The model— fidelity over headcount

A model of your market — built on what people actually bought.

It isn't about how many personas you can generate. A billion plausible shoppers is a party trick.

The only question that pays your team back is whether the model reproduces your real sales — and ours is built, and tested, to do exactly that.

Foundation
Real transactions

Grounded in what shoppers actually bought — not what a survey said they might do.

Method
Calibrated

Fit to your sales until the model's numbers match reality — not just look plausible.

Held-out accuracy · R²
0.64

How well it predicts real market shares on sales it never saw. A blind test, published.

01How the model is built
Two layers. One turns believable into accurate.

Most synthetic-shopper tools stop at the first layer and hope it's right. The second layer is where a plausible simulation becomes a decision-grade one.

01 · AGENT CHOICE LAYER
Thousands of shoppers who actually choose.

Simulated buyers weigh brand, price and features against the alternatives on the shelf and pick — the way real people trade off a name they trust against a dollar saved. This is where behaviour becomes believable: substitution, loyalty, the pull of a promo. But believable is only the starting point.

Output: a market that looks right.
02 · CALIBRATION LAYER
Fit to your actual sales, until the numbers match.

We estimate a structural demand model (BLP-style) against your real transactions, tuning the model until its predicted shares and price responses line up with what the market truly does. This is the moat: not a prettier simulation, but one whose outputs are anchored to measured reality — and checked on data it never saw.

Output: a market that is right — believable → accurate.
02Data sources
What the model is grounded in.

Purchases and market structure — the record of what happened, and the shelf it happened on. No scraped reviews, no borrowed opinions.

SRC 01
Your POS & sales history

Products, prices and units sold in your own category. Even a few weeks anchors the model to your real market.

SRC 02
Public transaction datasets

Scanner-style purchase panels — like Nevo's cereal and dunnhumby's frozen pizza — to extend and stress-test the model beyond your own history.

SRC 03
Category & product structure

Attributes, prices and pack sizes that define how products actually compete — the axes shoppers trade off.

SRC 04
Competitive pricing & distribution

What rivals charge and where they show up — so the model reflects the shelf your buyers really face.

03The calibration proof
Watch believable snap onto accurate.

Raw agent simulation lands far from the truth. The calibration layer pulls it onto real market shares — measured on sales the model never saw.

◇ Simulation vs real sales (held-out)

Bar length = how close each lands to the true market share. Watch calibration snap onto reality.

Raw sim
off by ~26×
Calibrated
≈ real
Real sales
truth
Raw agent simulation looks believable but lands far from the truth. Adding our calibration layer moves share-recovery from −0.35 to +0.44 — the gap between “looks right” and “is right.”
Held-out accuracy · R² (0–1)
0.64
How well we predict real market shares on sales the model never saw. 1 = perfect; strong for a blind test — and something no “synthetic shopper” tool publishes.
4 of 5×
calls which way real sales move when a price changes
2
independent public supermarket datasets validated on
−0.35→+0.44
share error recovered by the calibration layer
Why not just more personas?

A million plausible shoppers still can't tell you your elasticity.

Persona count — the party trick

You can generate as many synthetic shoppers as you like and they'll all look convincing. But believable is not the same as accurate. Scale a plausible guess and you get a very large plausible guess — with no way to know how far it sits from your real sales.

Fidelity — what actually pays

Your real price elasticity, your real substitution, your real promo lift — those come only from fitting the model to transactions and checking it against data it never saw. That held-out proof is exactly the step synthetic-shopper tools skip. It's the difference between a demo and a decision.

04Honest scope
What this model is — and what it isn't.

Decision-grade means we tell you where it's strong, where it's uncertain, and where it stops holding.

Read this before you trust a number
05Get started

See it calibrated to your market.

We'll run the live simulator on your own category, calibrated to your real sales — and send you the full reproducible benchmark report. Believable isn't enough; we'll prove it's accurate.

Request a demo →