A decision layer is the tier of the enterprise stack that sits above the systems of record and beneath the people who decide. It turns the data an organisation already holds into predictions with tested odds, early warnings routed to named owners, plans tracked against outcomes, and answers computed inside the asker's permissions. A system of record documents what happened. A decision layer changes what happens next, and keeps the score that proves it.
The term exists because a gap exists. Twenty years of enterprise software produced superb machinery for recording, tracking and reporting, and records are not decisions. The judgement that actually runs the business still lives in spreadsheets, inboxes and meetings, one floor above the systems that were supposed to inform it. The decision layer is the name for the tier that closes that gap.
The four requirements
A system qualifies as a decision layer when it meets four requirements, and each one is checkable.
- Predictions with published odds. Every claim about the future carries a probability that has been tested against what actually happened, so that of everything called 70%, about 70% lands.
- Deterministic numbers. The same inputs produce the same figure, every time, so two people can reconcile what they saw and an auditor can regenerate it.
- A named owner and a threshold. Every material signal reaches one accountable person with a deadline, so millions of events become a handful of decisions rather than a feed.
- A record of outcomes. Every call is scored against what happened next, so the organisation learns which of its predictions, and which of its people's overrides, deserve trust.
Two properties hold the four together. Everything runs on one event backbone, so a prediction, the intervention it triggered and the outcome it produced stay connected end to end. And every number is computed inside the viewer's entitlements, so each person sees their own honest total and nothing beyond it.
What a decision layer is not
The boundaries define the term as much as the requirements do.
Business intelligence describes. It answers what happened, in charts, and leaves what to do about it entirely to the reader. A copilot on a system of record summarises the past faster, which is useful and is not deciding. Workflow and process tools route tasks between people; they move the work without predicting anything or scoring what was decided. And a rules engine does decide, but without odds, calibration or learning, so it cannot tell you how often it has been right.
None of these are failures. Each does its job. But a dashboard with predictions bolted on, a chat window over records, or a governance platform standing beside the stack does not become a decision layer by adopting the word. If the odds are untested, the numbers change on refresh, nobody owns the exception, or no outcome is ever scored, the layer is missing exactly where it matters.
Analysts file this territory under Decision Intelligence, and the shelf is useful for finding vendors. The label undersells the job. Insight is the entry fee; a decision layer's product is owned, scored outcomes.
A system of record is complete when it describes the past. A decision layer is complete when the next outcome improves.
Prophesee is built as exactly this layer, with the four requirements as the architecture rather than the aspiration. Where records become decisions. See it on your own data.