Free preview · no account needed
Welcome
A decision model is an AI model that reads your text and picks one answer from a set you define. It doesn't write anything: what comes back is one of your answers, never a sentence.
- Routing, triage, flagging and checking — the same well-defined call, made over and over. Which team gets this ticket? Does this email ask to change bank details? Does this draft promise a refund?
- Fast and cheap per call — quick enough to sit inside a live process, and cheap enough to run on every item, not just a sample.
- A probability with every pick — the decision models you'll use here give one for each answer, so once you've checked what those numbers mean on your own work, your process can act on the clear cases and pass the rest to a person.
- Writing, explaining or summarising — it writes no text at all, not even a reason for its pick. That's your AI assistant's job.
- Maths, dates and counting — adding up invoice lines, or checking whether a notice period is over 30 days. A formula does these better.
- Open-ended judgment — anything without a fixed set of answers, such as "what should we do about this customer?"
- Anything it can't see — it knows only what you send it: not your order system, not last week's thread.
- Decisions about people's jobs — hiring, promotion, termination, or anything that rates or ranks a person. Those stay with a person.
Even on the calls it's good at, a well-formed answer isn't always a right one. Every pick is one of your options, and it can still be the wrong one, so you check it on your own work before you rely on it.
This phase starts from zero, for anyone who has never heard of a decision model, and everything in it works in a browser with no code.
The running example
One example runs through Decision Models: a support inbox. Every message that arrives raises the same three small questions, and a decision model can answer each of them.
- Which team? — billing, technical, account or other.
- How urgent? — anywhere from "a question with no deadline" to "the customer can't work at all".
- Does it ask for a refund? — yes or no.
Each one comes back in a different shape. You'll meet the shapes in the second module.
What this phase covers
- Uses and limits — what it is, how it differs from the chat assistants you already know, and which calls at work it can take.
- Answers and probabilities — what happens underneath, the three answer shapes and what its probabilities do and don't promise, and a first run on Laya (everyone) or Jev (if you have access). The tools themselves have their own part, Tools.
Start with the simplest question of all: what is a decision model?