Life Insurance Analytics

How each recorded call is scored: the predicted chance of a sale, how hard the customer pushed back, the re-contact lead list built from both, and how each agent executed the sales process.

Confidential

How the model's score is built

The model starts from the same point for every call and adds what the agent's and the customer's behaviour on this call say. The parts are on the model's internal scale: positive pushes toward a sale, negative toward no sale. The result is turned into the score on the right.

What moved this call's score the features with the largest effect, by role

Agent behaviour

23 measures of how the agent spoke: talking time, turn length, questions, pauses, sentiment, topics.

Customer behaviour

23 measures of how the customer spoke. This total is the "customer contribution" the lead list ranks on.
pushes toward a sale pushes toward no sale High / Low: this call's value against the median call. Hover a bar for what the measure means.

Where this call sits agent part against customer part, among calls the model never trained on

saleno sale ◯ this call
The model reads timing and turn-taking only, never the words. Its score orders calls by how sale-like they sound; it is not the chance of a sale in percent (the model was trained on equal numbers of sales and non-sales). Version: the current production model (v1).

How the resistance score is built

A taxonomy-based heuristic model on the call's behaviour labels: how hard the customer's objections were, how often they came back, whether a hard objection softened, and how hard the reasons behind them are to overcome. Every number runs the same way: higher means harder.

Where this call sits

calls that raised an objection │ dashed: strong / weak cut (median)│ cap: removed from the lead list above it ▼ this call

Objections raised

Difficulty on the taxonomy's scale, 0.10 the easiest objection to get past and 0.90 the hardest. Hover a name for its definition.

Underlying reasons

How hard the reason behind the pushback is to overcome, same scale. A call-purpose label (the customer called about something else) is kept as a flag and not scored.

How the list is built

at least any