Observe
Start from recorded actions in live communities and your own data: messages, purchases, departures. What people did, not what they said.
The problem
Surveys tell you what people say. Decisions depend on what they do.
Price changes, crisis responses and retention offers are usually tested with surveys and panels: individuals answering one at a time, reporting what they think they would do.
Real people do not decide in isolation. They react to each other: contagion, pile-ons, and the silent majority that quietly stops spending. Polysoma models that.
Context ingested
- pricing_deck_q3.pdfPARSED
- churn_export.csvPARSED
- community_guidelines.mdPARSED
- 2025_research_readout.pptxPARSED
4 files · 1.2M tokens · held in EU
Question
If we raise the mid tier from £9 to £12 in Q1, what happens to retention?
✓ Simulable: outcome defined, date attached
Proposed parameters
| Population | 30,000 agents | default |
| Segments | 4 | churn_export.csv |
| Runs | 1,000 | default |
| Horizon | 90 days | pricing_deck_q3.pdf |
| Confidence floor | 0.70 | default |
Every value is visible, sourced, and yours to override.
Result · retention at £12, 90 days
−4.1ppMedian
Spread −9.2pp to −0.6pp
Spread −9.2pp to −0.6pp
Confidence: moderate-high
Segment · lapsed returners−14pp
Segment · under-25+6pp
Follow-up · re-running · 400 runs
If we take the aggressive option, how does it land with the under-25 cohort specifically?
Under-25s absorb £12 with a median of +2.4pp retention, spread +0.1pp to +4.6pp. The cohort moves with its peers rather than on price.
Confidence: high (0.86) · holds unless peer churn exceeds 8%
Round trip on a follow-up: hours. The population stays live between questions.
730KNew signals per day
1.5MPeople observed
50MData points gathered
Populations are grounded in live behaviour from online communities, where behaviour under pressure is best instrumented, then combined with your own first-party data and public context.
How it works
Four steps.
Populate
Individual behavioural histories become agents, each carrying its history into the scenario.
Simulate
Put the decision in front of the population. Agents react to it and to each other, so contagion and pile-ons emerge instead of being averaged away.
Read the result
Many runs produce a distribution of outcomes with a stated confidence level: where the population lands most often, and how sure we are.
Simulation vs. survey panel
A population, not a panel.
| Polysoma | Survey panel | |
|---|---|---|
| Who answers | A population built from what people actually did, carrying behavioural histories. | Individuals reporting what they think they would do, under observation. |
| How they answer | They meet the scenario and react to it, and to each other. | One at a time, in isolation. |
| What you get | A distribution of outcomes with a stated confidence level. | An average with a margin of error. |
Where it’s used
In practice.
Pricing & packaging ↗
Who leaves, who upgrades, and who says nothing while quietly ceasing to spend. Test the change before your customers see it.
Crisis response ↗
Simulate three responses before publishing one.
Retention interventions ↗
Which intervention keeps people, and which merely delays their exit.
Deployment & data
The details.
- Inputs
- Three layers: live community behaviour, your first-party data, and public context.
- Output
- A distribution of outcomes with a stated confidence level, built for a leadership discussion.
- Turnaround
- The pilot depends on how much grounding your population needs. After that, follow-up questions take hours: the population persists.
- Engagement
- A scoping call, then a pilot question, then a standing capability where each new question costs less. Priced per engagement against research and decision budgets.
- Your data
- NDA before anything sensitive. Client data is processed under a signed agreement and never used to train models for anyone else.