Tobias Zwingmann

RAPYD.AI

Independent AI consulting to create profitable AI solutions to innovate and grow your business. Fast.

Managing Partner

O'Reilly Media

Author and Instructor

LinkedIn

LinkedIn Learning Instructor

D2A2

Chief of Research - Applied AI

Springboard

- Helping Springboard as a subject matter expert in AI, especially for data analytics
- Mentoring students who want to switch careers to data science / data analytics 
- Top Mentor Award in 2021 (Springboard Inspire)

AI SME & Data Science Mentor

Insurgence

Partner

Toptal

Independent Consultant

Data Science Dojo

ML/AI Instructor

Deutsche Messe AG

Systematic retrieval, analysis and utilization of (unstructured) data from diverse sources to build new products and processes.

- People movement analytics 
- Computer vision with focus on text retrieval

Senior Data Scientist

Building & maintaining predictive models. Generating and communicating insights about customer trends and behavior to product owners and senior management.

- Churn Prediction
- Next Best Offer Recommendations 
- Customer Segmentation
- Visual Analytics

Data Scientist

Leading various international trade show projects in e.g. China, India, Indonesia, Mexico, Myanmar, South Africa, South Korea, Turkey.

Project Director

Deutsche Messe Dubai Branch

Management Trainee

Author of AI-Powered BI (O'REILLY) | AI Coach & Mentor | I help organizations leverage AI and use data more intelligently to improve business performance. Sharing AI use cases & learnings.

I didn’t start my AI/ML journey doing “Profitable AI”.

I started it as a Data Scientist who built models that looked impressive on paper:

– Customer churn prediction with 95% accuracy ✅
– Next-best-offer systems with 15% uplift ✅
– Data-driven B2B products generating 6-figure revenues from day one ✅

But the hard lesson I’ve learned after 10+ years at the intersection of business and technology is that none of this matters if you can’t show results on the bottom line.

I’ve spent the last 5+ years in applied AI consulting where my work has taken me inside global enterprises in automotive and insurance, mid-sized companies with 500–1,000 employees, solopreneurs aiming to double their output, and small business owners uncovering $10k+ monthly savings in unexpected places. 

The contexts differ, but the patterns of what makes AI succeed (and fail) remain remarkably consistent.

Success is not about:
- Chasing the next shiny AI tool
- Hiring data scientists from Silicon Valley
- Hoping IT will figure it out

It’s about applying the principles of Profitable AI to deliver outcomes that matter: doing things faster, at lower cost, with higher quality — or scaling what you do to entirely new levels.

Along the way, I’ve shared many of these lessons in my books AI-Powered Business Intelligence (O’Reilly, 2022) and Augmented Analytics (O’Reilly, 2024), as well as in my weekly newsletter that is now read by 10,000+ professionals at companies like Google, Amazon, Gucci, H&M, Santander, and Mercedes-Benz.

So If you’re tired of the shiny AI demos and want to see it deliver real results instead — I think you found a good place.

We help you build amazing AI products faster.

Gemini 3 is making a lot of noise. And it's a really great model. "Is it better than GPT-5?!" My take: Use whatever works for you. Both are LLMs, both have their quirks. But if you want to know where each shines, here's what I found:

Gemini 3 = Messy Inputs → Clean Outputs
Throw it chaos: screenshots, transcripts, logs, mixed documents. It's great at compressing that mess into structure. Best when you're very specific about the output format you want. Seems like if you need to organize messy information, call Google.

Examples:
- Summarizing multiple meeting recordings into structured notes
- Extracting data from a messy pile of PDFs, screenshots, and spreadsheets

GPT-5.1 = Clean Inputs → Complex Thinking
Give it organized information and let it reason. It's great at planning, narrative, and multi-step decisions. Best when the task itself is complex, not the input.

Examples:
- Writing code (though Claude still beats it here)
- Building a strategy doc from a clean brief

Scenarios to combine both:

1/ Competitive analysis
Gemini 3 gathers data from competitor sites, pricing, and reviews; GPT 5.1 delivers the strategy

2/ Client onboarding
Gemini 3 cleans and organizes intake docs; GPT 5.1 produces the project plan

3/ Sales pipeline review
Gemini 3 compiles CRM data, calls, and emails; GPT 5.1 spots trends and recommends actions.

Don't try to find "the" best. 
Find whatever works best for you.

Fun fact: The infographic was done by me with Gemini-3 image gen ("Nano Banana"), first shot, Prompt in comments.

AI Advisor | O’Reilly Author | Instructor @ LinkedIn Learning | I find & realize profitable AI opportunities – here to share wins, fails, examples & frameworks.

Claude Code might be the most underrated AI tool right now. Not because it’s revolutionary, but because almost no one outside the AI bubble knows it exists. The name suggests it’s just for developers. It’s not.

I recently used Claude Code to set up my Nvidia Spark in 1 hour instead of 1 day. It handled all of the Linux work by checking configurations, reading logs, and generating the commands I needed to install tools like Docker and n8n. If you're on a Mac, you can run Claude Code locally too.

It felt like using ChatGPT except it has access to your computer.

You don’t need to be a developer to use it, but you need to be ok working with a terminal. If you are, it unlocks workflows that regular web-based AI tools struggle with:

- Working across multiple files and large projects
- Creating new files and docs across folders
- Getting insights from scattered datasets
- Pulling answers from many documents
- Organizing / sorting files in folders

I think it’s more of a local AI workspace than a coding tool.

If you’ve ever wished you could "upload your hard drive to ChatGPT", Claude Code comes close.

Give it a try if you don’t fear that blinking cursor.

Neil Sambol

I see a lot of confusion around AI workflows, AI agents, workflow automation, and agentic workflows. Most people can't even tell the difference. Which isn't a problem until you want to build an AI solution that is neither over-engineered nor insufficient.

So I put together a quick guide that breaks down each type and when to use what, along with a few examples.

Read the full blog post here: https://lnkd.in/erYqeVPi

AI Workflows vs. AI Agents

ChatGPT launched today 3 years ago. ONLY three years I should say. Feels like a decade compressed. To catch you up on what’s happened, I generated this little timeline with the help of ChatGPT Deep Research, had Gemini 3 double-check it, and Claude 4.5 generated the PDF.

Tasks like this would have taken a whole day in the "old" world.
Now done in less than 60 minutes with AI.

We're still far away from AGI.
But I don't care.
We're living in the most exciting timeline.
And I can’t wait to see what this timeline looks like 3 years from now.

3 Years Since ChatGPT

A while ago, I shared the AI Profit Stack – a simple framework to assess whether your AI projects are actually creating profit. People loved it, clients used it. Then someone asked me:

“How do I apply this to my Copilot licenses?”

Short answer:
You don’t.

Long answer:
Trying to measure all AI with the same ROI formula is exactly how you end up with fictional numbers, cynical finance teams, and AI steering committees that debate themselves into extinction.

Today I've shared the fix I'm using now in every client engagement – and the one I want you to have too.

Find it here here:

Open article: The Two Tracks of AI ROI by blog.tobiaszwingmann.com

How to separate 'AI that helps' from 'AI that pays'

taplio

Tobias Zwingmann's Linkedin Analytics