What this is
We build a population of simulated people out of behaviour that real people actually produced, put that population in front of a decision you are considering, and let them react to it and to each other. You get back the range of things that could happen, how likely each one is, and how confident we are. Not one person's opinion, and not an average of opinions. A population, reacting.
BUILT FROM WHAT PEOPLE DID. RUN AGAINST WHAT YOU ARE ABOUT TO DO.
A population built from what people actually did, carrying behavioural histories rather than demographic labels.
Individuals reporting what they think they would do, under observation, with an incentive to look reasonable.
They meet the scenario and react to it and to each other, so contagion, pile-ons and silent majorities emerge during the run.
One at a time, in isolation, and the group effect is averaged in afterwards if it is modelled at all.
A distribution of outcomes with a stated confidence level, a segment breakdown, and an account of what moved the result.
An average with a margin of error, and no way to see which parts of the population diverge or why.
Hours. The population persists, so a follow-up question re-runs against a model that already exists.
A new fieldwork cycle. Recruit, field, clean, report.
ALSO NOT: A FOCUS GROUP WITH EXTRA STEPS. NOT A POINT FORECAST. NOT A REPLACEMENT FOR JUDGEMENT.
Each of these is a decision with a cost attached, where the alternative is finding out after you have already committed.
The alternative
You are not choosing between this and nothing. You are choosing between this, a survey, a research agency, and finding out live. Here is where each one wins.
Use a simulation when
Use a survey when
WE WILL TELL YOU WHICH ONE YOU NEED ON THE SCOPING CALL, INCLUDING WHEN IT IS NOT US.
The Process
Scroll through the three stages. It starts with a record of what thousands of real people actually did inside live communities. It ends with a forecast of how those same people would react to something you are considering doing. The simulation is everything in between. We call the record the shell and the forecast the ghost.
Scroll to run
SHELL = WHAT WE OBSERVED. GHOST = WHAT WE PREDICT. AGENT = ONE SIMULATED PERSON, BUILT FROM ONE REAL BEHAVIOURAL HISTORY.
Stage 01
Every point is one thing a real person did inside a live community: a message, a purchase, a decision to stop showing up. Nobody was surveyed and nobody knew they were being counted, which is what makes it honest. It is also entirely in the past, and the past is not what you need.
Stage 02
Each history becomes an agent, and the agents are put in a room together with your scenario. They talk, copy each other, push back and change their minds. The clusters forming here were not designed by us. They are what happens when these specific people react to each other.
Stage 03
Run it a thousand times and you get a spread instead of a number. The bright contours are where the population lands most often. The width of the spread is how certain we are, and a wide one is more useful to you than a precise number we cannot stand behind.
Surveys, interviews and panels all record a person describing a decision after the fact, under observation, with an incentive to look reasonable. That gap between what people report and what they do is not noise you can correct for, and it is the reason the model has to start somewhere else.
Ours starts inside live communities, where our own products instrument behaviour as it happens. Every message, purchase, session and quiet departure is counted by something the community was already using, so nobody was recruited and nobody was performing for a researcher. That feed is layered against your own first-party data where it exists, and against public context, and no layer is ever used alone. Where two layers agree the signal gets stronger, and where they disagree confidence drops and the report says why. The model trains only on derived insights from communities that ELO’s products live in.
0
Million data points
0
Thousand signals a day
0
People observed
0
Live communities
CAPTURED INSIDE, NOT SCRAPED FROM OUTSIDE. GENERATED IN EUROPE, HELD IN EUROPE.
How the three layers are weighted, why a population is not a spreadsheet of independent columns, and how we separate evaluation from validation.
Altiera maintains the behavioural substrate, constructs populations, runs the interactions and calibrates its own confidence. Agents carry behavioural histories rather than demographic tags, so two people with identical demographics who behaved differently are modelled as different people. They influence each other during a run, which means contagion, pile-ons, silent majorities and tipping points appear because they were simulated rather than because they were assumed.
Using it takes six steps. You give it the documents you already have, ask one decision that can be wrong, check the parameters it proposes from what it read, and let it run. You get a range back with a confidence level, you push back on it, and the population re-deliberates and returns something sharper. The population persists between questions, so the second answer costs hours rather than a new project.
| Outcome distribution | The range across repeated runs, with the shape of the tail. |
|---|---|
| Segment breakdown | Which parts of the population diverge from the median, which is usually where the decision changes. |
| Reasoning trace | An auditable account of how it got there, for internal review or a regulator. |
The six-step walkthrough, all five capabilities, and the full deliverable set.
Accuracy and limits
Any vendor claiming their simulation is reliable everywhere is selling you a confident straight line drawn through a discontinuity. Relative comparisons are the strong case: which of three options lands better is far more reliable than the absolute number attached to any one of them. Group dynamics are the other strong case, because contagion and silent disengagement are what the model actually simulates.
Precise absolute forecasts are the weak case, and we give you a range because the range is the honest answer. Populations far outside anything we have observed are weaker still, and the confidence level reflects it rather than hiding it. This does not predict what any one person will do next, and it is not meant to. It estimates how a population's responses shift under stated conditions, which is the thing your decision depends on.
The difference between the two, what each one answers, and why a continuously refreshed substrate matters more than an annual one.
Questions
No. The population is simulated, the substrate underneath it is not. Agents are grounded in behaviour that real people produced, continuously, without being asked to describe it.
A panel asks simulated respondents what they would do, one at a time, and averages the answers. We place a population in a scenario and let it interact. The difference shows up precisely where it matters: contagion, tipping points and collective reactions, none of which survive averaging.
Not to start. The substrate is ours. If you have first-party data on the population you care about, it makes the result meaningfully sharper, and we will tell you how much.
Nowhere it does not need to. Data is generated and held in Europe. Client data is processed under a signed agreement and is never used to train models for anyone else.
You do not take our word for it. Every result carries a stated confidence, and we recommend running your first simulation against a decision whose outcome you already know. That is the cheapest way to calibrate your trust, and it is the test we prefer.
The substrate comes from gaming communities, which is where behaviour under pressure is best instrumented. The behaviours it captures, cooperation, defection, status, trust and belonging, are not gaming-specific. Where transfer to your population is uncertain, the confidence level reflects it.
