Score Explainability

Overview

Score Explainability opens up any score in the product. Every ranking, deal risk, awareness, customer health, technical risk, ecosystem fit, comes from one scoring substrate, and this audits it: it shows the drivers behind a number, the weight profile that produced it, and exactly how your learned weights differ from the shipped defaults. It re-scores nothing, it re-runs the same engine every surface uses, so what you see is what actually ranked the entity.

Drivers

Enter the signals for an entity and the profile it is scored on, and the drill-down shows each signal's contribution to the score, side by side under the default profile and, when you have a learned override, the tuned profile. This is the "why" a rep or a manager reads first: the score is never a black box, it is a sum of named, weighted signals.

Default vs tuned weights

The scoring substrate has a learning loop: a weekly retune proposes new weights that lean toward the signals that actually predicted outcomes, and you approve or dismiss them. This surface makes that change auditable. It lists every signal's default weight, the tuned weight, and the delta, with the changed weights first, so you can see exactly what the learning loop moved and defend the score in a review. When no override has been learned, the tuned view is absent and every delta is zero, honestly.

Proposed scoring changes

The weekly retune does not change your weights on its own. Each Sunday it learns new weights for deal risk, account health and prospect awareness from the records that have an outcome (won or lost deals, retained or churned accounts, converted or lost prospects) and writes them as a proposal. The card at the top of the page shows each weight that would move, from its current value to the proposed one, and how many decided records the proposal rests on.

Approve applies the new weights, and every score for that profile uses them from then on. Dismiss leaves the current weights in place. Until you choose, your scores keep the weights you last approved, or the shipped defaults if you have never approved a change. There is at most one open proposal per profile: a newer Sunday run replaces an older one you have not decided on, and a run that learns the same weights you already use removes it.

The history under the card lists each approval and dismissal with the date, who decided, and the evidence behind it. It is part of your data export.

Deterministic

The explanation is deterministic: the same signals, profile, and override always produce the same drivers and the same weight comparison. Nothing is invented, it reports only real contributions and real weight deltas from the same scoring code that ranks your book.