An LLM can read the ticket history; it can't tell you whether ticket #4521 will close for good or bounce back next week. Drop Cortex inside your agent and every "what happens next?" gets a real answer, predicted, simulated, called over a stateless API, and the agent acts on it. Not another agent. A forecaster, inside yours.
The forecasting API is generic: it works on any structured tabular data. What changes per vertical is the target your agent forecasts and the move it makes once it knows. A few of the patterns we see most often.
| Industry | Forecast | What your agent does with it |
|---|---|---|
| Customer support | Ticket reopen risk | Escalates the tickets likely to bounce back before they close |
| Customer success | Account churn / renewal risk | Triggers a save play on the at-risk renewals |
| Support ops | Ticket volume & surge | Staffs the queue before the spike, not after the backlog |
| E-commerce | Demand per SKU | Auto-adjusts inventory orders before the stockout |
| SaaS | Customer churn probability | Triggers a retention campaign on the at-risk list |
| Marketing | Campaign conversion rate | Reallocates budget across channels |
| Finance | Revenue forecast | Generates board-ready projections |
| Hospitality | Room occupancy rate | Adjusts dynamic pricing |
| Supply chain | Lead-time variability | Reroutes shipments proactively |
| HR | Employee attrition risk | Flags at-risk employees for managers |
| Energy | Load demand | Optimizes grid dispatch |
Your vertical isn't here? The core handles any prediction problem: you upload data, name a target, and get an endpoint. Tell us what you forecast.
The same agent you ship today, but it stops guessing about the future. Cortex gives it a calibrated prediction it can act on, and a what-if it can run before it commits.
Manages catalog, pricing, and replenishment across thousands of SKUs.
Cortex adds → forecasts SKU-level demand seven days out and reorders before the stockout, not after it.
Resolves tickets and closes the loop with the customer.
Cortex adds → predicts which tickets will reopen, so it can escalate the risky ones and prove resolution actually stuck.
Runs conversion experiments across landing pages and segments.
Cortex adds → simulates which variant lifts conversion per segment before the test ships, then ranks them by predicted impact.
Answers "where is my order?" and manages delivery exceptions.
Cortex adds → predicts delivery failures before they happen, so the agent reaches out proactively instead of after the complaint.
However your agent is built, embedding Cortex tends to land in one of three patterns. Pick the one that matches how much autonomy you've given it.
The agent calls predict and presents the result as-is: value, confidence interval, and the drivers behind it. The human decides.
The agent calls predict and, when the forecast crosses a rule you set, takes the action itself: places the PO, opens the ticket, reroutes the shipment.
The agent calls simulate across several what-ifs, compares the deltas, and recommends the optimal move with the evidence attached.
Predict, simulate, and monitor over a stateless API. Your brand, your margin, we own the modeling.
Questions first? Email [email protected].