Why Raati exists

The market is
an expensive teacher.

Marketers have learned what works the same way for fifty years — by paying for the answer. Raati is built on the bet that they no longer have to.

A methodology validated at Harvard Business School, Stanford, Ipsos, and The New York Times — made operational for teams that ship every week.
02 · The bet

The cost of learning is paid in distribution.

For most teams, distribution is the feedback loop. You ship copy, watch the numbers, adjust. The next campaign is informed by the last campaign's invoice.

This is not a bad system. It is just the most expensive one available.

A performance team spends $40K to learn which of three hooks lands. A B2B marketer launches a 90-day campaign and finds out — in the QBR — that the value prop didn't translate. In both cases, the lesson arrives after the budget has been spent.

We started Raati because the alternative has existed for years. It just hasn't been operational for most teams.

03 · The methodology has a paper trail

Synthetic audience research is not a new idea.

It is a quietly mature field — backed by work from Harvard Business School, Stanford-affiliated labs, Ipsos, and The Times. Four findings, each load-bearing.

[1] · Harvard Business School · 2025 calibrated
Larger, Faster, Cheaper: The Future of Market Research with AI
LLM-based synthetic customers approximate real-respondent preference data on product attributes — when calibrated against real-world responses to correct for known biases. (LLMs over-rate novelty. The researchers correct for it. So do we.)
Calibration matters
[2] · Stanford / DeepMind · 2024 85%
Digital twins from two-hour interviews
Over 1,000 digital twins built from two-hour interviews with real people. The agents answered surveys with roughly the consistency of humans retaking the same survey two weeks later.
Test-retest accuracy
[3] · The Times / Electric Twin · 2025 92%
Synthetic panel from a 642,000-subscriber base
A synthetic panel built from the Times' subscriber database, validated against a 10% holdout human group. Traditional human surveys, for reference, sit at ~93%.
Holdout validation
[4] · Ipsos / Stanford GSB · 2025 in trial
Synthetic respondents in market research
A formal partnership validating synthetic respondent panels for market and public-opinion research — including the risks and mitigation strategies that come with the methodology.
Industry partnership
0%
The headline number, in context

The Times' synthetic panel matched a holdout human group at ~92%. Traditional human surveys, run the same way, hit ~93%. The gap is one percentage point.

The methodology also beats demographic-only modeling — the realistic alternative for most marketing teams — by 14 to 15 percentage points, cuts time-to-insight by roughly half, and reduces cost by about a third.

HBS d3 · 2025 Stanford / DeepMind The Times R&D · 2025 Bain insight
04 · What we built

The research shows the methodology works. We made it operational.

The studies above run inside research teams — HBS labs, Times analytics, Ipsos panels. They take weeks to set up and require hand-curated audience data, careful calibration, and statistical care. They are not something a marketing team calls up on a Tuesday afternoon.

Raati is the operational layer.

In the lab
Weeks of setup. Manual calibration. A statistician on call.
  • — Source interview transcripts
  • — Hand-curate audience dossiers
  • — Calibrate against a holdout
  • — Run panel, interpret, write up
cycle time · 2–6 weeks
In Raati
Paste a draft. Read structured verdicts in minutes.
  • — 1,400-word dossiers, pre-built and tuned
  • — Bias calibration in the panel composition
  • — Verdicts return as structured JSON
  • — Synthesised variations target weak points
cycle time · ≈ 4 minutes

The methodology is the same. The friction is gone.

05 · What this is, and what it isn't

Synthetic audiences are not a replacement for live research.

The papers say this. The platforms doing this work say this. We say it too.

What Raati replaces is the first round — the part where you find out, expensively, that the hook doesn't land or the value prop doesn't translate. By the time real humans see your work, the obvious mistakes have been removed.

If you have a research team, Raati makes them faster. If you don't, Raati catches the first 80% of what a researcher would have flagged.

06 · Same lab, different surfaces

Two kinds of teams. One lab.

Raati serves performance marketers and B2B marketing leaders. The panel infrastructure is identical. The shape of the question changes.

For performance marketing
Three hooks, twelve hours.

Surface the one with a chance before it touches a media plan. Read the disagreement between personas — it tells you which segment is at risk before you spend on it.

Loop
Draft → panel → ship
Replaces
Burn-to-learn ad spend
For B2B marketing
A TAM of 12,000 humans.

You can't A/B test your way to truth at this scale. Build a panel from your actual buyer profiles — finance, IT, operating — and find out which proofs land before the campaign ships.

Loop
Draft → panel → launch
Replaces
Post-launch QBR surprises
Same panel infrastructure · Same verdict format · Different question
07 · One more thing

We made Raati because we wanted to use it.

We are not a research company. We are operators who needed a lab and built one.

If you want to see a panel run on something you have already shipped — or something you haven't — we will set up a session. Bring the draft. No slides.