AI agents help construct and attack models, not impersonate real populations or predict individual behaviour. One agent translates workshop language into candidate causal structures; another searches for missing actors and feedback loops; others inhabit explicitly defined incentive models to expose possible strategic responses. Retrieval connects assumptions to primary research and internal evidence. A critic detects when several agents share the same hidden premise or when a simulation's narrative exceeds its mathematical support. Language generation explains differences between worlds and proposes discriminating experiments, while humans decide which structures are legitimate and which consequences matter. The system should increase epistemic humility without making action impossible.
Counterfactual
“Can leaders test important decisions before reality makes them expensive?”
The beginning
Counterfactual is a decision studio for exploring not one forecast but a structured field of plausible outcomes. Important choices are made inside systems where actors adapt, evidence is incomplete and the intervention itself changes behaviour. The product would let a team express several competing models of that system, intervene on each and observe where their consequences diverge. It does not promise to show what will happen. It reveals what would need to be true for a particular future to emerge, which assumptions carry the decision and what early evidence would distinguish one world from another. The aim is to transform uncertainty from a rhetorical disclaimer into material that can be inspected, debated and acted upon.
Strategic decisions are often defended with a single spreadsheet and a narrative written after the preference is already known. Uncertainty is compressed into a sensitivity tab, second-order effects disappear and disagreement becomes political because alternative world models are never represented explicitly. Scenario planning can produce memorable stories without causal discipline; forecasting systems can produce precise numbers that conceal structural uncertainty. When reality differs, organisations rarely preserve enough information to learn whether the decision process was weak or the world simply took a low-probability path. Expensive commitments are therefore made without identifying reversible probes, leading indicators or conditions under which the choice should be reconsidered.
What it could become
The product would be a collaborative simulation canvas where teams define actors, resources, constraints, causal relationships, uncertainties and possible interventions. Rather than producing a single polished model, the workspace keeps rival structures side by side and highlights where they imply different actions. Users can change an assumption, freeze a variable, introduce an adversarial actor or remove a dependency and watch consequences propagate. Every scenario produces a decision record containing its evidence, sensitive parameters, expected signals and exit conditions. The system then recommends small real-world probes whose results would reduce uncertainty before the organisation commits. Later outcomes return to the model, allowing the team to evaluate calibration and revise its representation of the system.
For whom
- Executive and strategy teams
- Public policy organisations
- Investors and portfolio operators
- Product leaders making platform bets
Core capabilities
- Causal scenario modelling
- Multi-agent simulation
- Sensitivity and assumption analysis
- Decision journals and leading indicators
The project becomes meaningful only when a new technical possibility is translated into a clear human advantage, an experience people can understand, and a system capable of earning trust over time.
Intelligence and mathematics
Structural causal models distinguish observation from intervention and make counterfactual questions explicit. Monte Carlo simulation propagates parameter uncertainty; global sensitivity analysis reveals which assumptions dominate results; game theory and agent-based models represent strategic adaptation; and robust decision methods search for actions that perform acceptably across incompatible plausible worlds. Bayesian model comparison updates the relative support for rival structures as evidence arrives. Real-options analysis values reversibility and staged commitment. Outputs are distributions, thresholds and phase changes rather than point predictions. The decisive mathematical object is often a boundary: the smallest change in a parameter or behaviour that causes the preferred action to reverse.
How it might live
Counterfactual could begin as enterprise decision software for a narrow class of costly, repeated choices such as market entry, product-platform investment or capacity planning. Early deployments would combine software with structured facilitation because the hardest problem is eliciting honest assumptions, not running simulations. Subscription value grows as organisations accumulate reusable causal components, calibrated priors and a decision history that can be audited without hindsight bias. A portfolio edition could help investors compare assumptions across companies without forcing them into one template. The company must be paid for improving decision quality, not for producing impressive certainty; commercial incentives should reward transparent limits, reversible experiments and learning after outcomes arrive.
For me, a venture is more than an interesting technology. It needs a narrow first user, a repeated problem, a distribution path, a credible advantage and a reason to improve as more people use it. I would test those conditions before deciding whether this idea should become a company, a product, an open technology or an ongoing research programme.
Rules for making it real
- 01
Never present a scenario as a prediction.
- 02
Show which assumptions drive the result.
- 03
Prefer reversible experiments to irreversible confidence.
- 04
Score decisions by process as well as outcome.
From question to company
- 01Frame
Prototype causal canvas and scenario comparison.
- 02Prototype
Pilot on product and market-entry decisions.
- 03Prove
Add leading-indicator monitoring.
- 04Build
Build calibration analytics across decision portfolios.
What could go wrong
Serious imagination includes the possibility that an idea should change radically—or should not exist. These are the tensions the project would need to resolve:
- Sophisticated simulation masking weak assumptions.
- Decision-makers using scenarios to legitimise a preferred choice.
- Sensitive organisational data leaking into external models.