Freelance Growth, Paid | +100 entreprises transformées | +5 millions générés | Formation Growth

Passionate about the human relationship and the new technologies, I acquired several years of experiences as a commercial. Over the years I have noticed an evolution of the buying process, buyers are primarily looking on the internet before contacting a company. Aware of the difficulties companies have in being online, I decided to help them in this transition.


Jeremy LEPLU

Marc Simoncini refuse d'investir dans un domaine qu'il ne connaît pas. 

Pendant longtemps j'ai trouvé ça fermé d'esprit. 

Puis j'ai lancé 3 projets dans des domaines que je ne maîtrisais pas : 
- Un outil pour payer son hébergement Webflow moins cher
- Une app de home staging par IA 
- Un outil SEO 

À chaque fois le même film. 

Je passais des semaines à comprendre les bases du marché. 
Les vrais problèmes. Les codes du secteur. Les gens à qui vendre. 

Pendant que j'apprenais, mes concurrents exécutaient.
J'avais l'énergie. Les skills techniques. La motivation.

Il me manquait le seul truc qui ne s'apprend pas en 6 mois : le contexte. 

Aujourd'hui je construis Smart Caller. 
Un setter IA qui contacte et qualifie les leads B2B en moins de 2 minutes. 

Et pour la première fois, je ne découvre rien. 
6 ans en growth B2B. 100+ entreprises accompagnées. Des milliers de leads générés.

Je connais le problème par cœur : 78% des leads signent avec le premier qui répond. Et la majorité des boîtes mettent 24-48h à rappeler.

C'est mon unfair advantage. Pas une techno. Pas un brevet. Pas du funding. 

Juste 6 ans à voir le même problème tous les jours chez tous mes clients. 

Le meilleur projet c'est pas celui qui t'excite le plus. 
C'est celui où t'as un avantage que personne ne peut rattraper.

Le setter IA pour les équipes B2B | Qualifie, relance et book vos leads en <​ 2 min | +100 entreprises accompagnées | Fondateur Smart Caller

Ton client dépense 15 000€/mois en acquisition. Ses campagnes génèrent 300 leads.
CPL : 50€. Correct pour du B2B.

Maintenant regarde ce qui se passe après.

Sur ces 300 leads, 40% ne seront jamais contactés. Ça fait 120 leads à 50€ pièce. 6 000€ jetés à la poubelle. Chaque mois.

Sur les 180 restants, la moitié sera contactée après 48 heures. À ce stade le lead est froid. Taux de conversion divisé est par 10. Autant dire que ces 90 leads à 50€ ne valent plus grand-chose. 4 500€ de plus gaspillés.

Il reste 90 leads traités "correctement". Mais correctement ça veut dire quoi ? Un mail template envoyé le lendemain. Pas un appel. Pas une conversation.

Au final, sur 15 000€ investis, ton client tire peut-être 2 000 à 3 000€ de valeur réelle.

Et c'est toi qu'il blâme.

Ce calcul je l'ai fait des dizaines de fois avec mes clients. À chaque fois la réaction est la même "Comment on règle ça ?

La réponse c'est pas plus de budget. C'est pas de meilleures campagnes. C'est pas un nouveau canal d'acquisition.

La réponse c'est de traiter chaque lead comme ce qu'il est : un billet de 50€ qui perd de la valeur chaque minute qui passe.

C'est la philosophie derrière Smart Caller. Chaque lead est contacté en moins de 2 minutes. Pas parce que c'est un beau chiffre. Parce que c'est là que la valeur est maximale.

Tu dépenses déjà l'argent pour générer ces leads. La question c'est : combien tu laisses sur la table ?

J'ai généré des milliers de leads pour mes clients.
Et j'ai regardé 80% d'entre eux mourir dans un CRM.

Pendant 6 ans, j'ai monté des stratégies d'acquisition pour +100 boîtes B2B. LinkedIn Ads, cold outreach, ABM, lead magnets… Je sais faire rentrer du lead.

Mais il y a un truc qui m'a rendu dingue pendant des années.

Tu livres 200 leads qualifiés à un client. Et 3 semaines plus tard ? 150 n'ont jamais été rappelés. 30 ont reçu un mail générique. 20 ont été traités correctement.

Pendant longtemps j'ai cru que c'était un problème de volonté. Qu'il suffisait de secouer les commerciaux.

J'avais tort.

Le vrai problème c'est la guerre silencieuse entre le marketing et les sales.

Le marketing génère des leads. Les sales disent qu'ils sont pas qualifiés. Le marketing dit que les sales ne rappellent pas assez vite. Personne ne bouge. Les leads refroidissent. L'argent est gâché.

J'ai cherché une solution pendant des mois. Ça n'existait pas.

Alors je l'ai construit. Ça s'appelle Smart Caller.

C'est un agent IA qui contacte et qualifie tes leads par SMS, WhatsApp ou Instagram en moins de 2 minutes après leur inscription.

Parce que la donnée est claire : un lead contacté dans les 2 premières minutes a 9x plus de chances de convertir.

La durée moyenne d'un contrat agence marketing-client en France : 8 mois. 

8 mois. Même pas un an. 

La raison numéro 1 de rupture ? "On ne voit pas les résultats." 

Pas "le CPL est trop haut." 
Pas "les créas sont nulles." 
Pas "le targeting est mauvais." 
"On ne voit pas les résultats." 

Traduction : le client ne voit pas de RDV. Il voit des leads dans un tableau. Pas des gens en face de lui. 

Et c'est là que l'agence a un angle mort. 
Elle optimise le CPL, le CTR, le taux de conversion landing page. 
Tout ce qui est AVANT le lead. 

Ce qui se passe APRÈS le lead, le temps de rappel, la qualification, la prise de RDV c'est "côté client."

Sauf que le client s'en fout de savoir de quel côté est le problème. 
Il veut des RDV. Il n'en a pas. Il churn. 

Les agences qui contrôlent toute la chaîne du clic au RDV ont moins de churn. 
Parce que le client voit du concret. 

Celle qui s'arrête au lead vend un produit à moitié fini. 
Et elle le paie tous les 8 mois.

I'm a founding partner at Highline Beta. 

Highline Beta is a corporate venture studio, working with large companies to innovate beyond the core. We help build new ventures internally and externally, and invest in pre-seed/seed stage startups that have a meaningful connection to our corporate partners.

I'm an entrepreneur, investor, product expert and author. I've worked in startups (VarageSale, GoInstant and others) and large companies (Salesforce), with the knowledge and experience to bridge the gaps between them.

In 2013 I co-authored Lean Analytics (http://leananalyticsbook.com), to help startups and large enterprises build better products faster with data. I've spoken globally and consulted with many companies around implementing Lean Startup and Lean Analytics.

In 2010 I co-founded Year One Labs, a pre-seed startup accelerator. One of our portfolio companies, Localmind, was acquired by Airbnb. Year One Labs was the first accelerator to apply a rigorous Lean Startup methodology.

Areas of expertise: venture studios, corporate venture studios, venture building, corporate innovation, startups, business model innovation, Lean Analytics, Lean Startup, product management, design thinking, startup investing, startup accelerators & recruiting.

Ben Yoskovitz

We're not prepared for the destruction of junior roles because of AI.

I'm definitely not. I have two adult kids in university. (BTW, they're brilliant and 100% worth meeting and hiring!)

If entry-level jobs are replaced by AI, what are we expecting graduates to do? Become influencers? Start podcasts about their feelings? 

Let's be realistic.

We need young people joining the workforce and levelling up. How else is true expertise shaped? How else do they gain the experience needed to uncover new opportunities, chase new ideas and create new value? We're not all going to sit around and prompt AI mindlessly.

AI will replace jobs. But before we destroy everything, I think AI should focus on training young people to be better at their jobs.

Use AI as the ultimate tool for increasing how quickly new employees become effective. Training is expensive and time consuming. Improving time to value for new employees is one of the most underrated metrics in any organization.

I didn't have AI to help train me to become a product manager. But new product managers can leverage it in incredible ways, if it's done deliberately. Same is true for any industry, role or type of worker.

I want AI to be the ultimate training tool. Something that helps people, especially young people, learn and level up faster. So they can become craftspeople, not drones.

If we get this wrong, we feed tons of smart, capable people to the wolves and eliminate any chance of unlocking their potential.

And selfishly? My kids will graduate university in 1-3 years. I'd like there to be somewhere for them to go. Somewhere that gives them the tools to excel, explore and expand their capabilities while creating value. Or maybe they'll start companies...

Founding Partner at Highline Beta | Author of FocusedChaos.co

What have you built using AI this week?

My list:

1️⃣ Worked on my LLM Optimizer: The tool scrapes a website and evaluates it against best practices for LLMs and SEO. There are a lot of tools out there that do this, many startups that have raised capital. I decided to try it on my own.

I'm learning a ton about best practices for site architecture and AEO, GEO, SEO. All the "EOs". 😅 

If you're interested in testing this out, let me know. Drop a comment below.

2️⃣ Created more Claude Skills: I'm late to the Claude party, but decided to invite myself and see what's going on. Claude is good. I'm finding it's already stronger as a thought partner for writing content, analyzing things, etc. (Btw, this post is 100% handwritten by yours truly! That alone should give it a boost!)

3️⃣ SEO Optimization Scripts: Based on what I learned about Highline Beta's website through the LLM Optimizer (which also does lightweight SEO analysis), I built a few scripts using Claude Code to optimize basic SEO stuff: titles, meta descriptions, etc. Simple stuff, but integrated into Webflow and a library of best practices, it works pretty seamlessly, which is easier than doing a ton of manual fixes.

4️⃣ Carousel & Image Designer Using Claude: Pretty simple, but I wanted to take blog, newsletter and LinkedIn posts and generate images or carousels. These need to be on brand. If you check out my content, you'll see it's getting better and better. Carousels are generated automatically from Claude once I've written the copy. It's using elements from the website, our brand colors and logo. Again, this is fairly straightforward, but to do a carousel well before was painful.

The goal isn't to optimize everything perfectly. Or manage a slew of agent bots aiming to spew spam content and run my life.

The goal is learning.

And optimizing some things. 😊 

Happy building!

40 years ago McKinsey defined the concept of the "T-shaped" person. Since then, this has been (for the most part) the definition of an ideal candidate for many roles.

* The vertical bar of the T = deep expertise in one domain. Your specialty. The thing people hire you for.

* The horizontal bar = enough breadth to collaborate across disciplines. The ability to sit in a room with people from other functions and actually contribute to the conversation.

The core idea: be great at one thing, and be collaborative enough to work well with people who are great at other things.

When Tim Brown from IDEO popularized the concept, he described the horizontal bar this way (paraphrasing):

"It’s the disposition for collaboration across disciplines, built from two things. First, empathy. The ability to stand in someone else’s shoes and understand a problem from their perspective. Second, enthusiasm for other disciplines, 'to the point that they may actually start to practice them.'"

AI is blowing this up.

Why?

Because AI makes it possible for everyone to actually start doing other things. In fact, the way the world is going, employers will demand that specialists become generalists.

It used to be tough to learn other disciplines, gain confidence in them, and try a bunch of things. AI makes that much easier.

I'm building & trying things I never thought possible even a year ago. From simple automations to full-blown web applications. I'm deep diving concepts like LLM optimization through doing, not being a passive learner or relying on others.

The T-shape is changing into something else: where the horizontal bar BECOMES the thing that actually matters.

AI compresses learning time. It gives everyone a chance to become functionally capable in something new. Right now. But it doesn't auto-install expertise.

Specialties still matter, but I think we'll see a big shift towards hiring competent, curious, "aggressive" generalists that are comfortable figuring things out on their own.

How does this change the skills you need? What sits in that fat horizontal T-bar?

1️⃣ Comfort with ambiguity: People who need clear instructions before starting anything will struggle. Specialists get locked in on "what they do" without the ability to explore further.

2️⃣ Speed of experimentation: People who aren't excited about trying, failing & learning by doing will become obsolete.

3️⃣ Judgment about AI outputs: AI doesn't have all the answers. But you can't figure out what it's good at or where it drifts, without using the tools actively.

4️⃣ Cross-domain curiosity: People who aren't curious -- and then take ACTION on that curiosity -- are toast. If you say, "I've always been interested in X," but show no signs of actually learning X (through doing), no one will hire you.

5️⃣ Self-directed learning: Learning is going to become a huge measure of progress and ability, more than it ever has been.

Specialties still matter, but your generalist capabilities will matter more.

When do we get exhausted by startup pitches that mention "AI agents" every 2 seconds?

I'm already there.

I get it. You use AI.

So does everyone else.

I'm not sure it's the differentiator you think it is. Maybe you're obligated to say it, and repeat it over and over, because everyone is looking for AI and "AI agents" are hot...but sigh. 

I'm not convinced it's doing what you think it's doing.

While buyers may also feel the pressure from their bosses to "use AI!" your agentic platform of magic-sauce could very well be a checkbox on their list of requirements and not the differentiator that wins deals.

Everyone has agents.

Most of them are mediocre.

Even more of them are simply workflows (which is OK!)

I get the hype. I believe in Agentic AI and a pretty automated future. But the endless pitches are exhausting.

Today I'm sharing a special "What I'm building with AI" post.

This one is for my wife, Jessica Howard.

Over the last year or so, Jess has been focusing more on her art business.

Art is a tough business.

But art is awesome.

We need art.

She's working on getting her art into more galleries, shows and events. It takes time to find the "Requests for Artists", figure out which ones are a fit, and submit applications.

Of course, I thought, "AI can improve this process."

So I built a tool. 😃 

The tool searches for artist requests across specific geographies, scores them (based on Jess's type of art and a few other parameters) and makes it easier for her to apply. 

More applications = more chances to get into galleries and shows = more exposure = more sales.

This took me less than an hour to build.

If v1 works, I'll keep iterating on it, adding more capabilities to streamline various operational tasks she has to do as an artist; freeing her up to spend more time creating. This is a perfect use case for AI-assisted building. 

Btw, if we've ever done a virtual meeting before while I was at home, you saw her art behind me. Or if you've been to Highline Beta's office (Highline Hub), the art there is hers too.

She also has a Shopify store, which I'm happy to share with you (hint, hint). And coming soon, keep an eye out for her art in galleries across GTA.

Why should big companies build new ventures?

There are a host of reasons, but I want to focus on one today: Creating differentiated value propositions.

In a bunch of industries there are a lot of companies that sell undifferentiated products. Think: insurance, banking, legal, accounting, etc. "A mortgage is a mortgage is a mortgage." Not 100% true, but pretty close. Or, "Every life insurance policy is the same, I can't tell the difference." I've done research in this space and that's pretty much what people say.

So how do you differentiate in a world of vast competition, where everyone is selling something very similar?

Better Superbowl ads?

Maybe.

I think there's another way: Building new ventures.

New ventures give you, as a big company, the opportunity to solve new problems for your users/customers. You can create differentiated value propositions, outside the core, in a way that engages customers more actively. 

Very few people engage with their insurance company on a regular basis. At renewal, and when there's a problem. Not good enough.

But those same people, who buy insurance, have a host of other, related but separate problems and pain points that an insurance company may have a right to solve.

So big companies can use new ventures with new value propositions to:

1️⃣ Retain existing customers and increase revenue/customer. Those are good things.

2️⃣ Acquire new customers and then convert them into buying from the core. That's also a good thing.

For this to work, the new ventures cannot be glorified lead funnels. New ventures aren't simply new marketing campaigns. 

New ventures need to:

✅ Solve a net new problem in a meaningful way
✅ Be able to stand independently so they can create net new value without being pushy about up/cross-selling
✅ Survive on their own, with a viable business model

If these conditions aren't true, new ventures tend to fail. They become money pits and someone will eventually say, "Couldn't we put up more bus ads, or do more Instagram posts?"

If you can meet those conditions and find the appropriate (not pushy) way to up/cross-sell between new ventures and the core (and vice versa) you have a powerful differentiator from your competition. 

Suddenly you have many more entry points into your company, because each venture is a net new way of engaging users/customers. The core is a massive asset, because it has the biggest number of customers (compared to the new ventures) and you can create many more meaningful touchpoints.

This model works.

It's not easy, but it's doable.

And it differentiates companies in a sea of sameness.

This week in my vibe coding adventures, I built:

1️⃣ A Web Page Interlinking Tool: I've wanted this for a while and finally realized, "Maybe I can build it myself?" 

Interlinking between pages on your site can strengthen SEO, and to some extent LLM optimization. You pick a handful of pillar pages that you want everything to link up to in a hierarchy, some key words + variations to use, and link everything. 

It's annoying to do manually. With Claude Code I designed a strategy, based on best practices, and then an interlinking tool that scrapes every page on our site, identifies the key pillar content and secondary content of importance, and proposes hyperlinks. Through the Webflow API it can add the hyperlinks and publish the content.

It works really well and was crazy easy to build. I'm sure there are more sophisticated ways of doing this, but I think this is a decent start. I had looked awhile ago for a tool that did this and couldn't find one.

2️⃣ Key Takeaways Blog Optimization Tool: Another tool designed to help with LLM optimization. It reads a blog post, identifies 3-4 key questions from the post and puts them at the top of the post in a "Key Takeaways" section. 

It also automatically adds FAQ Schema to the blog post so that we're following best practices on technical site structure.

It publishes back into Webflow (you can review first) into a custom field so it doesn't mess with the content of the post itself. I found it works quite well and the content is solid.

3️⃣ LLM Optimizer: I added more functionality to the LLM Optimizer. Now it runs Google and Bing searches against prompts to see how you rank. I'm not trying to replace any of the great tools out there for this sort of thing, but just give people a quick heads up on how their site is performing. 

The LLM Optimizer now:

* Scans your whole site and provides a score based on a few key criteria
* A lot of it is focused on technical architecture
* It provides recommendations on improvements to make
* It runs bi-weekly tests against 4 LLMs (Perplexity, OpenAI, Gemini and Claude) to see how you show up for a bunch of prompts and provides a report
* It does a bit of SEO assessment too, with recommendations on fixing meta titles/descriptions and more
* It's far from production-ready, but as a test app for Highline Beta and our portfolio it's quite solid. And I've learned a ton about optimization

4️⃣ Granola to Google Drive Integration: I use Granola for meeting notes. I added the ability to move those notes into a Google Drive doc through Zapier, and then from Google Drive, sort the notes into the right folders based on basic mapping. Not the most elegant solution but it works, and I think it will save me time managing meeting notes and even tracking meetings (what was said, agreed to, etc.) more easily.

Most of this is done with Claude Code and running on my local machine. The LLM Optimizer is done using Lovable.

Happy building!

AI has turned everyone into product managers.

A few years ago, people declared product managers obsolete. The work to figure out what to build, managing the build, getting the product into users' and customers' hands and collecting feedback would be done by...other people. 🤷‍♂️ 

Product management went through an existential crisis.

What is product management?

Is it project management? A lot of people think so.

Is it a waste of time? Nope!

With AI coding assistance everyone can build products. That's a good thing, because more people with more ability to make stuff is positive. But the repercussion is a lot of useless and crappy products.

Enter: Product Managers. 💪 

Product management is the job of building the right thing for the right user/customer at the right time. 

The development of a product is way easier, but figuring out what to build and how to distribute it hasn't gotten any easier. If anything those two tasks are harder. With 100x more products flooding the market, users/customers will be overwhelmed by choice. Differentiating becomes harder. Getting your message out is harder. 

Product managers are supposed to:

✅ Validate a real problem; an unmet need where there's legitimate whitespace (good luck finding that!)
✅ Build the right-sized product (not endless features because you can)
✅ Get the product into the hands of early adopters and confirm it's creating value (and that you can change their behavior)
✅ Iterate and learn constantly throughout this process

Sounds like EVERYONE should be doing that. 

Now everyone is a product manager. Who'd a thunk it?

What did you build this week with AI?

I continue to dive in and experiment, learn and try new things:

1️⃣ Investor Portal: I've been working on this for a couple weeks. The tools for managing a fund are outdated and expensive. We run a small fund. So we thought it would make sense to save money and build something ourselves. Instead of Lovable, I went with Claude Code + Vercel (+ Resend for emails). This is going live at the end of the month, so it has to be production-grade compared to other prototypes & experiments I've built. That means staging + production servers, deep dives on security and live data.

2️⃣ New Blog Post Indexing Tool: Continuing on my efforts to deep dive on LLM optimization (AEO, GEO) and SEO, I built a simple workflow automation with Weblow + n8n that automatically pings Google and IndexNow when we publish a new blog post. The best practices suggest that it's worth proactively "telling" the search engines of new content.

3️⃣ Market Sizer: In the Highline Beta Toolkit, I added a new tool: the Market Sizer. With a few inputs on what you're looking to build, target ICP and geography, it calculates TAM, SAM and SOM, with full references to how it's finding the data. It also does a bit of "bottom-up budgeting" (which I've always felt is more important than top-down market sizing), based on price point, market penetration, conversion numbers, CAC data and more. It aggregates relevant data it can find to build a "lightweight" business model.

This is live today! If you want to use it, comment below.

4️⃣ Trend Analyzer: This is not live yet, but it will be another tool in the Highline Beta Toolkit (which already includes: Starburst, Assumption Tracker, Problem Validator, Concept Statement Builder, Problem Interview Generator, Solution Scorecard and Pre-Mortem Analysis -- all free to use).

The Trend Analyzer looks for your target audience's communities and uncovers trends, interesting topics, etc. that may be valuable. It works primarily with Reddit, but can work with some web communities too, and I'm hoping to get it working with FB Groups and other online "gathering places" as well.

5️⃣ HLB Toolkit MCP Server: The tools we've built are discrete things that anyone can use quickly. But I want to "string the tools" together into an interface that allows me to kick off a project, engage the right tool at the right time, and maintain context. So I'm building an MCP Server that'll integrate with Claude (and ChatGPT) that uses the tools when appropriate during a project/conversation. The idea would be that I can jam with Claude on a project, and it'll know when to use a specific tool we've built. 

Effectively, each tool goes from being a discrete web application to an agent.

This isn't live yet, but I'm testing it now.

➡️ So what have you built this week? (I know Meir Dick built stuff. 😃) Who else?

Most corporate innovation programs aren't failing because of bad ideas.

They're failing because someone turned innovation into a factory.

I've watched this happen more than once. A well-intentioned leader designs a rigorous process: stage gates, deliverable templates, mandatory outputs at every step. Then they hire entrepreneurial people to run it.

Those people last about six months before they're exhausted, demoralized, or gone.

Here's what overly rigid innovation processes actually do:

1️⃣ They shift focus from outcomes to outputs. 

People stop asking "did we find a real problem?" and start asking, "Did we check all the boxes?" 

I worked with one client who specified exactly how many customer interviews had to happen at each stage. Teams hit the number. They didn't actually learn anything.

2️⃣ They drive out the very people you need. 

You can't attract entrepreneurial talent and then strip out all the entrepreneurial autonomy. That's not a culture problem. It's a design problem.

3️⃣ They create the illusion of progress. 

Things get built. Presentations get made. Reviews get scheduled. But no one can tell if any of it is creating real value. 

The process becomes the product.

Here's what actually works:

✅ Have a clear evaluation framework. 

What spaces should you be exploring? 

What advantage does your core business bring to a new venture? 

These are useful filters.

✅ Have a simple funnel. 

Problem finding to ideation to concept testing to MVP. 

But be honest about how many hoops you're forcing people to jump through.

Trust your people with tools, not checklists. Give them Lean Canvas, Design Thinking, JTBD. Then let them decide what to use and when. If they can't figure it out, they probably shouldn't be building ventures.

✅ Accept that you can't measure your way to victory. 

New ventures are messy. Track the vital signs. Build a basic portfolio framework. 

But you cannot systematize uncertainty out of existence.

✅ Corporate ventures aren't widgets.

Process is a tool. Not a goal.

What does innovation governance look like at your company? Is it helping your team move, or slowing them down?

Innovation checklists killing the best innovators

If AI is levelling the playing field, what actually differentiates you?

Everyone has access to the same AI tools now. Same ability to generate, summarize, research and build.

Michelle Moon wrote something that stuck with me: in the AI era, domain expertise is the last defensible moat. As general-purpose models get better and cheaper, the technology stops being the differentiator. What's irreproducible is lived experience: knowing how customers think, where workflows actually break, and which "best practices" fall apart in the real world.

She's right. And I'd add one thing: domain expertise doesn't just show up as knowledge. It shows up as instinct.

It's the quiet voice saying "we've seen this before" when the data looks clean but something feels off. It's knowing what questions to ask before you run the experiment. It's recognizing a false positive before it costs you six weeks. It's understanding not just what to do, but when, why, and what not to do.

That's not in a dashboard. It's earned through repetition, failure, and time spent deep in a specific problem space.

Domain expertise is actually an operating system for intuition.

We're applying this thinking directly at Highline Beta. One of the reasons we're building vertical venture studios is because we believe deep domain expertise in a specific industry creates better outcomes. It's not enough to know how to build ventures. You need to know the edge cases, the buyers, the regulatory landmines, the players who matter. That knowledge compounds. We're betting on domain expertise as a structural advantage, not just a nice-to-have.
The risk is that AI flattens everyone to the same starting line and we mistake that for equality. The person who has spent years in insurance, or recruiting, or supply chain logistics has something the model doesn't.

That's the moat. Develop it deliberately. Don't let the availability of AI tools convince you that you can skip the hard work of building it.

I've spent 15+ years preaching build, measure, learn. I co-wrote a book on analytics. I genuinely believe data is one of the most underrated communication tools in any organization.

And I still got burned by trusting it too much.

Early in my career, I worked with a startup where one user segment was converting way better than all the others. The data was clear. Double down, right?

But something felt off. Usage patterns weren't sticky. Behavior was erratic. Retention was soft. My gut kept saying "this isn't right" but the numbers kept saying "go."

Six weeks later that segment evaporated. Low-quality third-party traffic we hadn't accounted for. A false positive we had missed entirely.

The mistake wasn't using data. The mistake was ignoring the instinct that something wasn't adding up.

In Lean Analytics, Alistair Croll and I wrote: Instincts are experiments. Data is proof.

Data doesn't make decisions. It's an input into decision-making. Ultimately you decide what to focus on, how to run experiments, and when to just ship something versus running a more rigorous process. If everything was blindly data-driven, you'd be outsourcing your judgment entirely.

And judgment is increasingly the thing that actually matters.

In a world where AI gives everyone access to the same tools and information, the differentiator isn't who has the data. It's who knows what to do with it. That's domain expertise. It's the quiet voice saying "we've seen this before" or "this looks good on the surface but it won't last." It's not in a dashboard. It's earned through repetition, failure and time.

Domain expertise is actually an operating system for intuition.

The goal isn't to choose between gut and data. It's to use your instincts to ask better questions, then use data to test them.

Start with a hunch. Articulate the hypothesis. Run the experiment. Collect the data. Make a decision. Reflect and iterate.

That loop tightens over time. But it only works if you're honest about what you actually know versus what the dashboard is telling you.

Have you ever shipped something against the data? How did it go?

Shopify runs rigorous experiments. They track data carefully. And sometimes, when the results come back inconclusive, they ship anyway.

I appreciated that insight from Archie Abrams, VP of Product at Shopify, which he shared on Lenny Rachitsky's podcast.

They're not reckless about it. They're not throwing out the data. They're using their own conviction that the value will compound over time even if the short-term metrics don't show it yet.

That's not a failure of process. It's judgment.

Most teams treat inconclusive data as a stop sign. Kill the experiment, move on, pick something with a cleaner signal. But there's a second option, similar to what Archie is describing: ship on conviction, with your eyes open.

I want conclusive data too, but sometimes you have to plough ahead without it. Here's a framework that might help:

1️⃣ Is the experiment measuring the right thing? Sometimes inconclusive means the metric is wrong, not the idea. Check whether your One Metric That Matters (OMTM: the metric you're optimizing for in this experiment) is actually connected to your North Star Metric (NSM: the metric that defines overall business health). A flat OMTM doesn't always mean the change is worthless.

2️⃣ Is the timeframe long enough? Some changes compound slowly. A feature that improves onboarding quality might not show up in weekly retention for months. If you believe the mechanism is sound, extend before you kill.

3️⃣ What does your gut say? If experienced people on the team feel strongly that this is right despite the numbers, that's data too. Articulate why. Write it down. That's domain expertise showing up as instinct.

4️⃣ What's the cost of being wrong? If the downside is recoverable, ship and keep measuring. If it's not, wait for better signal.

5️⃣ Are you improving the OMTM in service of something bigger? Sometimes a change doesn't move your immediate metric but strengthens the system overall. That's worth pursuing even when the experiment doesn't "win" cleanly.

Blind data-driven decision making sounds rigorous. But it can quietly become an excuse to avoid hard judgment calls.

The goal was never to let the data decide. The goal was always to make better decisions. Data is one of the inputs. Conviction, experience and a clear-eyed view of the long game are the others.

When did you last ship something the data didn't fully support? How did it turn out?

Shipping With Conviction Over Data

The best product roadmaps are growth tools.

They're a bet on what will change customer behavior, move the business forward and create value that compounds over time. Features are just the mechanism.

The problem is most roadmaps don't look like that. They look like a backlog of customer requests, stakeholder asks and engineering tickets. Ordered by whoever shouted loudest last sprint.

That's not a strategy. That's a queue.

Your product roadmap should be inspired by your business model, because your business model drives actual success (it either works or doesn't).

So how do you build a roadmap that actually drives growth?

1️⃣ Hunt for the riskiest assumption. It's almost never a technical one. The biggest risk in most businesses is whether anyone will care and whether you can reach them. Start there. What would make someone feel stupid for not having adopted your product already? What feature completely reframes how your market thinks about value? That's where the roadmap should start, not with the feature backlog.

2️⃣ Think like a zero-day marketer. Uber paid black car drivers to idle before demand existed, manufacturing supply that made the product feel inevitable from day one. Burger King made its $0.01 Whopper redeemable only from a McDonald's parking lot, forcing customers to download the app, enable location services and enter payment info in a single move. Neither of those were customer requests. Both belonged on a product roadmap. Features are growth levers. Most teams just never think to pull them.

3️⃣ Don't be afraid to be weird. The comfort zone is where your competitors live. In an attention economy where everyone is producing content and shipping features, novelty is the differentiator. The best growth strategies often feel uncomfortable. That's exactly why they work. Stop worrying about being too unconventional. You're probably not being weird enough.

4️⃣ Track what actually moves the business. Connect every feature and experiment to a metric that matters. Your One Metric That Matters (OMTM) tells you if a specific change is creating value. Your North Star Metric (NSM) tells you if that value is moving the overall business. If nothing ever bubbles up to the NSM, you might be building the wrong things entirely.

Product roadmaps are not checklists. They're growth engines.

h/t to Alistair Croll for working with me on these ideas.

Who wants to share their product roadmap stories (or horror stories)? Who has figured out how to manage all the stakeholders and "voices in the room"?

taplio

The Battle

Jeremy LEPLU

Ben Yoskovitz

Followers

Posts

Engagements

Likes

Performance ScoreJeremy LEPLU's LinkedIn Performance

Performance ScoreBen Yoskovitz's LinkedIn Performance

Ready to grow your LinkedIn presence?

taplio

The Battle

Jeremy LEPLU

Ben Yoskovitz

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);}Jeremy LEPLU's LinkedIn Performance

Performance ScoreBen Yoskovitz's LinkedIn Performance

Ready to grow your LinkedIn presence?

Performance ScoreJeremy LEPLU's LinkedIn Performance