Introducing Till-1.1

Your next product launch has already happened. We simulated it this morning.‍

We’re now simulating the American consumer thousands of times a day, then using last night’s receipts to grade and improve the model every morning.

Today I’m announcing Till 1.1, our proprietary demand model built to understand and predict real consumer behavior. It’s the result of seven years of building, testing, getting things wrong, improving them, and investing heavily in data science, machine learning, compute, training, and model tuning.

The demand gap

There’s a basic problem in this industry that still hasn’t been solved. Companies spend billions trying to understand what people want, but most of that understanding still comes from what people say they’ll do, not what they actually do.

Those two things are often very different.

A consumer can tell you they love a product, hate a brand, plan to buy something next month, or care deeply about a certain feature. Then the receipt tells a completely different story. That gap between stated intent and actual behavior is where a lot of bad decisions get made.

We spent the last five years building around the other side of that gap. Instead of asking people what they think they’ll buy, we start with what they actually bought.

Research

Attain did the unglamorous work first. We built a primary-sourced dataset: billions of transactions, trillions of dollars in purchase value, dozens of categories from retail and dining to financial services and CPG. Millions of Americans opted in and got paid for their own data, because I believe the people generating the most valuable dataset in capitalism deserve a cut of it. We built all of it to measure, optimize, and personalize marketing outcomes.

Nvidia sold gaming chips for two decades before the world noticed they'd built the engine of the AI age. Amazon wired up server infrastructure to sell paperbacks and it became AWS. The pattern repeats: the asset built for one job becomes the launchpad for a much bigger one. Somewhere in the last eighteen months, staring at our own pipes, we saw it.

The dataset we built to measure marketing was also the largest labeled training set of American buying behavior ever assembled. Surveys capture what people say. Our corpus captures what they did: receipt by receipt, timestamped, verified, refreshed in real time.

So we invested in experimentation. 

Thousands of hours of engineering and research nobody saw. We turned billions of transactions into a massive training corpus. The team built agents that interview real, verified buyers at the exact moment behavior changes: the first purchase, the brand switch, the quiet churn. We stood up an evaluation harness that grades every prediction against the purchase that follows it. And we run compute that never sleeps, because a model that retrains daily on what the country just bought doesn't take weekends.

The result is a specialized model of consumer demand. It runs thousands of simulations a day. It gets sharper every morning. And unlike the synthetic twins and AI panels raising billions right now on rented data, it doesn't grade its own homework. Reality grades it. We just kept the receipts.

The "terminal" demand never had

Here's what I have been telling the team here at Attain: marketing will be the smallest thing this model ever does.

Bloomberg gave institutional capital a terminal four decades ago. Every tick of every asset on earth, streamed to every desk that matters. Demand never got one, even though what three hundred million people decide to buy next is the force that actually moves the economy. We've been reading it through surveys, quarterly filings, and vibes. That ends now.

A model built on real purchases reads the consumer economy before the government prints the number. It calls revenue while the quarter is still breathing. It prices credit on how people actually live instead of how a bureau file remembers them. It tells a retailer what to stock before the truck leaves the depot. It lets a policymaker simulate a tariff against real household behavior before it lands on real households. And when shopping agents start buying on behalf of every family in America, those agents will need ground truth about human preference. We're the layer they'll query.

We sequenced the human genome and it rewired medicine. Nobody ever sequenced the human checkout. That's the project: a living model of demand connected to an instrument that can read it. Every prediction gets settled against a real purchase. When we're right, the till proves it. When we're wrong, the till proves that too, and the model wakes up smarter. This is the scoreboard for a new economy.

The demand for demand

I expected the market to need convincing. I was flat wrong. The pull has been the loudest signal of my career. A few weeks of quiet conversations, and more than a dozen enterprises are already on the waitlist. Conversations that began as research pilots are ending as negotiations for model access. And the callers aren't who we pitched: funds want the demand signal, retailers want the forecast, platforms want the ground-truth layer under their own AI. When customers you never called start describing your roadmap back to you, you've caught a wave.

What we believe

A few things we hold at Attain, stated plainly.

People deserve to know how their data gets used and to get paid when it does. Full stop. The training corpus of the AI economy is being strip-mined from the public for free. We built the counter-model: opted in, compensated, dignified.

Models that grade their own homework will have a good year. Models graded by reality will have a good few decades.

The economy has plenty of opinions about itself. What it's missing is an instrument.

Come look

Attain made its name measuring and powering the outcome. Till anticipates what comes next. If you're a brand tired of testimony, a fund tired of stale panels, a builder who wants to work on the model with an answer key, or a skeptic, especially a skeptic: we built the scoreboard for you.

Agents to humans to Attain.

Bring receipts!