AI Revenue Operations
A working pipeline-risk engine for a fictional $40M-ARR SaaS company. It reads the opportunity table, applies deterministic risk signals, quantifies forecast exposure, and writes the CRO's Monday brief — with the reasoning visible at every step. Move the thresholds and the entire chain recalculates.
Quarter to date, 42 open opportunities. Everything below is computed live from the table.
These are the deterministic rules. They are tunable because every company's benchmarks differ — and because a risk model nobody can tune is a risk model nobody trusts.
Sorted by risk score. Each flag shows the rule that fired — no score appears without its reason. Click a column header to re-sort.
| Account | Segment | Owner | Stage | Forecast | Amount | Stage age | Last touch | Risk | Signals fired |
|---|
—
Open pipeline against the quarterly target for each segment.
This is the output that matters. The numbers come from the engine; the narrative structure is the part a revenue leader actually reads on a Monday morning.
The most common way an AI pipeline-review project fails is by handing the whole job to a model. Below is the boundary I hold.
Signal detection, scoring, exposure, coverage and concentration are all rules and arithmetic over CRM fields. They are auditable, reproducible, and identical for everyone who opens the report. A number a leader will defend to a board cannot come from a probabilistic generator.
The model reads the scored output plus call notes, next-step text and email threads to explain why a deal stalled, spot themes the rules can't encode ("three deals lost the same champion"), draft the narrative, and propose the action list. It never invents a dollar figure — it is handed the figures.
The forecast call stops being a recitation of the pipeline and becomes an argument about five deals. That is the whole point.