330 million synthetic residents, each placed in a real census tract, each carrying the demographic, housing, health, and economic characteristics that decide who buys, where, and when. No personal data. 98% correlation with the real U.S. population.
The Twin architecture was engineered at Harvard University over two years on Harvard's high-performance computing clusters, and is licensed by its co-inventors through 9 Foundations. It preserves how characteristics co-occur within a person, at a geographic resolution precise enough for action and stable enough to stay accurate over time.
Privacy rules make every population dataset you can buy pick one. Individual-level data shows how characteristics interact within a person, but it is released only at coarse geography. Neighborhood data is precise about place, but reports one characteristic at a time, averaged across everyone. So decisions get made on incomplete data, and budget gets spread across markets that are not equally valuable.
Safe but vague. "Average income in ZIP 10001." One characteristic at a time, averaged across everyone who lives there.
Specific but risky. Heavy regulation, incomplete coverage, privacy liability. And a list still only knows what is on the list.
330 million synthetic residents, each modeled as a whole person, each placed in a real census tract. The patterns are real. The people don't exist.
Twin began in a Harvard lab as a public-health question: which children are at risk of lead exposure? Risk concentrates where two things overlap, young children and pre-1980 housing. The standard approach takes the top quartile for each characteristic separately and calls the overlap high-risk. Twin models both characteristics jointly inside each synthetic resident, then places that resident in a real census tract.
Across the U.S., the standard approach missed 73% of high-risk tracts: 3.28 million at-risk children, overlooked. That is when we understood what the method could find for anyone.
The same method now finds customers. Who are you missing in your market?
Take the top quartile for each characteristic separately. Call the overlap high-risk.
Model every characteristic jointly inside each synthetic resident, then place that resident in a real tract.
at-risk children the standard method missed nationwide.
The map at the top of this page is this archetype in New York City: 830,661 children, placed tract by tract.
Co-inventor of the Twin architecture, originally built in his Harvard lab. Author of Healthy Buildings and 100+ peer-reviewed papers, regular contributor to Harvard Business Review, and advisor to senior executives at JPMorgan Chase and Amazon.
Co-inventor of Twin and its computational core, originating in her doctoral research at Harvard on spatial microsimulation, statistics, and machine learning. She built the statistical logic that lets synthetic residents mirror real populations at tract resolution.
Co-inventor. Quantitative methods, building simulation, and synthetic population generation. Built national-scale residential models of the U.S. and U.K.
Environmental modeling and analytics. Leads the Advanced Analytics team and advises Fortune 500 companies on environmental health; upholds the technical standards Twin's partners require.
Work backwards from limited sales or engagement data to infer the customer profile most associated with strong performance. Not just by age and income: insurance type, disability status, language spoken at home.
Find where that profile is concentrated, in the markets you already serve and in the ones you are weighing entering.
Overlay weather, environment, and your own locations on the same tracts, so activation is timed to when demand is projected to move.
Where are the customers who want what this product does — and when?
Who people are, how households are composed, and where they came from.
3 packs · 586 hidden variables
Schooling, occupation, income, and economic position.
3 packs · 1,575 hidden variables
Spending, banking, credit, assets, and financial resilience.
6 packs · 1,584 hidden variables
Coverage, care use, conditions, function, and prevention.
13 packs · 2,547 hidden variables
Activity, consumption, sleep, food, caregiving, and family life.
8 packs · 447 hidden variables
The structure, its systems, its condition, and what it costs.
9 packs · 1,316 hidden variables
Equipment, fuels, consumption, cost, and energy burden.
9 packs · 490 hidden variables
How people move, what they drive, and where they are connected.
7 packs · 848 hidden variables
Community safety context.
3 packs · 22 hidden variables
The hidden variables are the characteristics that separate your customer from everyone else: the ones neighborhood averages can't see and lists don't carry. What kind of insurance they carry. Whether someone in the home has a disability. What language is spoken at home. Twin carries 9,415 of them.
Choose a geography. Define the customer. Add the hidden variables and the spatial context. Twin shows where they are, tract by tract, and when.
Federal surveys and the decennial census.
Synthetic residents, not records. Nothing to breach, nothing to consent, nothing to de-identify.
84,000+ census tracts, about twice the resolution of ZIP codes.
Average Pearson R² of 0.98 across all census tracts. Kullback–Leibler divergence below 0.08 for every variable tested.
A national packaged-foods manufacturer needed to rank grocers across Michigan to decide where to invest marketing, messaging, and logistics. The available methods ranked stores by household income or by total population, and neither captured the segment they were trying to reach.
A roof window manufacturer entering the U.S. market needed to know which homes its product can physically go into, and which owners are close to a renovation decision. Household income and home value answer neither question.
A regional health insurer needed personas for its priority member groups to guide communications, media, and product development. The research available was quantitative and third-party; neither showed where those members live or how their local environment shapes their care.
A home mobility company wanted to test whether a 5,000-piece direct-mail pilot across Massachusetts could complement its digital lead generation. Available lists were built on age and homeownership alone, which do not capture the combination of conditions that drives the purchase.
A prescription delivery company needed to know who was driving demand for each drug across Massachusetts, why some service areas under- or over-performed against their footprint, and how to change operating windows. Sales data showed where orders came from, not who placed them.
the customer you are trying to reach, defined variable by variable.
the hidden variables that separate them from everyone else.
your locations, third-party points, weather and environment.
the states, DMAs, or regions the decision actually covers.
scoring, ranking, or inference built for your question.
Build archetypes, see concentration, export priority tracts. Password-protected, issued per user.
Pull Twin outputs into your own environment, your agency's platform, or an activation partner's system.
9 Foundations builds the archetypes, runs the analysis, and delivers a decision-ready report. Most engagements start here.
Delivered as CSV, Excel, GeoJSON, API feed, or scheduled email alert.
Or write to twin@9Foundations.com.