By IDEASCANNER · 2026-09-22 · Category: AI & Entrepreneurship
What if entrepreneurs could simulate major decisions before investing time, money and energy? This is the idea behind our vision of a World Model for Business: understand companies better, play out possible developments and learn from real outcomes. The article explains the opportunities — and why getting there is more technologically and managerially demanding than it first appears.
Our path to a “World Model for Business” — and why the toughest questions aren’t only technical.
Imagine you want to take your company to the next level.
You could expand sales. Specialize your offering. Develop recurring services. Or first structure the organization so growth doesn’t just add more work to your desk.
All of these routes can make sense. But which one fits your company? In what order? And under what conditions would an initially convincing decision turn into an expensive mistake?
I’m therefore occupied by one question:
Could we test business decisions in a model first, before spending money, time and the energy of a whole team?
That is exactly the vision we pursue with IDEASCANNER: a World Model for Business.
Not as a promise to predict the future, but as a way to better understand possible actions, examine their likely consequences and systematically learn from implementation.
What is a world model anyway?
Put simply, a world model represents relevant states of an environment and tries to model how they change — especially when someone takes action. In AI research this approach is used to simulate outcomes in virtual environments. The basic idea: a system should be able to play out possible developments instead of only reacting to events that have already happened.
Our question is how this concept can be meaningfully transferred to businesses.
It’s not about a photorealistic copy of the office or production hall. It’s about a model of business relationships: between offering, customers, competition, sales, organization, resources and economic results.
The aim would be to investigate questions such as:
What could happen to my company if I implement this measure — and what would need to be true for it to work?
Business simulations are not a new invention. At MIT, dynamic business models and so-called management flight simulators have been developed and used for decades. Our goal is to combine such ways of thinking with AI and company-specific knowledge and make them accessible for everyday use in medium-sized companies.
Where we are today — and where we want to go
At IDEASCANNER we currently work with a structured view of more than 150 value-relevant characteristics. Central are, among other things, a company’s unique position, its sales development and the construction of a more capable, evolving organization.
This is our starting point. It is not yet a validated world model that reliably simulates the consequences of business decisions.
Structuring an analysis of a company is different from robustly modeling its development under different decisions.
We don’t want to hide that difference behind a new term. We want to make it the subject of further development.
Our direction can be described in three words: Understand. Simulate. Learn.
Today we help identify potential and fields of action. Building on that, we want to make possible development paths comparable. In the long term, comparing with actual results should help improve the underlying models.
A concrete example: Is more sales really the next step?
Take a fictional technical service provider.
Demand is decent and the team well utilized. Many orders are custom solutions, and important decisions rest with the owner. The company wants to grow.
An additional salesperson could bring in more orders. Under the described conditions, however, they could also worsen existing bottlenecks: more coordination, longer delivery times and extra strain on key people.
Stronger specialization could open other customers and better prices. But the company might have to give up parts of its current business for that.
Standardizing certain services could free up capacity. Initially, though, it would require time that is lacking in day-to-day operations.
A useful company model would need to account for these interactions. It should help examine how different paths affect profit, liquidity, team and owner dependency.
A valuable answer might be, for example:
Given these assumptions, it makes more sense to standardize certain services first and then expand sales. Crucial is whether the expected relief can actually be achieved. That should first be tested in a limited area.
That would not be certainty, but a concrete, testable decision hypothesis.
Perhaps the greatest benefit sometimes lies precisely in recognizing what must happen first for the next step to work at all.
What opportunities this could create for entrepreneurship
For me it’s about much more than better forecasts.
We could experiment more boldly. Whoever recognizes the critical assumptions behind an idea can test more selectively. Instead of immediately approving a large investment, a small trial could show whether the decisive prerequisite is even met.
Strategy could become an ongoing learning process. We would not only ask whether a measure was implemented, but also: What effect did we expect? What actually happened? What must we reassess as a result?
Entrepreneurial knowledge could become more shareable. Such a model could help make experiences, assumptions and dependencies visible to a leadership team. This is also interesting for succession: Which considerations support the business model — and which have only existed in the owner’s head so far?
And finally, we want to make this form of decision support available to companies that cannot build their own strategy or simulation departments.
The opportunity would be better risk management. Not necessarily to take fewer risks, but to decide more consciously which risks an entrepreneurial opportunity is worth.
The hurdles are substantial — and part of the vision
As attractive as this perspective is, much work lies between a convincing scenario and a reliable decision model.
Patterns are not causes
If successful companies often have specialized offerings, that does not automatically mean specialization makes every company more successful. Maybe some companies can specialize precisely because they already have a strong market position.
A decision model must do more than detect correlations. It must help distinguish what co-occurs from what actually causes change. This distinction is central to research on causal learning.
For us this means: even a plausible-seeming causal chain must first be treated as a hypothesis.
We need implementation experiences — including failures
To learn from decisions, we need more than a snapshot of a company.
We must be able to trace how the starting point looked, what was decided and actually implemented, and which results followed. Even when a measure failed.
At the same time, this must not become a costly documentation project. A system whose data maintenance costs more energy than the decisions it improves would defeat its purpose.
This also includes: companies must be able to understand which of their confidential data is used for what. Learning across companies must not mean sharing trade secrets.
Companies operate in a world that reacts
Customers change expectations. Competitors respond. A decision can alter the very conditions on which its original assessment was based.
Research also describes this fundamental problem as performative prediction: when people act on a forecast, they can change the predicted outcome.
That’s why we’re not only interested in a supposedly optimal future path. We want to see which decisions hold up across several plausible developments — and when a model needs re-evaluation.
Precision must not be mistaken for reliability
A number with two decimal places can look convincing without being well-founded. A range is not robust simply because a simulation outputs it.
Our standard must therefore be to make assumptions, data gaps and uncertainties visible. That includes the ability to say: for this question, we still know too little.
We want to test models against later results and simpler planning approaches. What matters is not how impressive a simulation looks, but whether it demonstrably leads to better decisions.
The goal remains a human decision
What does “better” even mean?
More profit? Higher company value? Less personal dependency? Secure jobs? More time for family? Long-term independence of the company?
A model needs goals. It should not quietly fix them for us.
For me the vision therefore includes making goal conflicts visible rather than hiding them behind a seemingly objective recommendation. Responsibility for the decision stays with people.
Our measure is not perfect prediction
We want to develop this approach on concrete business questions. Step by step, verifiably and with willingness to correct assumptions.
A helpful World Model for Business would not have to predict every development. It would need to help us ask better questions, recognize critical prerequisites earlier and learn more consistently from experience.
My hope is that this creates more room for what entrepreneurship means to me: seeing opportunities, daring something new and creating something valuable for others.
We don’t want to take decisions away from entrepreneurs. We want to help them better understand what they are deciding about.
Which concrete decision in your company would you like to test in a model before you make it in reality?