.png)
Almost every business owner right now is asking some version of the same question, are we ready for AI? And honestly it feels like the right place to start so nobody questions it, they just go looking for the answer and start spending money based on whatever they find.
But that question sends you in the wrong direction from day one and the investments that follow it reflect that.
Gregory Van Duyse of Leap AI is direct about this. The real question isn't are we ready, it's are we building toward where AI is going? Because here's the reality most businesses aren't factoring in:
AI's capacity to do real work is doubling every four months right now and that means by the end of 2026 we're looking at an 8 to 10x improvement in what AI can actually do compared to where we started the year. If you're only asking whether you're ready for AI today, you're completely missing the trajectory.
What businesses actually need to be asking is this, "Do we have the foundation in place to take advantage of what AI will be able to do tomorrow?" And that question leads you somewhere very specific and very different from where most companies are currently focused.
It leads you to data. And it leads you to something called Recursive Self-improvement.
What RSI Actually Means Outside a Lab
The Feedback Loop Explained
RSI sounds like something happening inside a frontier AI lab and Gregory is straightforward about that, it is. But the concept it's built on is anything but new and most business owners have actually heard of it under a different name.
Think about Toyota. In the 1950s they were building cars out of a war-torn Japanese economy with almost nothing going for them competitively and they went on to produce what most people consider the best cars in the world.
The reason wasn't technology, it was continuous improvement. Running experiments inside the business, measuring what worked and what didn't, going again and again until improvement became the operating system of the entire company.
RSI is that same loop made automatic and exponential. At the AI lab level what they're working toward is getting AI to improve AI, AI researchers designing experiments, AI coders writing and testing changes, benchmarks measuring whether it worked, and the whole thing feeding back into itself getting smarter with every cycle. The reason labs can move so fast is that everything is digital so there's nothing physical to wait on and the loop runs at machine speed.
For a business the loop is slower because you're dealing with humans, physical operations, and sales cycles that might take two months to show a result. But the principle is exactly the same,
Measure → Learn → Adapt → Repeat → back to Measure
You implement something, you measure what changed, you learn from what the data tells you, you adapt based on that, and you go again.
Every loop builds on the last one and over time that compounding is what creates an AI competitive advantage. Your competitors genuinely cannot close quickly, not because you have better technology but because you have more loops behind you and better data feeding each new one.
What Breaks the Loop Before It Starts
Here's where most businesses hit a wall before the loop ever gets going and it comes back to the same three problems almost every time:
-
Fragmented data across teams with no agreed definition of what anything means. If your sales team and your operations team are measuring the same thing differently, the measurements you get out of the loop are unreliable and you can't trust them to make decisions from.
-
Missing historical data. Without a real benchmark of where you were before, you have no way of knowing if you're actually getting better. You know where you are today but the before-and-after comparison that makes the loop meaningful simply doesn't exist.
-
Decisions made verbally and never recorded. This one is the most underestimated and we'll come back to it in detail because it's where Gregory gets most specific and most urgent.
The Gold Mine Most Companies Let Evaporate
What Gets Lost Every Week
Gregory describes businesses as fundamentally doing two things, having meetings and doing work. And in those meetings something enormously valuable is happening that almost nobody is capturing.
People are reporting on results; they're sharing what they're seeing in their departments. They're debating options where nobody knows for certain what the right move is. They're making judgement calls based on years of experience and context and institutional knowledge that exists nowhere else, and then the call ends and all of it disappears.
Think about that for a second. A senior manager makes a decision in a meeting, explains exactly why, references something that didn't work three years ago and why this approach is different, and then everyone goes off to act on it and that reasoning is just gone. It's not in any document. It's not in any system. It lived in that room for forty-five minutes and then evaporated.
Now think about where AI is heading. Gregory puts it plainly, within two years AI will be 100 times more capable than it is today and it will start genuinely helping leadership teams analyse options, model probabilities, run experiments and make better decisions faster.
When that happens, wouldn't it be enormously valuable for that AI to understand how your business has made decisions in the past? What was the reasoning? What got tried and failed and why? What are the real priorities when things get difficult?
That context is sitting in your meetings right now and it's evaporating every single week.
What Capturing It Actually Looks Like
Gregory's answer here is simpler than most people expect, record everything and start now.
Every meeting, at every level of the business, should be recorded. Management meetings, departmental meetings, one-on-ones between managers, daily standups, company-wide sessions, all of it. Recorded, stored, kept confidential with proper access controls, and treated as business data the same way you'd treat financial data.
There's even hardware available now that people can simply wear at work that captures everything passively throughout the day. The point isn't to build a complicated system, it's to stop letting this data disappear.
Beyond meetings, the same logic applies to the rest of your AI data strategy. A data lake that pulls together streams from your ERP, your CRM, your spreadsheets, your email, not replacing those systems but connecting them so that AI has one place to access your business instead of having to navigate five different siloed locations.
And KPIs that are bankable, meaning tied to real outcomes like profit, revenue per employee, client retention, not vanity metrics that look impressive on a slide but don't actually tell you if the business is improving.
The goal across all of it is the same, give AI the data it needs to eventually help you run the loop properly and keep running it faster.
The Compounding Advantage
Here's what makes this urgent and not just interesting, the gap between companies building now and companies waiting is already opening and it doesn't stay the same size. It grows every quarter.
Every cycle of the loop a business runs gives them better data, better measurements, cleaner decisions, and a faster next cycle. The business waiting to feel ready is still at zero cycles.
And because AI's underlying capability keeps doubling every four months, the companies running loops now are doing it with tools that keep getting more powerful underneath them. The advantage compounds on itself.
Gregory is clear about what this means at the industry level, whoever reaches recursive self-improvement first in an industry will dominate that industry. They'll do things better, faster, and cheaper in ways that keep improving automatically while everyone else is still figuring out where to start.
And this brings us back to the most important line in the whole conversation around AI data strategy and why it matters right now:
"You can't optimise what you didn't record." — Gregory Van Duyse, Leap AI
Every week you're not capturing data, not measuring outcomes, not building the feedback infrastructure, that's a week of the loop someone in your industry is running that you're not. And unlike a technology gap which you can close by buying the same tool, a data gap compounds in the other direction the longer you leave it.
The Sequencing Mistake That Poisons the Loop
A lot of AI consultants tell companies to rethink and redesign their workflows at the same time as they're integrating AI, which might improve the process while we're in there. Gregory pushes back on this hard and the reason is measurement.
If you change your workflow and implement AI simultaneously you now have two variables changing at once and no way of knowing which one drove the result. Did efficiency improve because of the AI or because you redesigned the process? You genuinely can't tell and if you can't tell, you can't learn from it, which means the loop is broken before it starts.
The right sequence is capture first, then govern, then optimise, then innovate. Take a clean benchmark of how the current process performs with humans doing it, quality, quantity, time, cost. Then integrate AI into that same process without changing anything else.
Now you have a real before and after and you can see exactly what the AI integration did. Once that's validated you optimise and then eventually you innovate on top of a system you actually understand.
Gregory uses a customer service example that makes this very concrete. A company built an AI system to handle support tickets, it looked incredible, and they fired 80% of their customer support team because the AI was handling everything so smoothly. A few months later they started rehiring those same people.
Why? Customers had quietly stopped doing business with them, not complaining, just leaving. The trust had eroded. The AI was technically closing tickets but it wasn't solving real problems and nobody had realised that the human support agents were also the company's most important source of product improvement feedback. Both things were lost at the same time and the damage only showed up later in churn numbers.
They changed too much too fast and they had no way to measure which change caused which outcome. That's what skipping the sequence costs you and it's a pattern that shows up across industries whenever companies rush the transformation before they've nailed the integration.
What to Build Before Your Next AI Project
Before you launch your next AI project, whatever it is, there are three foundations that determine whether it feeds into a compounding loop or just sits there as a standalone tool that doesn't connect to anything.
-
Get your data connected. Not perfect, connected. Your enterprise AI roadmap doesn't require a perfect data lake built all at once. Start with the most important stream, probably your ERP or CRM, and bring that data together into one accessible location. Add streams as new projects create the need for them. Build it incrementally rather than waiting to build something perfect before you start anything.
-
Build KPIs that are actually bankable. Work with your department heads to find the measurements that tie directly to real outcomes, revenue, profit, efficiency, client lifetime value. If you don't have a dashboard that tells you clearly whether the business is going better or worse, you can't run meaningful experiments and you can't prove that anything you're building is actually moving the needle.
-
Start capturing everything right now. Every meeting, every decision, every management conversation. Not because you can use all of it today but because the AI that will be helping you run your business in the next two to three years will need it and the businesses that started capturing early will have a depth of institutional knowledge that businesses starting from scratch won't be able to manufacture quickly.
This is exactly what Gregory's Seven Strategic Pillars framework is built around, creating the conditions where AI can do what it's actually capable of, including eventually running the improvement loops that compound your AI competitive advantage every single quarter. The pillars aren't a technology checklist. They're a sequenced AI implementation strategy that builds toward RSI from the very first project.
The AI race is a data race. And it's already started.
Download the 7 Strategic Pillars framework to see exactly how to build the data foundation RSI requires.
