Foresight - the Business World Model
Foresight predicts what your next move will do. It keeps your company’s past — each decision, what it predicted and what really happened — and turns it into a forecast: a range, what drives it, and a version you can hold it to. Your model is built on your own history, and no one else’s model learns from you.
Problem
Decisions get made, results come in, and nothing connects the two. Six months later, nobody can say what the price change did, which channel really worked, or how long that launch took. So the next decision gets made like the last one: from memory, and confidently.
I record both sides every time: where things stood, the decision, what I predicted, and what happened. That record is what makes prediction possible at all, and Foresight is what it becomes.
How it works
Four layers we own turn your company’s record into an outcome: what an action will do, which decisions tend to work, how far the number can be trusted, and what the futures look like before you commit. Pick a layer to see what it does.
Drag sideways to see the rest
Ours
Prediction
Name a move and it tells you what that move does to your company: a range with the drivers behind it, never a bare sentence. How likely this mission is to land, whether this deal closes, how far the timeline slips, where the cash goes.
These four layers predict your results. In your workspace they’re tuned on your company’s own history, and they stay inside it.
Yours
One model per company, tuned on that company’s own history. The longer you use it, the better it gets at your business, and at nobody else’s.
Tuned on your history
Your decisions, your results, how sure it should be about you. The model fits how your company actually behaves, which is far more useful than getting better at companies in general.
Inside your company
That tuning stays inside your company's own space and is deleted with it. It never leaves, so there's nothing to leak and nothing to hand to anyone else.
Never sold, never borrowed
We don't sell customer data. API and enterprise workspaces are never used for shared training, at any setting, and helping the shared model is a choice you make: the shape of decisions, never your content.
What you get
Foresight never gives you a bare number. Each prediction is something you can open:
Prediction record
foresight-1Saved the moment it’s made: range, drivers and model version, sealed before the answer is known, so you can read back what it said once the result is in.
Worked example
Decision D-104 · good conversations
Drag sideways to see the rest
expected range · sealed when you decided
Aster, week two. The model predicted 8–14 good conversations for the design-partner path, and gave that number with its drivers and its version attached, because a number nobody can question isn’t proof.
The decision record froze it the moment the decision was made. Not summarized, not revised once the answer was known: kept as it was.
Day 60: eleven. Inside the range, and you can trust that only because the range was written down in week two and couldn’t be edited afterwards.
Today a person scores a prediction against its result. Automatic grading, and the confidence record it would feed, aren’t running yet.
Kept honest
Each prediction carries its range, its drivers and the model version that made it, written down before the result exists. That’s what makes “how sure” checkable at all: the record is fixed, so nobody can quietly rewrite it to match how things turned out.
What I know feeds the model. Its prediction goes on record with the decision, and stays there to compare with the result.
The hard question
Your model trains on your company, stays inside your company's own space, and is deleted with it. The shared model has exactly one way in: a structure-only export — the shape of a decision, what kind of proof it had, its prediction and its result — with names and IDs stripped, an anonymity check in front of it, and only from companies whose consent setting allows it. API and enterprise workspaces are never used for shared training, at any setting, and we don't sell customer data. Your documents, your messages and your customers' records are not training material.
Because I made them checkable before you asked. Each prediction is saved the moment it's made, with its range, its drivers and its version, and that record is sealed: it can't be changed once the answer is known. Checkable isn't the same as checked. Automatic grading against results isn't running yet, so we publish no accuracy number, and you shouldn't assume one. When it runs, the same separate Verify that checks the work will grade each prediction — never the model being graded. And BusinessBench publishes how we measure, so anyone can argue with it.