics:
– every metric has a definition
– related data lives together
– you know where truth lives
– past decisions stay explainable

Libraries don’t optimize for storage. They optimize for understanding.
Great analytics works the same way. It’s not about having more data. It’s about making learning feel inevitable. Because mess spreads quickly. But order compounds quietly.

So if your dashboards feel heavy, your metrics feel fragile, or your team keeps asking the same questions again and again — you don’t need a new tool.
You need shelves, labels, and silence.
Build the library.

------

👋 I’m Nick, Fractional CDO. I write about building calm, decision-ready analytics systems. Follow if this is the kind of data thinking you want more of.

Most companies don’t waste money on data tools.
They waste it on the wrong data hires.

Last year (had to do it!) I released a book called Your Fractional CDO — and if you’re a founder, CxO, or anyone involved in decision-making, it can save you a lot of time, money, and frustration.

It’s a practical 101 to data analytics, written in plain language. But one section matters more than most: data roles.

Your first data hire can either accelerate your business or quietly burn months of runway. Not because they’re bad — but because each role solves a different layer of the system, and there’s a natural order to building a data team.

Here’s the short version:

Data Engineer — pipelines, integrations, storage, identity stitching. They build the foundation everything else depends on.

Data Analyst — investigations, insight generation, decision support. They tell you what changed — and why it matters.

Analytics Engineer — transformations, models, metric definitions. This is where CAC, ROAS, LTV, and retention become trustworthy.

Data Scientist — forecasts, experiments, advanced modeling. Powerful, but only once the basics are stable.

CDO / Head of Data — strategy, architecture, prioritization. The role that ensures data work drives ROI, not noise.

In the book, I break down each role in detail: what they actually do, when you need them, what problems they solve, and the risks of hiring them too early or too late.

I also share a simple framework for assembling your first data team based on your company stage and the decisions you’re trying to make.

If you’re hiring your first (or next) data person, read this first.
Your Fractional CDO is out now on Amazon: https://lnkd.in/dczU_G34

Hot take:

You don’t need a data team.

Your nephew’s “pretty good at Excel,” right?
He can whip up a dashboard, eyeball some trends, and tell you what’s “probably working.” No need for data governance, definitions, or version control — just vibes and pivot tables.

Who needs analysts when you’ve got family?

Head of RevOps | Recently helped SaaS organisation 19% faster growth, +10-20% sales productivity, and 30% lower GTM costs with Revenue Operations

My Name is Ali and I love building things. Whether that be data solutions, my business, or helping other organizations at the end what I'm doing is building myself into the person I want to become. 

Our thesis is that data is the lifeblood of any organization and will only work if it's made custom to your goals, needs and priorities at any stage of your growth.

Unfortunately, many businesses don't have that as they 

-  Rely on off-the-shelf solutions that are not custom to their needs and don't adapt 

- Their data is all over the place resulting in not able to make any data driven decisions.

- Can't rely on their data to make rational logical decisions. 

- They track surface-level KPIs and not actionable insights that would give them a competitive advantage 

- They don't know how to bridge the gap between business and data expertise.

Well at aztela we are that bridge that connects business and data together to develop a moat inside of your business resulting in: 

- One of our clients generated over 20x ROI since starting working with us resulting in over $10m in revenue attributed 

- Increasing profits due to seeing everything at one place which would normally take you 100h a month

- Decreased churn and increased LTV able to see bottlenecks, and root cause as opposed then spinning in circles 

- Fixing inefficient operations due to removing data silos 

We do that by utilizing data you already have and turning it into actionable insights. 

We implement custom analytics dashboards (BI) and advanced solutions to achieve business results just by utilizing data you already have.  

If you want to create a moat, competitive edege in your market click on this link to speak to our team- > 
https://calendly.com/alizu/15-minute-meeting-aztela?month=2024-04I

Ali Z

I recently met a manufacturer with 500 employees and 800 dashboards. And they were still flying blind. Your production run is still late, you still bleeding margin and your inventory is still wrong. 

If your data doesn't trigger an action, it's just noise. 

Most manufacturers, asset heavy orgs are drowning in dashboards.
Every kpi looks great but you’ve got 50 dashboards for each.

But the same issues keep showing up quarter after quarter:

Late production runs nobody catches until it’s too late.
Maintenance firefighting that kills uptime.
Inventory volatility eating your margins.
Leadership meetings where nobody trusts the numbers.

The truth?

Dashboards don’t fix problems.
They just make them prettier.

You’ve built systems of 'reporting' when what you need are systems of action.

Dashboards show you what happened.
But dashboards are just the medium, the format. 

The old model of “data → dashboard → decision” is too slow.
By the time you ask for data, ask for a change via Slack or Email, gets back to you in 2-weeks priorites already changed.   

There is no time for bureaucracy . 

Because  your data is only valuable if you can perceive, react, and re optimize  

Without it you’re not “data-driven.”

Here is what need to do.

1. Start with problems not tools. 

Don’t start with data or tools. 
Start with the decisions that move the P&L called data strategy. 

Examples:

When to switch production lines.
What is our profitability to save time on reconciliation . 
When to reorder raw materials.
Speed up drug releases  

If a dashboard doesn’t directly impact a decision or an action, kill it.
Don't built it, even if someone asked for it. 

2. Merge IT and OT for Real-Time Feedback.

Your ERP, MES, SCADA, and shop floor systems are all talking,  but not to each other.

That’s why you’re still operating at “email or Slack speed.”
IT+OT so insights can trigger changes instantly.

You are drowning in data but, using it 10% 
That’s not future tech, that’s just modern integration done right.

3. Build a Data Fabric, Not Another Pipeline.

Factories don’t need more dashboards.
They need a nervous system.

An industrial data fabric connects every machine, application, and metric through one unified data layer.

That’s how you don't need to rebuild dashboards 50 times.

4. Shift from Dashboards to Decision Systems

Don't get stuck in 'dashboard la la land' only.  
I seen orgs have more dashboards then FTEs but still can't act or trust it.

Your goal is to operationalize data to act. 

Example: 
Automated alerts when downtime exceeds thresholds
Real-time material variance notifications 
Providing users access  data_marts.

Data only valuable when it's used to built data products. 
Dashboards, ML models, AI platforms are tip of the iceberg. 
The real work is below surface.

The companies winning in 2026 aren’t the ones visualizing data better.
They’re the ones who leverage data to business impact that detect, decide, and act at scale.

Ali Šifrar

CEO @ aztela | Fractional data leader helping scaleups+ mid-market, gain control, value from their data to accelerate their data, digital & ai journey.

Recently met a manufacturer having over 1,000 connected machines. Terabytes of data. Modern data tools. But when asked COO: “Can you see your cost per unit now?". He went silent. You don't have visibility. You are collecting expensive noise.

Every manufacturing exec says they’re “data-driven.”
They’ve connected every factory, sensor, and system.
Petabytes of data streaming in real time.

Then you ask:

> “Can you tell me which line will miss its delivery target tomorrow?” 
> “Can you see cost per unit in real time?”
> “Do all your dashboards agree?”

Silence.

Manufacturers have never had more data, and never trusted it less. 
Scale has become noise.

One manufacturer in healthcare we worked had +1000 machines operating worldwide.
Over 1000s are connected and streaming data.

And even their CIO said it plainly:
“We needed a foundation of data to make insights meaningful.”

That’s the reality every manufacturing CIO is facing right now.

The last decade was about connecting everything.

The next decade is about making sense of it.

Most buy all the tools, get all the data just in case and think ‘problem solved’

But you’ve got:
-MES systems calculating OEE five different ways.
-IoT platforms pumping out unread sensor logs.
-“Digital threads” that look more like spaghetti.
-AI pilots choking on inconsistent data.

Everyone scaled data collection.
Almost no one scaled context.

You don’t have a visibility problem.

You drowning it 

1. List out inventory

Count every dataset you have.
Now count how many are actually used in decisions.
In most orgs, it’s under 10%.

The rest? 
Digital exhaust.

If your team can’t tie a dataset to a decision, kill it.

2. Kill Redundant Data Streams

Every plant collects the same metrics three times, MES, historian, IoT.
Pick the source closest to reality.
Kill the copies.
Redundant data doesn’t create trust; it kills it.

3. Align on a Single Definition of Truth

If your OEE in Plant A isn’t the same as Plant B, you’re not measuring performance.
You’re measuring chaos.
Build a semantic layer that defines metrics once.

Let every dashboard pull from it.

Truth is not a dashboard; it’s a definition.

4. Contextualize OT + IT data.

OT tells you how something happened (sensors, machines, PLCs).
IT tells you what happened (ERP, MES).
Combine both, and you can finally answer why it happened, and fix it.

5.Move from dashboards to data products.

Dashboards show you data.
Data products act on it.
Example: automatically alert engineering when defect rate exceeds design tolerance. That’s impact.

Manufacturers spent the last 10 years connecting machines.
The next 10 will be about connecting meaning.

More data won’t make you smarter.

Better context will.
Before you fund another “data lake” or “smart factory” initiative, ask one question:

“Does this make the signal stronger or just the noise louder?”

That’s the manufacturing data paradox.
And most of your competitors are drowning in it.

CEO @ aztela | We help manufacturers, logistics and asset-heavy organization mid-market build scalable AI ready data foundation so they leverage data to decrees cost and gain edge in the market.

The winners in 2026 won't be the manufacturers or asset-heavy org with the best AI models. They will be the ones who leverage their Proprietary Data that major LLM hasn't seen. 
 
Everyone’s talking about AI.
Most manufacturers and asset heavy companies aren’t remotely ready for it.
They have a data architecture problem.

Walk into any org today and you’ll see the same story on repeat:

17 PLCs all speaking different languages
ERP that only talks to itself
A historian that logs numbers no one looks at
Excel reports sitting in SharePoint folders named “Final_v3_Updated”

And yet, the board is asking, “When are we deploying predictive maintenance or generative AI?”

Here’s the truth no one likes to say out loud:

You can’t train AI on data you don’t understand.

AI doesn’t learn from isolated points.
It learns from relationships, between machines, jobs, operators, and conditions.

Here is simplified 3-Step AI Readiness Path for manufactures, nd asset heavy companies

We’ve helped dozens to go from “we have tons of data” to “edge with propitery.”

1. CONNECT,  Get the Data Moving

Every AI story starts with the same pain: the data is trapped.
In PLCs. In historians. In MES systems. In vendor clouds.

Actionable Moves:
-Connect sensors and ohters using modern, lightweight protocols.
-Stop dumping data into yet another proprietary historian. 
-Standardize the basics, timestamp, source ID, event type, value. 

If your data isn’t moving, nothing else matters.

2. CONTEXTUALIZE, Give the Data Meaning

Here’s where most companies fail.

They connect everything and celebrate- but all they’ve done is move chaos from on-prem to the cloud.

AI doesn’t care that you have 10 million data points.
It cares whether those points have meaning.

Actionable Moves:
-Build a (UNS) - a single, logical data model that organizes information by plant, line, machine, product, and property.
-Capture who, what, when, and why -  not just “Temp001 = 78°F.”
-Map relationships -  e.g., Machine A feeds Machine B; Job 1005 ran with Operator X using Material Y. 

That’s contextualization.
It’s what separates “data movement” from “data understanding.”

When your data starts speaking a common language, AI starts listening.

3. CORRELATE, Turn Context into Action

This is where it all pays off.
Once you’ve connected and contextualized your data, you can finally see cause and effect.

Actionable Moves:
-Merge OT data and IT data (from your data lake) into one model.
-Correlate across systems — production → quality → energy → supplier → environment.

Start small. Choose a use case where correlation drives ROI. 

Every manufacturer says they “want AI.”
Few are willing to do the groundwork that makes AI possible.

Winners won’t be those with the best LLMs or tools, but those who exploit their proprietary data that major models aren’t trained on

Comment me “AI Readiness” and I’ll send our 3-Step AI Readiness Playbook

Recently met a COO who wanted to scale data across multiple sources, and 12 plants. But their integration is running on a Python script written by "Joe" three years ago. Every time you want to onboard a new source, it takes months and it breaks. 

Every manufacturer says the same thing:
“We want to scale data, AI across all our plants.”

But here’s what they don’t say out loud:

“We can’t even connect half our sources, machines to the same system.”

And that’s the real reason AI isn’t scaling.
Not the algorithms.
Not the data scientists.
It’s the integration debt.

Every new line, every new machine, every new system another one-off integration, another script, another point of failure.

By the time your new data source connects the first line, the business has already moved on.

If you want scalable AI, you need scalable connectivity.
And that means one thing: Connectivity Factories.

Manufacturers understand scale better than anyone.
You don’t build one widget at a time you design a production line.

So why is your data integration still a custom craft?

Every company we talk to treats integration as a project.

-New MES? Write new connectors.
-New ERP? Rewrite the integration.
-New site? Rebuild it all again.

Each project “solves” one local pain but it creates global chaos.
Now, you’ve got hundreds of one-off integrations, no documentation, and no idea why your AI use case takes six months to go live.

Even worse if “Joe” who built this in ops/IT leaves and now you just in mess. 

The outcome?

A pile of “AI pilots” that never make it past the POC stage.
Data teams chasing protocols and APIs instead of insights.
Executives wondering where the ROI went.

We’ve seen this movie.
And the only companies escaping it are doing one thing differently they’re industrializing connectivity itself.

The companies that are scaling aren’t doing more integrations, they’re manufacturing integrations.

They’ve built standardized, repeatable systems that make connecting machines, systems, and data sources..

- Standardized integrations templates
- Use AI to assisted with mapping
- Automate deployment across plants (Use AI hint)
- Reuse logic across every integration
- Every plant “installs” connectivity, not rebuilds it

Integration becomes scalable, measurable, and fast, like any other industrial process.

The results 

-Connectivity setup time drops from weeks to hours
-Integration stops being a bottleneck.
-Data becomes reliable, consistent, and scalable
-Data and AI products finally speed up 10x.

Here’s the truth no one wants to say:
Integration is the bottleneck of every data or it team. 

Until you treat it like a scalable system not a project you’ll stay stuck in pilot purgatory.

Stop building integrations.
Start building clear process to move faster. 

We broke down the full playbook in our latest newsletter.
Check it out below ↓

Dashboards are dead. By the time your teams reacts, already lost +100k and the COO is already screaming. This is the reality for 90% of manufacturers. You don't need another dashboard, you need action. 

The line goes off spec or route delivery is late. 
The dashboard shows it three hours later.
By the time maintenance reacts, the scrap’s already in the bin and the CFO is on the floor asking what went wrong.

That’s data-delayed.

One manufacturer and supply-chain org we worked decided to fix that.

And in doing so, we quietly built what every manufacturer will eventually need:
an ‘operational’ data warehouse that runs their factory floor like a reflex not a ‘dashboard’.

After that was talking to VP of Manufacturing and COO, when one of them said this:

“We used to respond to issues in days. Now we respond in minutes. That’s the difference between a steady plant and a profitable one.”

They didn’t add more dashboards.
They redesigned how data moves, reacts, and drives action.

Here’s what changed:

-Signals that used to sit in siloed PLCs now stream continuously into a unified time-series system.

-Machine data gets contextualized instantly which product, shift, material lot, operator.

-Analytic models push recommendations back to the domain experts, not just to dashboards that are opened once a month in meeting when it’s to late.  

So when a temperature, route delay, stockouts or vibration reading goes off spec the operator sees it in seconds, not hours.

They didn’t “modernize BI.” lol 
They operationalized their warehouse.

Or what we call it used the propitery data and turn it into intelligence

Here’s the framework any asset heavy org can copy. 

1. Identify Where Latency Costs Money
Ask one question: Where does a 15-minute delay cost us $50K?
Start there. That’s your ROI case for real-time

2.Connect Machine + Business Data
A temperature spike means nothing unless you know what batch, material, and shift it belongs to.
Fuse your PLC or historian data with ERP/MES context in near-real-time

3. Stream, Don't Wait
Move from nightly ETL to continuous micro-batches.
Every minute counts. You don’t need Kafka you need freshness.
Ask yourself always “What does it cost us” 

4.Close the Loop
If your data ends at a dashboard, it’s dead.
Data → Decision → Action must be one continuous loop.
Example:
If temp deviates 2°C, the system notifies maintenance instantly.

5.Make Operators Data Domains 
When the person who is domain expert, in front line sees the data in real time they act faster than any dashboard ever could.

It doesn’t matter if you have best data stack if you only look from hindsight. 
Data has to be used to be valuable, not reported only on. 

This manufacturer and other supply chain firm didn’t get ahead by chasing buzzwords.
They got ahead by closing the latency gap between data collection → action

That’s what true data-driven company looks like
Not another BI tool

99% of Your factory Data Is Useless. You have 80,000 sensor tags. But  can't answer basic question "Why did throughput drop 4% last week?" without a 3-hour meeting arguing about who is right. You don't need 80,000 tags, you only need 4 schemas

Every manufacturing leader says the same thing:

“We’re swimming in data.”

But when you ask a simple question like

“Can you tell me, right now, why your throughput dropped 4% last week?”

You get ten different answers… and none of them are right.

That’s because 99% of factory data is completely unusable.
And it has nothing to do with AI, tools, or dashboards.

Manufacturers have a data usability problem and it’s killing trust, and ROI.

Manufacturers generate more data than almost any other sector
But less than 1–5% of that data is ever used.

Each machine speaks its own “language.”
The same tag name means different things across lines.
Quality data arrives eight hours late.

You’ve got 80,000 tags, but nobody knows what machine half of them belong to.

Meanwhile, leadership spends millions chasing "AI” or some "vertical platform " that’s built on this mess and wonders why nothing scales beyond a single pilot.

Before you can do AI, predictive maintenance, or digital twins
you need your factory data to speak a common language.

That means automating how you collect, clean, and standardize data from the shop floor.

The manufacturers who are actually scaling AI (think Coca-Cola’s operations) all did the same thing first

They implemented a data-first architecture that structures everything into just four schemas:

1.Cycle Schema: Every discrete event every few seconds, across every machine.
2.Downtime Schema: When stopped, why it stopped, and how long.
3.Part Schema: How materials move and transform across.
4.Defect Schema: Where quality issues occur and where they’re detected.

That’s it. Four schemas.
Not 80,000 tags. Not 15 disconnected systems.

Once structured, data becomes instantly usable for analytics, AI, and even GenAI copilots that let operators ask questions in natural language

“Why did Line 4 slow down yesterday?”
“Which shift has the highest defect rate?”

That’s when “data-driven” actually means something.

Here’s the Playbook That Works

1. Automate Data Engineering: Stop manually mapping tags. Use streaming pipelines that structure everything in real time.
2. Standardize Once, Scale Everywhere: One model across all plants build once, reuse everywhere.
3. Empower the Frontline: Data is worthless if it doesn’t help operators act in real time.
4. Measure ROI in Productivity, Not Dashboards: 1% improvement in OEE = millions in recovered value.

If your data team is still debating which "ai tool" to buy, you’ve already lost.

Real question isn’t “Which model should we use?”
It’s “Can our data even be used?”

99% of your factory data is sitting idle, because it’s not, trusted, or usable.

Fix that, and every AI pilot, dashboard, predictive model suddenly starts working

Recent manufacturer spent +$1millions on "Data Transformation." But you operators still trust a $2 notebook more than your dashboard or predictive model. You can't buy trust. 

Walk into any “digitally transformed” asset heavy company and you’ll see the same thing:

Robots. Dashboards. Sensors. AI pilots.

But behind the curtain?
Spreadsheets.
Email chains.
People saying, “We don’t trust the data.”

Let’s be clear you can’t automate what you don’t trust.

For the past decade, manufacturers have been told that digital transformation means buying more tech.
So they bought robots, IoT, and “predictive AI.”

And yet:

-Operators still take notes on paper.
-Finance and Ops can’t agree on OEE.
-Dashboards are beautiful and ignored.

That’s not transformation.
That’s expensive stagnation.

Because automation without trust just means you’re accelerating bad decisions.

Any Transformation Starts With Trust

Every manufacturer that’s actually scaled AI, reduced downtime, or unified global KPIs they didn’t start with tech.

They started with data trust.

Data trust means:
- People believe the numbers before they act.
- Systems agree on one version of truth.
- Decisions can be traced and defended.

If you skip that foundation, you’ll keep “digitizing chaos.”

If your “digital transformation” isn’t moving, stop buying tools, failed AI POCs and start fixing trust.

1.Fix Visibility Before Velocity
You can’t optimize what you can’t see.
Connect your systems, clean the data, and make performance visible before you automate anything.

2.Audit Where Data Trust Breaks Down
If two systems show two answers, no one acts.
Map your data flows. Document the disconnects.

You either have:
Inconsistent data - No standardization across
Delayed data - Stale and outdated
Or Opaque data- No one knows where the numbers came

You’ll find 80% of your problems are definition-process problems, prior data hitting the cloud not system problems.

3.Standardize Language Before Systems
Create a data dictionary units, KPIs, naming rules.
Make everyone speak the same language before they plug anything in.
Boring? Yes.
But it’s the single biggest unlock for scaling automation.

4.Bridge IT + OT
Transformation fails in the gap between IT and the plant floor.
Build one governance team that owns both.
Data doesn’t care who your org chart says reports to whom.

5.Build a Culture of Verification, Not Blame
If people get punished for “bad data,” they’ll hide it.
If they’re rewarded for fixing it, trust compounds.

Digital transformation isn’t about robots.
It’s about confidence in your data, in your people, in your decisions.

You can automate lines and connect machines all you want.
But if your people don’t believe the data behind it?
You’re not transforming. You’re decorating.

Automation amplifies whatever data you feed it.
If your data is wrong, automation just makes the wrong faster.

Start with trust.
Everything else scales from there.

"We’ve got data everywhere, but no answers." A COO of a $100M manufacture told me this last week. They had an $1m ERP, 400 dashboards and 5,000 sensors BUT still couldn't answer why the production line went down on Tuesday or drop in OEE across.  

Let’s be blunt most manufacturers and asset-heavy companies don’t have a “data strategy.”

They have 20 years of data chaos.

ERP, MES, SCADA, spreadsheets, PLCs none of them talk.
Every team has their own version of the truth.
And every “digital initiative” dies in pilot purgatory.

Yet when executives hear “data strategy,” they imagine a two-year consulting project full of diagrams and jargon.

That’s not what this is.
An industrial data strategy is not a report.
It’s the operating system of your factory.

A mid-market manufacturer we met had spent years investing in tech new MES, shiny dashboards, “AI readiness.”

But they couldn’t answer three simple questions:

- What’s our real-time OEE across lines?
- Why did that line go down last Tuesday?
- Which supplier issue is driving our scrap rate?

The COO summed it up perfectly:

“We’ve got data everywhere, but no answers anywhere.”

That’s an architecture problem.

The most successful mid-market manufacturers the ones scaling AI and digital aren’t the ones buying more software.

They’re the ones that finally said:

“Enough chaos. Let’s fix the foundation.”

They follow a 4-layer data blueprint that actually works:
Ingest → Store → Contextualize → Act

1.Ingest Capture What Matters, Not Everything

Stop streaming every sensor tag “just in case.”
Decide what data actually drives a business decision.

Start by connecting:
PLCs (machine states, alarms)
MES & ERP (production orders, downtime)
Quality systems (defects, batch IDs)
Transform and filter at the edge before it ever hits your cloud.
Collect less. Use more.

2. Store tructure Your Data or Drown in It

You don’t need a “data lake.” You need data discipline.

Use the Bronze–Silver–Gold model. 
With clear constitutional  layer. 
Your "predictive model"  is only as good as the structure beneath it.

3.Contextualize Turn Data Into Meaning

Most manufacturers already have the data.
They just can’t relate machine → operator → supplier → part → defect.

That’s where contextualization comes in:

Use a Unified Namespace (UNS) to standardize naming and structure.
Build a Graph to link data across systems.
You can’t run predictive maintenance or AI until your data understands itself.

4. Act Stop Reporting. 
Dashboards are fine. Closed-loop action is better. 
Empower domain experts. Delayed or not used data is useless  
Data → Transformation → Trigger

Because in 2026, the manufacturers winning aren’t the ones with more sensors
They’re the ones who know how to turn proprietary data into intelligence and action.

→ Comment “Blueprint” 

We’ve built a free 90-Day “Manufacturing Data Strategy” the exact 4-layer plan mid-market manufacturers, supply chains & asset heavy organization.

8 out of 10 "Predictive Maintenance" projects fail. Recent manufacturer wasted $400k and 12-months on model nobody used. Not because the AI is broken. But because of your data architecture. 

Every slide deck every pitch sounds the same:

“We’ll predict when your machines will fail before they fail, inventory or delivery delays” 

But here’s the truth:
8 out of 10 predictive maintenance projects fail before they ever deliver value.

Not because AI doesn’t work.
Because your data doesn’t.

A $400K Lesson in Skipping Basics

A mid-market manufacturer spent 18 months, $400K, and hired a small data science team to predict bearing failures on a critical asset.

They had sensors.
They had dashboards.
They had an ML model running in the cloud.

But they didn’t have context.

Maintenance data lived in SAP.
Telemetry lived in the PLCs, IoTs, WMS and POS.
Failure logs were Excel sheets in someone’s inbox.

The model’s predictions were useless.
Technicians ignored them.
The system was shut down six months later.

The post-mortem was simple:

“We tried to predict the future without understanding the present.”

Why Most Predictive Maintenance Projects Fail

Let’s cut through the consulting jargon the real reasons are painfully simple:

Data Silos Everywhere – You can’t train an ML model if your maintenance, MES, and sensor data don’t talk to each other.

No Context Layer – You have data points, but no lineage no way to connect a failure to operator, supplier, or process.

Jumping Straight to AI – You skipped the foundational stages: OEE visibility, diagnostic analytics, and process understanding.

Bad Edge–Cloud Strategy – You’re streaming every millisecond of data into your lake turning it into a swamp.

No Trust from the Floor – Maintenance teams don’t act on alerts they don’t believe.

The Playbook: What Smart Manufacturers Do Instead

Smart manufacturers don’t start with AI.
They start with data discipline.

Here’s the exact playbook they follow:

Step 1: Build Real-Time Visibility (Descriptive Analytics)

Connect assets. Build OEE visibility. Fix your real-time awareness before predicting anything.
Start with least amount of variables. 

Step 2: Contextualize Your Data (Diagnostic Analytics)

You are lacking why. 
Map maintenance, MES, and IoT data into a unified model often a Unified Namespace or Graph.

Step 3: Standardize and Clean Early

Set a single schema for units, tags, and statuses. Enforce it at data ingestion, not reporting time.
This ensures your data is usable for analytics from day one.
You can’t have one source temperature in Celsius and another in Fahrenheit.

Step 4: Push Small One Asset

Don’t boil the ocean. Pick one high-impact thing, route, machine, inventory, train a model, and prove ROI in 8–12 weeks.

Step 5: Close the Loop

If can’t act on it, it’s usless. 
Automate the workflow. 

If your predictive maintenance project is failing, it’s not an AI problem.
It’s a data architecture problem.

The worst way to find out your data is broken is from an angry COO. Global manufacturer 5+ plants, 10,000 suppliers, 5TB of data. Hundreds of automated pipelines and dashboards. If your VP finds the bug before you do, you, it's to already to late. 

Not because of an outage.
Not because of a cyberattack.
Because of one dropped column.

A single schema change in a production system one field someone thought “wasn’t needed anymore” silently disappeared.

By the time anyone noticed, it had already cascaded:

Dashboards went blank.
Planning models failed.
Forecasts froze.
Supply chain teams reverted to manual SAP transactions.

Factories couldn’t see what to produce next.
Managers couldn’t trust their data.
And the global ops team spent days firefighting instead of running the business.

That’s how one schema change broke the business.

This isn’t a data story. It’s a reliability story.

Every company thinks their data problems are “technical.”
They’re not.
They’re operational.

If your business depends on ERP, BI, or AI systems you’re one schema change away from chaos.

Why?
Because every data pipeline is a set of unwritten contracts.
When those break, everything downstream does too.

The field that disappeared wasn’t the problem.
The lack of enforcement was.

The real lesson: Every schema is a contract.

You’d never let a supplier change a part spec without notice.
So why do you let your systems change data structures without validation?

Your data is your supply chain.

Schemas are your supplier specs.
Pipelines are your logistics network.
Dashboards are your distribution.

And when your “data supplier” changes specs without a contract your entire operation stalls.

That’s what happened here.
 How to Stop This From Happening to You

Here’s exactly how we prevent this chaos for manufacturers, and asset heavy orgs. 

1.Enforce Data Contracts

Define exactly what data producers deliver fields, types, update frequency, ownership.
In DBT, that means schema + test definitions.
In business, it means accountability.
No contract, no data.

2.Shift Testing Left

Don’t wait for dashboards to fail.
Validate schema changes before deployment.
Treat data pipelines like code versioned, reviewed, and tested before release.

3.Automate Drift Detection

Use DBT, Great Expectations, or even AI (LLMs trained on metadata) to flag schema drift automatically.
If a column changes, you should know before your planners do.

4.Make Ownership Non-Negotiable

No one should say “that’s not my data.”
Ops, Finance, and IT all own reliability.
Because when data fails, everyone loses.

The next major business outage won’t come from a ransomware attack or a system crash.

It’ll come from a schema change no one noticed.

That’s the quiet risk every executive is carrying right now.

Your dashboards won’t crash they’ll just lie.
Your forecasts won’t stop they’ll just be wrong.
And your teams won’t notice until they can’t deliver.

Your supply chain is broken. They are automated, AI-enabled but completely blind. Ask a COO why they’re out of stock. You’ll get 14 different excuses from 5 different systems. The problem with modern supply chains isn't logistics or tools. It's data. 

You don’t need more sensors or $2m digital twin.

You need to see your data move.

One manufacturing firm we worked figured this out 

Their CIO and COO told a story that stuck with me:

“We can trace diaper from the store shelf all the way back to the exact microsecond it hit the glue gun in the factory.”

That’s not a metaphor.

It’s data lineage every transaction, every handoff, every timestamp linked from the consumer back through manufacturing, distribution, and finance.

Here’s how it works for them

When a parent scans a product pack for loyalty points, that simple scan connects:

-the store it came from,
-how long it sat in distribution,
-what plant produced it,
-whether that batch had issues,
-and what it cost to get there.

That’s real visibility.
That’s how you prevent supply chain fires before they happen.

And no you don’t need to be giant conglomerate like P&G or spent $5m on tools. 
Wrong approach. 

You just need to stop treating data as IT’s job

We’ve helped manufacturers build the same visibility using the systems they already have 

1.Start with a single pain point.

Don’t “launch a dashboard or project.” 
Start with urgent problem, not tools, costing you. 

“Why do we keep running out of SKU 123?”
“Why does lead time swing by 40% region to region?”

If that question crosses multiple functions, perfect. 

2. Map your “data DNA.”

Don't go into built mode yet.
List every system and entity that touches that process.
ERP → MRP → WMS → TMS → Finance.

Who owns each? Where does the data change? What rules apply? Where used downstream?
This gives you visibility into how your supply chain actually runs behind scenes before building it.

3.Instrument the flow.

Add timestamps, anomaly checks, and ownership.
Every data handoff should say: who, when, accuracy, and next hop.

4.Tie lineage to data products  

Example 
Problem Repeated out-of-stock in key stores
Use Case: Trace product flow from DC → shelf → sales
ROI: +15% inventory accuracy 

or facing 

Customer complaints on product defects. You can trace SKU to manufacturing batch & time stamp and cut issue-resolution time from weeks to hours 

That’s the business case. 

5. Assign business ownership

IT can’t own data lineage
If everyone owns their data node, the lineage keeps itself healthy.

6. Build modular, not monolithic. 

Start with urgent problems like “order to delivery”
Prove it in 4–6 weeks. Then add

Most supply chains don’t fail because of bad logistics
they fail because leaders can’t trace their data

Future-ready supply chains aren’t the ones that predict the future or best tools.
They’re the ones that leverage data

Comment "Supply Chain", and I’ll send you  the exact 6-week playbook manufacturers are using to fix this

Until IT and OT speak the same language, your $1M dashboards are just expensive "useless digital decorations" Recent COO asked "Why did yield drop 4%?" and nobody could tell him. 

A few months ago, I sat in a manufacturing  conference room watching two teams argue about the same problem.

Your digital transformation didn’t fail because you picked the wrong tool.
It failed because your IT and OT teams don’t speak the same data language. 

They just didn’t know it yet.

IT is building data lakes, deploying Snowflake, and talking about dashboards and OPEX. 
OT is wiring sensors, logging PLC tags, and talking about machine uptime.

IT thinks in tables and transactions
OT thinks in tags and tolerances.

They were both convinced the other side didn’t understand the business.

And they were both right.

Then COO walks in, looks around, and asks one question:
“Why did yield drop 4% on Line 3 last quarter?”

Silence.

IT pulled up reports showing production volumes, OEE, and downtime trends.
OT pointed at machine logs and said, “Temperature spikes, but we don’t know when.”

No one could connect the two.

The data existed.
The insight didn’t.

That’s when it hit me, they weren’t arguing about systems. They were speaking different languages.

Here’s the truth no one likes to say out loud:

Digital transformation doesn’t fail because of your tech stack.
It fails because your teams don’t share a common data language.

IT speaks “business.”
OT speaks “machines.”

They both produce data, just in completely different dialects.

IT

What: Orders, materials, costs
When: Daily jobs
Where: Cloud systems
Why: Visibility

OT

What: Sensors, cycles, tolerances
When: Millisecond updates
Where: On the shop floor
Why: Uptime

You can’t unify that by buying another platform.
You have to translate it into a shared language, one built on context, not tools.

We started small.
No buzzwords. No 2-year roadmap.

Just one question:

“What’s the one decision you can’t make because your data doesn’t connect?”
We picked yield analysis.

Here’s what we did in 60 days:

1. Mapped the language
Defined what “batch,” “machine,” and “tolerance” actually meant — across MES, ERP, and IoT.

2. Unified identifiers
Synced Part IDs, Batch IDs, and Timestamps so systems could finally talk to each other.

3. Built one contextual model
Merged design tolerances from PLM with live machine data  so when a deviation occurred, we knew which part, which shift, which batch.

4. Closed the loop
When a tolerance breach hit, it didn’t go to a dashboard — it triggered alert and created a feedback ticket to engineering.

After a while COO came back.
Asked the same question:

“Why did yield drop 4% on Line 3?”
This time, now were able to see full picture.

“Machine 12 ran outside tolerance for 2 hours on Batch 431A, same issue as the supplier lot from March.”

No debate. Just facts.

That’s what happens when IT and OT stop reporting separately and start understanding each other’s data

Recently talked to CIO of an F1 team on why they win. He didn't say due "perfect data architecture." But due speed. Most burn $500k, 18-month designing the "perfect" architecture that delivers zero impact. By the time they finish, the business has moved on.

You don’t win by building the most complex car. 
You win by iterating faster than your competitors.

Every team has access to the same tech, data, and talent pool.
But the fastest ones don’t out-engineer they out-learn.

Meanwhile, most data teams are doing the opposite.

They spend 12–18 months designing “perfect architecture” multi-cloud diagrams, vendor evaluations, governance committees and by the time it’s ready,
the business has already changed direction.

Architecture matters. Data foundation matters. 
But only as a means to speed, not an end in itself.

Most data programs die because they optimize for elegance instead of execution.

I’ve seen companies burn $500k+ on replatforming projects that produced zero usable impact because no one asked:

“Are we moving faster? Are we decreasing delivery times?

The F1 world understands this better than anyone.
Every race weekend, teams run thousands of experiments.
They test, fail, learn, and push updates in real time.

Not once a year.
Every week.

That’s the mindset Data and CIOs need to steal

1.Launch fast, learn faster.

Stop over-architecting.
In F1, no car is “finished.” They deploy updates constantly.
Do the same deploy data products in 2–4 week sprints, not quarters.
Build, test, iterate. Treat every release as an experiment.

Measure success in adoption, trust and results. 

2.Reduce drag first.

The slower a race car, the more drag it carries.
Same with your data stack.

Cut tools. Kill redundancy.
Don't mesure success in tool count.
Most CIOs think more platforms = more capability.

Wrong. 

It’s more integration overhead, more latency, more failure points.

3. Optimize for marginal gains.

F1 wins come from 0.2% improvements done 1,000 times.
Your data team should operate the same way.
Small wins compound into trust, adoption, and ROI.

4. Train like a pit crew.

Pit stop takes <3 seconds because everyone knows their role.
That’s what your data operations should look like.
Small, cross-functional teams aligned to outcomes not tickets.
Reward precision under pressure.

5. Balance short-term with long-term.

F1 teams design next year’s car while racing this one.
CIOs must do the same: ship value weekly while evolving the foundation.
You are not “strategic”, if you’re not delivering in real time.

Architecture wins awards. 
Agility wins championships.

Your competitors aren’t beating because they have better tools.
They’re beating you because they learn, adapt, and ship faster.

If you’re leading a data team, ask yourself:

Are we optimizing for speed to value?

Business doesn’t care how elegant your data model is.
They care how fast it helps them win.

Stop building like an enterprise IT department.
Start building like an F1 team.

Your Supply Chain is running on a 12-person email thread. By the time you figure it out, it's to late  Your nice dashboard show you what happened But not what to do next. 

Every ops leader knows the moment:
It’s 2:13 a.m.

Your phone buzzes “URGENT: shipment delayed.”

You scroll through a 10-person email chain, 12 replies deep.
Everyone’s copied, nobody’s deciding.

By the time you re-route, the production line’s idle, the freight’s expedited, and the cost has already hit the P&L.

Here’s the truth:
You don’t have a delay problem.
You have a manual decision problem.

Supply Chain Delays Are Inevitable. Manual Reactions Aren’t.

Most manufacturers and distributors are still reacting to delays like it’s 2015.

Planners manually reschedule orders.
Logistics teams call carriers and “wait for updates.”
Managers decide expedites in a Teams thread.
Finance finds out three days later when the bill hits.

Everyone’s working hard. Nobody’s working smart.

Your systems can tell you a shipment’s late.
But they can’t tell you what to do about it.
You have nice dashboards but they are usless. 

Don't get me started on IBP or other parts. 

That’s the missing layer:
Automated decisioning.

What Automated Decisioning Actually Means

It’s not “AI.”
It’s not a magic dashboard.
It’s simply this:

When an event happens (a delay, shortage, or disruption)…
-It’s detected in real time.
-Your rules or models evaluate the impact.
-The next best action happens automatically.

No waiting for planners. No “what should we do?” meetings.

The human becomes the exception handler, not the bottleneck.

1. Map Where Decisions Actually Happen

Don’t start with tech. Start with truth.
List every time someone manually reacts to a delay:

“If supplier X is late, we expedite with Y.”
“If truck misses cutoff, alert customer service.”

2. Build a Real-Time Event Backbone

You can’t automate what you can’t detect.

Stream your key operational events (shipment, supplier, production) into a central.

No delay should be trapped in a spreadsheet or silo.If data doesn’t move in real time, your decisions never will. 
That’s your “nervous system.”

No more email alerts. Real-time triggers instead.

3. Codify the Decision Logic

Turn tribal knowledge into rules

Don’t start with some complex AI or ML model. 
Turn common sense turned into code. 
Don't force AI where it makes no sense. 

3.Connect Decisions to Action Systems

A decision without execution is just another alert.

Connect your logic from central source of truth ERP, TMS, MES, or CRM so the system can actually do something:

Reschedule production.
Reroute shipment.
Notify customer automatically.

That’s how you close the gap between insight and impact.

Here’s what happens in real life:
Delay event detected instantly.
System calculates risk.
Auto-reroute triggered.

A Tier-1 auto supplier we worked with cut reaction time from 12 hours to under 10 minutes.

Every delay costs money.
But slow decisions cost more

Your 2026 AI strategy is already dead. You are missing the one thing that actually makes AI work. This year, the winners won't be companies with the best AI model. But those who have context.

“We’re going all in on AI this year.”
Except 9 out of 10 of those “AI strategies” are already dead.

Not because of bad models.
Not because of missing talent.
Because your data doesn’t know how to talk to itself.

Your CRM calls it “customer.”
Finance calls it “account.”
Ops calls it “site.”
And your AI? It just calls it confused.

For years, companies believed if they centralized data, they’d create intelligence.
So they built warehouses.
Then lakes.
Then lakehouses.

Now, they’re realizing:

→ Centralized ≠ Connected.

Your data warehouse tells you what you have.
Your knowledge graph tells you how it all connects.

Without that connective tissue, your AI models are operating blind:

Forecasting demand without understanding product hierarchies.
Predicting churn without linking customer interactions.
Detecting anomalies without connecting to operational events.

That’s why AI pilots look great in demos, and fail in production.
Because AI isn’t about bigger models.
It’s about connected meaning.

If you want AI that works, stop chasing models.
Start teaching your data to speak your business language.

Here’s the 5-step playbook:

1. Build a Business Ontology (Not Another Dashboard)

Map how your business actually thinks: customers, orders, assets, suppliers, SKUs.

“Product” → linked to “Manufacturing Batch” → linked to “Quality Event”

This isn’t IT work. It’s strategy work.
Run a 2-day ontology sprint with your functional leaders.

2. Link Your Structured + Unstructured Data

AI doesn’t just learn from tables.
It learns from emails, PDFs, call notes, logs.

Maintenance logs mention “compressor 14,” which is tagged in OT data as “C14.” 

Connect them through entity tagging and relationships.

3. Create a Semantic Layer (Your AI Translator)

Your existing data catalog describes data. Knowledge graph interprets it. 
This layer turns relationships into queryable context.

So when you ask, “Which suppliers caused downtime in the last 6 months?",  it actually answers correctly.

4. Anchor Every Data Product in the Graph

Your “Customer 360” dashboard isn’t 360 unless it’s connected across domains.
Graphs make products dynamic and self-healing when structures change.

Like 
Dynamic supplier risk scoring (linked across purchase orders, logistics) 

5. Leverage proprietary data 

Those who gone win are not with best LLM models or AI tools. 
But those who leverage their proprietary data. 
Every data point, variable is a edge in your market. 

If your AI strategy doesn’t include a knowledge graph, it’s not a strategy.

You can’t build intelligence on top of disconnection.

The future of AI won’t belong to the companies with the best prompts.
It’ll belong to the ones whose data knows how it relates to itself

90% of manufacturers are jumping straight to "AI". But But takes them 5 days and 20 meetings to explain why a number changed. Half of operators are still in excel. You are not 'data-driven' and here is how to change that in 2026

You’ve got Power BI reports, a shiny Snowflake instance, maybe an AI pilot running in a corner of one plant.

But when I ask execs one simple question

“Can you trace one KPI from source to decision, end to end?”

90% of the time, the answer is no.

Or it takes 5 days, 5 people and 20 meetings. 

That’s not a technology failure.
That’s a maturity gap.

Everyone wants to jump straight to “predictive maintenance” or “AI optimization.”

But manufacturing, logistics orgs data maturity doesn’t work like that.
You don’t go from chaos to AI in one quarter.

You pass through five unavoidable stages each one with different pain, ownership, and ROI.

Here’s how to know where you actually are 

Stage 1: The Dark Ages The Blind Factory

Data exists everywhere, but no one uses it.
It’s in SCADA, MES, Excel sheets, and people’s heads.
Operators rely on intuition, not information.

Symptoms:

Endless firefighting
Paper logbooks
Reports created manually once a week
manufacturing or supply chain. 
Your job is to assign ownership and Inventory your systems.  

Stage 2: The Conflicting Era

You can see data now… sort of.

Dashboards pop up everywhere, but they all tell different stories.
Finance shows 92% OEE. Ops shows 78%. Nobody knows who’s right.

Symptoms:

“Multiple versions of the truth"
Dashboard fatigue
No agreed KPI definitions

Your job is to standardize definitions, resurface and redundant dashboards or sheets. 

Stage 3: Diagnose The Context Stage

This is the turning point.
You finally understand why things happen not just what happened.

Symptoms:

You can link downtime to shifts, batches, or recipes
Teams start trusting reports
OT and IT actually talk to each other

Your job here is to build your context layer.
Context isn’t metadata it’s meaning.
It’s how you connect sensors → equipment → process → cost.
Without it, you’ll never scale.

Stage 4: Predictive Stage

Now you can see ahead.
Predictive maintenance, quality forecasting, energy optimization they’re real.

But this is where most teams stall.

Why?
Because they forgot to govern context.
Models degrade. Data drifts. Trust erodes.

Your job here is version your context.
Treat it like code. Keep feedback loops tight.

Stage 5: Operationalize 

This is the endgame. Closed-loop optimization.
Data drives action automatically, end users able to access the data with no bureaucracy.

Not everything is a 'dashboard'

It’s about culture, governance, and compounding context

Data maturity is leadership work, not technical work.
And once you know your stage, you can stop chasing hype and start building capabilities that last and impact today and set you for long term success .

Last week COO of a $250M manufacturer said to me: “We’ve got SAP, ERP, WMS, IoT but when something breaks, can’t see it.” It took them 7 meetings and 20 spreadsheets  to uncover root cause or trace a defect. Here is what most manufacturers and supply chain miss 

He wasn’t exaggerating.
His team had more dashboards than decisions.
And they still couldn’t trace a defect without six people, three spreadsheets, and a week of digging in meetings. 

Your problem isn’t visibility.
It’s traceability. 

Manufacturers or supply chain orgs don’t fail because they lack data.
They fail because their data lives in isolation.

MES knows what happened.
ERP knows what order it was for.
SAP for you fleet optimizations. 
QMS knows which batch failed.
Maintenance knows when the line went down.
IoT knows the temperature and vibration.

But no one, and I mean no one can tell you how those things connect.
That’s why your teams spend hours pulling CSVs, merging spreadsheets, and arguing over “whose data is right.” 

You’ve built a data warehouse. You’ve visualized KPIs.
But you never built a knowledge graph.

Everyone’s obsessed with collecting more data.
No one’s focused on connecting it.

You don’t need another system of record.
You need a system of relationships.


Think of it as your factory’s digital DNA:
Machines → Batches → Materials → Suppliers → Quality → Customers in one connected truth.

Forget OT + IT. It’s bigger then that. 

The best manufacturers and supply chain organizations, retailers we work with don’t chase “more data.”

They chase connected data.

Example:

“Show me all defects in the last 90 days tied to Supplier X’s batch, produced on Line 3 during night shift.”

If you can’t answer that in seconds, you don’t have traceability.
You have data chaos with prettier charts.


You stop asking “what happened?” and start answering “why.” 

We built a knowledge graph for a precision parts manufacturer last year.
Before that, root cause investigations took 7 days on average.
Seven.

After 90 days, they could trace a defect to a supplier in 30 minutes.

They didn’t buy a new platform.
They just connected MES, ERP, and QMS data into a graph model.

That one change saved $1.2M in rework and scrap in the first quarter.
Not because of AI.
Because of context.

The smart ones don’t start with tools.
They start with relationships.

1. Map Entities and Relationships .
2.Connect Existing Systems into Graph
3. Tie to urgent business needs and questions. 
4. Data has to act. Stop reporting only dashboards.

Everyone in manufacturing is chasing AI.

But AI is useless when your data can’t explain itself.
If your systems can’t talk to each other, your machine learning model will just automate confusion.

Because until your data knows how it connects, it doesn’t matter how much you have, it’s just noise.

If your team still needs a week to find the cause of a root cause, you don’t have a visibility problem.

You lack context

Over 80% of manufacturing data modernization's fail. The moment you add a new source, tool, plant everything breaks. You built architecture for stability. But your business is built for growth. 

When your company doubles in plants, products, or markets- your data foundation collapses under its own weight.

Pipelines break. Metrics drift. Dashboards contradict each other.
And AI pilots stall because nobody trusts the inputs.

It’s because your architecture was built for stability, not growth.

You start your “modernization” journey with a clear plan:

- Move to the cloud.
- Centralize ERP, MES, and historian data
- Build a few “data products
- Launch shiny dashboards and AI PoCs.

Then growth hits.

-You acquire two plants with different historians.
-One uses OPC-UA, the other Modbus.
-Every site has its own definition of “OEE.”
-Finance calls it “downtime.” Ops calls it “line changeover.”
-Suddenly, your “single source of truth” has 10 dialects.

This constantly happens in asset heavy industries. 

And now your so-called “modern data stack” becomes the same mess just with higher AWS costs.

This is where 80% of manufacturing modernization efforts stall.
Because the foundation wasn’t designed to flex with growth.

What the top-performing manufacturers do differently:

They don’t chase tools.
They build systems that absorb chaos.

Here’s how they do it:

1. Design for expansion by default.

Don’t hard-code your integrations or models.
Build metadata-driven ingestion and transformation frameworks that can adapt as new systems, schemas, and sites come online.

If your pipelines can’t onboard a new plant without rewriting code, you’re not modernized- you’re fragile.

2. Harmonize semantics before harmonizing systems.

If “OEE,” and “margin” mean different things in every source or team no data lake can fix that.

Start with a business dictionary.
Run cross-site metric harmonization meetings.
Document it in your catalog.
Make the data mean the same thing before you centralize it.

3. Integrate OT early, not later.

Stop treating historian and sensor data as a “Phase 3.”
That’s the data that drives every AI and predictive use case. 
If it’s not integrated, you’re modernizing for reporting not intelligence.
Of course align it to business impact. 

4. Scale ownership, not just engineering.

Central data teams can’t keep up with hyper-growth .
Assign data stewards and champions by domain (Ops, Quality, Supply Chain).
Give them visibility into data trust in format they like. 
Get data in domain expertise hand, without them needing to email or slack anyone. 

That’s how you scale governance without bureaucracy.

5. Treat modernization as a capability, not a project.

There’s no “done” state.
Your platform, models, and governance must evolve with every product line, plant, and AI use case.

If your foundation can’t handle new systems, or new product lines- you haven’t modernized anything.

You’ve just moved legacy to the cloud.

taplio

The Battle

Nikolay Valiotti

Ali Z

Followers

Posts

Engagements

Likes

Performance ScoreNikolay Valiotti's LinkedIn Performance

Performance ScoreAli Z's LinkedIn Performance

Ready to grow your LinkedIn presence?

taplio

The Battle

Nikolay Valiotti

Ali Z

Followers

Posts

Engagements

Likes

Performance Score.css-goub6s{font-size:var(--chakra-fontSizes-md);font-weight:500;color:var(--chakra-colors-gray-600);display:block;margin-top:var(--chakra-space-1);}Nikolay Valiotti's LinkedIn Performance

Performance ScoreAli Z's LinkedIn Performance

Ready to grow your LinkedIn presence?

Performance ScoreNikolay Valiotti's LinkedIn Performance