A housing policy, a transportation investment, or a new development shapes an area for decades. The planners who do that work deserve evidence they can trust, and they should not have to spend weeks putting it together. That’s why we built EIKOS.
This year we spoke with planners across the US about how they build the evidence for their plans, studies, and grant applications. Many told us the information exists, but getting it together takes time.
Those conversations shaped what went into EIKOS.
What EIKOS shows
EIKOS shows planners how residents are likely to spend an ordinary day in the area they plan for. Each dot on the map is a simulated resident with a home and a daily schedule, and a job and a commute where they have one. Press play and the day runs hour by hour.
Click a resident to see who they are: their age, education, job, income, household size, and what they are doing at that moment. Every resident is synthetic, modeled statistically from census and national survey data. The simulation contains no personal data.
Why simulated residents help
Public data usually comes as separate tables. One gives you age, another income, another how people travel to work. Each is useful, but none of them tells you how those facts combine for the same household. Each resident in EIKOS carries all of them at once, in the same combinations they appear in the real population.
The Report
Next to the simulation, EIKOS has a planning Report with two tabs: Comprehensive planning and Urban design and planning. Its sections cover topics a plan has to address, such as population, housing, and employment. Every section names its sources, and the Report cites the state planning law that applies.
EIKOS does the analysis. The judgment stays with you.
Planning is a commitment to the people who will live with its decisions. The purpose of EIKOS is to help you understand the lives your decisions will shape, and bring that understanding into planning for the future.
Early access is now live - https://t.co/0t1IvX0dVB
Day 14 of #30DayMapChallenge: Borgesian map.
Jorge Luis Borges wrote a one-paragraph story, “On Exactitude in Science”, about cartographers who make a map of an empire as large as the empire itself. EIKOS comes close in its own way: it draws every resident.
Each person is one of the 660,511 residents EIKOS models for the city of Boston. None of them are real. Resident #250,064,758 is 45, finished high school, works in installation and repair, earns $89,700 and takes a 20-minute taxi to work at 6:02am.
Day 13 of #30DayMapChallenge: Interactions.
It is 12:15 on a Tuesday in downtown Providence. The lunch rush is on. The dots are people and vehicles, and each person has an age, a household, a job and a day of trips.
Ask a language model to simulate 100 women and 100 men on how entrepreneurial they are, and it pushes the two groups more than three times further apart than they actually are. It is not describing a person. It is describing the label.
@Forbes cited that research last week. → https://t.co/AE4QDq74yN
GPT-3.5 and 4.1 gave more varied answers than GPT-5.4 and 5.5.
The better models get at coding, the more they behave as if every question has one right answer.
That is a problem if you want a model to represent a population rather than find an answer.
Our co-founder, Michael Birdsall covered this in Perspectives on AI, a webinar with @APDUorg.
Watch here:
https://t.co/dcKsxAt4Dk
Exploratory work is well served by a quick generated persona. Nothing is committed, and a wrong answer is cheap to discover.
A media budget is a different job. A quarter's campaign, a regional launch, a pricing call. Something happens because of the answer, and the money goes either way.
The honest distinction is not capability. A language model will write you a convincing persona. What it cannot do is show which real population that persona came from, so there is no variable to inspect. That gap stays invisible until somebody asks you to justify the decision, which on any decision carrying value is always.
Whichever tool produced your answer, the questions are the same.
Which dataset?
What sample, at what geographic resolution?
What error band?
When it was last refreshed?
→ https://t.co/LDtd0CZozW
America's highest household incomes are not spread out. Of the 20 richest US counties, 8 are Washington DC suburbs and 4 are in the San Francisco Bay Area. Loudoun County leads at $178,707. https://t.co/kBO4f6MsFL → https://t.co/0Dg9rDtpdY
Bring context to your persona chats with images and URLs.
Test everything before the budget goes out. Every ad, every page, every idea, on an audience built from real data.
Start for free - https://t.co/YyVuXn5Os2
Georgia's richest county, Forsyth, has a median household income of $138,000. Its poorest, Randolph, $25,425. More than five times apart, the widest gap of any US state. A state average describes neither. → https://t.co/gOaVjrnYrm
It's never been easier to hear from your target audience.
In Cambium AI, you can now group your synthetic personas, label them, and pull just that group to poll or talk to. For example, Gen Z renters, on demand, ready to answer.
Built from real US public data. Hear from the people you're building for before you spend.
Los Alamos County, New Mexico: median household income near $143,000, poverty under 3%. The typical New Mexico county sits near $52,000. One county can skew a whole state average. → https://t.co/lslBcbQVNx