Modelling stakeholder behaviour

Organisations have spent a decade instrumenting everything except the humans who decide their outcomes. That is starting to change.

Most strategic failures are people failures. The data was available and the signals were public. What the organisation misjudged was a person: how a minister would react under pressure, how a regulator would read an ambiguous filing, how a counterparty would behave once the deal got hard, how a journalist would frame a difficult story.

Look at where the modelling budget goes, though. Companies model their supply chains, their cash flows, their churn and their networks in extraordinary detail. The people who actually determine outcomes, the ministers, regulators, investors, rivals and journalists, remain unmodelled. They are covered by monitoring, which tells you what they said yesterday, and by instinct, which is only as good as the last conversation your adviser had.

A stakeholder digital twin closes that gap.

What a twin is

In our usage, a digital twin is an evidence-bound model of how a named individual or institution reasons, communicates and decides. It is built from the public record and structured evidence: voting histories, speeches, interviews, written output, institutional incentives, relationships and revealed behaviour under pressure. It carries a fidelity grade, so you know how much evidence sits beneath it before you weigh its output, and it updates as new evidence arrives.

It is probabilistic by design. A twin estimates likelihoods and shows its reasoning. It does not issue prophecies, and any model that claims certainty about a human being should not be trusted with anything.

What it changes in practice

Anticipation improves first. Before a select committee appearance, you can put your brief to a twin of the chair and see which lines of attack are most probable and which answers hold. Before a regulatory submission, you can test how that regulator has treated similar arguments and where your framing is likely to land.

Message testing accelerates. Rather than waiting weeks for fieldwork, a message can be run against synthetic audiences, modelled populations segmented the way the electorate or a customer base actually splits. The output is directional rather than definitive, and it is cheap to rerun a dozen variants before lunch. Polling still matters. It matters most once you already know which three messages deserve the fieldwork.

Negotiation preparation becomes rehearsal. A simulator built on twins lets a deal team play the other side of the table repeatedly before the real meeting, stress-testing sequencing, anchors and concession patterns against a model of the specific people opposite rather than a generic adversary.

And twins watched over time provide early warning. A minister's language shifting on an issue, an enforcement pattern tightening, a rival's positioning drifting. Pattern change usually precedes announcement, and a persistent model notices drift that episodic monitoring misses.

A different question from data platforms

The last decade's intelligence infrastructure, with Palantir the most prominent example, solved a real problem: organising what an enterprise already knows and connecting it. That is a question about your own data. Twins address the other question, the one that determines most outcomes: what will the people outside your organisation do next. The two are complements rather than substitutes, but only one of them tells you what Thursday's meeting will actually be like.

Where we are strict

Trust in this category depends on discipline, so the limits are built in. Twins speak only where evidence exists. Where the record is thin, the twin says so and the fidelity grade reflects it. Every material forecast we issue carries a probability and enters a scored register, checked against reality when reality arrives. We would rather be measurably right at 70 per cent than unfalsifiably impressive.

Organisations that model the people who matter will out-prepare those that only model their own data. Preparation compounds.