Simulix
LiveLive residents modeled248,932,104ZIP codes33,144Block groups220,000Attributes per agent120+

Pick a ZIP. Ask any question.
See what they would do.

Simulix builds a population of American AI agents from public Census data — for any place in the country — and answers questions about them in minutes.

Running·simulix.com/run
01 / 04
Geography
Atlanta, GA
Midtown · Atlantic Station · 38 block groups · 24,512 residents
Population · sample of 12standing by
Marcus W.
34
Aisha C.
29
David C.
41
Keisha R.
37
Jordan P.
28
Tasha B.
52
Diego S.
31
Patricia H.
45
Tyler M.
26
Renee J.
49
Brandon C.
33
Lauren D.
38
The question
Verdict · 10,000 agents · 4.1sawaiting
cancel
downgrade
keep paying
Midtown will trade Netflix for Tubi before they trade rent for it. Five dollars is the line.
Switch place
I.The old way · the new way

The same question, asked two ways.
One of them is going to feel strange in a year.

Brief · Panel No. 1142April 1, 2024
“Would renters in Brooklyn tap a $39 meal-kit ad promising ‘dinner in 12 minutes’?”
MethodIn-person focus group, two cities
Sample24 recruited respondents
Timeline14 days · 9 working days after recruit
Cost$48,000 · plus $6,000 facility
MethodologyBehind NDA, deck-only summary
DeliverablePDF, 38 pages, 3 weeks later
The way it has been done for forty years.
Simulix · run.simulix.comMay 13, 2026 · 14:32 pt
“Would renters in 11211 tap a $39 meal-kit ad promising ‘dinner in 12 minutes’?”
MethodCalibrated agent population, four DTC test markets
Sample142,000 households across four markets
Timeline5.8 seconds
Cost$0.18 · per agent answered
MethodologyOpen at simulix.com/method
DeliverableJSON · segments · verbatims · today
The way it is being done now.

We did not build this to disrupt anyone. We built it because the firms we work with told us they could no longer afford to spend $80,000 to learn which creative works. They wanted to know on Tuesday what the panels would have told them in three weeks.

— The editors
II.How it works · a single case, walked through

One question.
End to end.

A direct-to-consumer brand came to us before their next paid-social flight. Six creative variants, four markets, one $80,000 test budget. They wanted to know which combinations would land — before lighting the budget on fire.

01
The setup
Geography
30309 · 11211 · 78704 · 60614
Atlanta · Brooklyn · Austin · Chicago
Four DTC test markets · 142,000 households · Methodology open
Asked of every household

You see this ad on Instagram for a $39 monthly meal kit promising ‘dinner in 12 minutes.’ Do you tap it?

Resolution chosen to fit the spend. Multi-market, multi-variant — the way paid social actually runs.
02
The run
Sample of 224 · from 142,000
5.8 s · ± 1.4 pp
How they answered
Tap the ad
18%
Save for later
24%
Scroll past
58%
Every household carries 120+ demographic, household, and behavioral attributes. Returned in 5.8 s.
One of the 224

Jasmine R.

29 · Brand designer · Brooklyn, NY · 11211

20 of Jasmine’s 120+ attributes.

Household
  • Renter · 2-bedroom · lives with partner
  • Household income · $94,000
  • No children · no pets
  • Moved to ZIP in 2022
Work
  • Designer at a 40-person agency
  • Hybrid · in office Tues / Wed / Thurs
  • Commutes by subway · 22 minutes
Consumer
  • Subscribes to: Spotify, NYT, ClassPass, HelloFresh
  • Cooks 3 nights a week · orders in 2
  • Last meal-kit signup: 2023 · churned in 6 weeks
Media
  • Instagram · 47 min/day
  • TikTok · 22 min/day
  • Saw 3 meal-kit ads this week before this one
Values
  • Time over money · convenience tax tolerable
  • Skeptical of subscriptions she has churned before
  • Will save attractive offers to ‘decide on payday’
Her answer
Twelve minutes is the whole pitch. I’ll save it and check on payday. I’ve burned out on three meal kits already — I’m not signing up at 9 PM on a Wednesday.

Every agent in the simulation carries this kind of shape. Demographics, household, work, consumer behavior, media diet, values — all anchored to a real place. The model writes the voice; the data decides who is in the room.

03
The disagreement
Renters, ages 25–34 · Brooklyn
Tap or save. Convenience wins.
Tap
41%
Homeowners with kids · Atlanta suburbs
Skeptical of the price. Curious about the time savings.
Save
38%
Households earning $50–80K · Austin
The price is the friction. Not the concept.
Scroll
71%
Jasmine R., 29 · Brand designer · Brooklyn
Save

Twelve minutes is the whole pitch. I’ll save it and check on payday.

Marcus W., 41 · Operations manager · Austin
Scroll

$39 a month is two dinners out for our family. That’s not a meal kit, that’s a subscription I’ll forget about in three weeks.

Where the segments split is where the spend lives. Same creative, four markets, four different outcomes.
04
The decision
What they did with it

Killed two creatives before the flight. Reallocated $52,000 from Austin to Brooklyn and the Atlanta suburbs. Reworked the $39 price anchor in the Austin creative to “$1.30 per meal.” The campaign returned 2.3× the previous quarter’s CAC — measured against a control market they intentionally left untouched.

III.The public ledger

Every call we have made.
Including the misses.

Each row below is a real prediction we filed publicly before the outcome was known. Updated the morning the world catches up. Methodology open at simulix.com/method.

Calls filed
8
trailing 12 months
Within ± 2 points
6 / 8
tolerance band
Outside
2 / 8
posted publicly
Median Δ
2.1 pp
across all filings
Jan 23In band · +0.6 pp
Iowa primary turnout, statewide
Predicted 14.8%Actual 14.2%
Feb 04In band · −2 pp
Super Bowl ad recall, Madison demo
Predicted 38%Actual 40%
Mar 11In band · −3 pp
GLP-1 awareness, ages 55+, suburban
Predicted 71%Actual 74%
Mar 28Posted miss · +2.1 pp
Q1 Black-Friday spend, projected y/y
Predicted +3.2%Actual +1.1%
Apr 02In band · +2 pp
NYC mayoral favorability, age 18–29
Predicted 48%Actual 46%
Apr 18In band · +3 pp
EV intent, two-car suburban households
Predicted 22%Actual 19%
May 01In band · +1 pp
Coffee price elasticity, 2.99 → 3.49
Predicted −14%Actual −15%
May 06Posted miss · +5 pp
Local school-bond support, three Ohio districts
Predicted 63%Actual 58%

We do not retroactively delete the bad ones. When a model misses, we publish the miss and post the recalibration that followed in the next morning’s ledger.

See the full archive →
IV.The layer, not the dashboard

Three primitives.
Whatever you build next.

We are not the product on top. We are the population underneath. Research tools, audience platforms, agency dashboards, AI labs, political shops — they build on us. You can be running in ten minutes.

What you build
  • Research and insights tools
  • Audience-intelligence platforms
  • Agency-facing dashboards
  • AI labs using a real ground truth
  • In-house testing flows
  • Political and advocacy testing
asksanswers
~ 5.8 s
The Simulix layer
  • POST/v1/simulations· populations
    Build a demographically-anchored population for any geography — 240 million residents to draw from, 120+ attributes per agent.
  • POST/v1/simulations· ask
    Pose a question to that population. Get verdicts and transcripts.
  • POST/v1/simulations· agents
    Pull any individual respondent, with their full demographic shape.

One key. Three primitives. A free tier with enough room to ship a prototype, and pricing you read on the page.

V.Three honest questions

The ones we get,
in the same order, every time.

No. Every agent is built from public American demographic records for a specific place — household composition, income, work, consumer behavior, media diet, values. 120+ attributes per agent. The language model writes the voice; the demographics decide who is in the room. We hold out a fresh slice of real-world surveys and grade the population against it daily. The benchmark is on this page.
VI.Your turn

Pick a ZIP. Paste a question.
See who taps.

Free tier · no card · ten minutes to your first runsimulix.com/run