How it thinks
A prediction machine for your company.
Your brain doesn’t wait for input. It predicts what should happen and spends its energy only where reality disagrees. EndoBrain works the same way: it holds a model of your company, checks it against what happened and speaks up only when the gap matters.
The loop
One loop, running every day.
Expect
A model of your company predicts the day.
Notice
Only the gap between expected and real counts.
Stay quiet
Nothing new means no message at all.
Tidy up
At night, contradictions become questions.
One move
In the morning, one move with the numbers.
Learn
Your decision goes back into the model.
It expects
It holds a model of your company.
Money in and out, open invoices, leads, promises, the rhythm of your entries. From what it already knows, it builds an expectation for each of them. Every morning it compares that expectation with what actually happened. The model is not a chart you read. It is what lets the brain notice that something is off before you do.
| Expected | Happened | |
|---|---|---|
| Money in, October | €6,200 | €4,750 |
| Open invoices | 1 | 3 |
| Entries this week | 12 | 4 |
It notices
It only pays attention to surprise.
The gap between what it expected and what happened is the only signal it acts on. Each gap is weighted by how reliable the data behind it is. A month with only nine days of entries is not a collapse in revenue, it is a month you haven’t logged yet. So it doesn’t raise an alarm. It asks.
9 of 31 days logged
Not an alarm. A question: “Did you log everything for October?”
It stays quiet
Nothing new? Silence.
When nothing changed, there is no message, no notification, not even a call to the model. It gets used to what repeats, so the same advice doesn’t come back every day. A move you dropped stays dropped. A move you postponed comes back after a week, not tomorrow.
- Messages
- 0
- Notifications
- 0
- Calls to the model
- 0
Same as yesterday. Nothing to say.
At night
While you sleep, it tidies its memory.
Memory gathered from conversation contradicts itself sometimes. A client is still listed as a lead. A lost lead has a follow-up next week. At night it looks for exactly this. What it finds becomes a question for you in the morning. It never settles the matter by deleting something.
Atelier Weiss is a lead and, since March, also a client.
Question for youIs Atelier Weiss a client now?
The fast path
Some things can’t wait for the morning.
An overdue invoice. A payment date that passed. These don’t need thinking, they need speed. Plain rules in code catch them and reach your phone at once, without asking the model. Fast and boring on purpose, so it is never wrong about the obvious.
Invoice 031 for Nordlicht GmbH is 14 days overdue. €1,450.
In the morning
One move, with the numbers.
Everything it noticed is boiled down to one brief: the main thing in a sentence, one move and at most one question. Every figure in it comes from code, and a guard checks the text before you see it. You decide what happens next.
- Main thing
- Two September invoices are still open, €1,450 in total.
- One move
- Send Nordlicht a friendly reminder today. Last time, one reminder was enough.
- One question
- Is the spring campaign with Atelier Weiss still on for March?
It knows what it doesn’t know
Confidence is computed, not claimed.
How sure it is depends on how complete the data is and how well its last forecasts held. Code computes that score. The model doesn’t get a vote. When the forecast is unreliable, the brief speaks without amounts rather than with wrong ones.
- Month complete
- high
- Last forecast off by
- 18 %
- Registry quiet for
- 2 days
In progress
It learns from what you decide.
Done, later, drop: every decision is remembered, with your reason if you give one. Your preferences become part of its memory, not a passing instruction. Keeping score of which kinds of moves work for your company is the next step. No retraining, only your own history.
- Done
- Reminder to Nordlicht
- Later
- Raise the Basic price · back in 7 days
- Drop
- Cold-call list B · “not now”
Inspired by neuroscience
Predictive coding, precision weighting, repetition suppression, sleep replay, the fast path of the amygdala, metacognition. An engineering analogy, not biology: we take the principles that work, not the anatomy.