Scoring and analytics model

Speechlytics exposes two complementary call scores. They answer different questions and should not be combined without understanding their meaning.

Keyword score

The keyword score is rule-based. It answers whether configured words or phrases appeared under the expected conditions.

A keyword definition can include:

• one or more phrases;
• the channel to inspect;
• a score contribution;
• tags used for reporting;
• exclusive or out-of-scope behavior.

Speechlytics evaluates matches after transcription and aggregates the configured contributions into the call's score. Because the rules are account-specific, the same transcript can legitimately score differently for two accounts.

Analytics score

The analytics score is derived from structured conversation analysis. It answers how the conversation performed against the account's Sales or Customer Support profile.

Inputs can include communication quality, outcome, resolution, sentiment, business value, risk, next steps, and other profile-specific dimensions. The normalized result is returned as analyticsScore.

Key differences

	Keyword score	Analytics score
Source	Configured phrase rules	Structured transcript analysis
Main question	Were expected phrases present?	How did the conversation perform?
Account dependence	Keyword definitions and weights	Analysis profile and available fields
Result field	score	analyticsScore

A high keyword score does not guarantee strong conversation quality, and a strong analytics score does not guarantee that every required phrase was spoken.

Scope and exclusions

Keyword rules can mark a call out of scope or apply exclusive behavior. Consumers should read inScope together with the score instead of treating every numeric value as directly comparable.

Agent aggregation

The Intelligence module aggregates eligible call analytics into agent-level results. Averages are based on calls where the relevant value is available; unavailable values are not automatically treated as zero.

Where scores appear

• GET /api/v1/calls/{id}/status;
• GET /api/v1/calls;
• Calls list and call details;
• analytics dashboards;
• Intelligence agent scorecards.

Use scores responsibly

• Compare calls processed under the same account rules and analysis profile.
• Review transcript evidence before taking action on an outlier.
• Use trends and representative samples instead of a single call.
• Account for audio quality, channel separation, and transcription accuracy.
• Reprocess historical calls when you need them evaluated under materially changed rules.

For a shorter business explanation, see Understanding call scores.