Scoring readiness so the agent fires on yes, not maybe

If the agent fires the instant any one signal crosses, it will annoy good accounts and burn trust faster than it earns revenue. The job is to turn the raw trigger library into a single expansion-readiness score per account, recomputed continuously, so the agent acts on a composite yes rather than a lonely maybe. Think of it as a lead-scoring model pointed inward at your installed base instead of outward at strangers.

Build the score from the dimensions the data keeps surfacing: capacity pressure (how close to a seat or usage ceiling), adoption momentum (week-over-week growth versus the account's own baseline), value realisation (milestones hit, depth of feature use), and breadth (how many teams or departments are live). Each contributes; none alone decides. Identifying the right offer for the right customer at the right time means correlating product usage against the customer's stated goals, not reading one metric in isolation.

Do not guess the weights and leave them. Wire the score so closed-won and closed-lost expansions feed back in. If capacity-pressure accounts close at three times the rate of pure adoption-momentum ones, the score should learn that. This is the part humans get wrong, because they cannot hold forty accounts' worth of correlations in their head, and it is precisely where an agent earns its place: it can hold the whole population and keep the weighting honest. Tightness matters as much as accuracy here. A product-qualified definition that converts below 15% is too loose and you are firing offers at accounts that were not ready; above 40% is too tight and you are sleeping through real intent. The score should be tuned to land the flagged cohort in that 25 to 35% conversion band.

Resolve the score into bands, because different scores deserve different actions. A hot band (high capacity pressure plus realised value) earns an immediate, named offer. A warm band earns a nurture or an in-product nudge that lets the customer self-expand. A watch band stays silent and keeps accumulating signal. Mapping bands to actions is what stops the agent oscillating between silence and spam.

Finally, prove the score before you trust it. A readiness score is a claim about the future, so test it like one: backtest it against the last few quarters of expansions. Did the accounts the score would have flagged hot actually expand, and did the watch band stay quiet? A score that cannot retro-predict your own closed deals is not ready to drive a real offer.