Predictive Inference
See where your numbers are going before they get there.
Predictive Inference projects a target metric forward in time — demand, revenue, capacity, volume — from your historical data. It gives operators a forward view to plan against rather than reacting after the fact. Use it when the buyer's question is 'what will X be?' rather than 'what is X?' or 'why did X happen?'
Shape
Operational dimensions
Each output waits on a human decision.
Fires on a clock.
Owns a system-of-record; expensive to migrate.
Consumes external data; does not write back.
Inputs
- historical time-series for the target metric
- exogenous feature data (seasonality, calendar, external signals)
- forecast horizon and granularity
- optional scenario inputs (growth assumptions, capacity constraints)
Outputs
- projected values over the forecast horizon
- uncertainty bands (confidence intervals or quantile ranges)
- scenario branches for alternative assumptions
- backtesting accuracy metrics for model transparency
Mechanism
Produces a forward-looking numeric or distributional projection over time (forecast, demand prediction, time-series projection, scenario projection) from historical structured data.
Why this is a primitive
Cannot be decomposed: the project-the-series-forward operation is a single inference. It is structurally distinct from per-item classification (which emits a label on a current record) because the output type is a future trajectory / distribution over a time horizon, not a label on a present item. Composing classification repeatedly does not yield a forecast and vice versa — different inference shapes.
Where it shows up
Related primitives
Tags
See where it fits.
Primitives are configured into named solution shapes for each client’s domain. The fastest next step is a conversation about which shape fits your problem.
Start a conversation