Understanding Call Scores & Analytics
This guide explains, in plain terms, how Speechlytics scores calls and what the numbers in your call results mean. No technical background needed.
The two scores on every call
Every analyzed call gets two separate scores. They measure different things, so it's normal for them to differ on the same call.
1. Keyword Score — "Did the important words come up?"
This is a rules-based score you control. You tell the system which words and phrases matter to your business (for example, a greeting, a compliance statement, or an upsell offer) and how many points each one is worth. When a call's transcript contains those phrases, the call earns those points.
Think of it as a checklist: the more of your important phrases appear in the call, the higher the score.
2. Analytics Score — "How good was the call overall?"
This is an AI-generated score. The system's AI reads the whole conversation and rates the call's overall quality on a scale, which is then shown as a percentage from 0 to 100. It reflects things like clarity, professionalism, and whether the customer's issue was resolved — a holistic judgment rather than a phrase checklist.
Think of it as a quality grade for the conversation.
How the Keyword Score is calculated
-
You define keywords. Each keyword is a group of phrases and is worth a set number of points. You decide the points.
-
The system searches the transcript. It looks for your phrases in the call, and can be set to search a specific side of the conversation (the agent's side, the customer's side, or both).
-
Points are awarded when phrases are found. A keyword can be set up two ways:
- Any phrase counts — finding even one of its phrases earns the full points.
- All phrases count — the points are split across the phrases, so you earn a share based on how many were found (find them all, get full points; find half, get half).
-
The call's final Keyword Score is the average of the keywords that matched. If nothing matched, the score is 0.
A few special keyword rules
-
Priority keywords first. You can mark certain keywords as high-priority. If any of those are found, only they count toward the score — the rest are set aside. This lets you make sure the most important phrases drive the score.
-
Out-of-scope keywords. Some phrases can mark a call as "not relevant" (out of scope). These don't add points; instead they flag the call so it can be filtered out of your relevant-call reports.
-
Instant alerts. A keyword can be set to email a chosen recipient the moment its phrase is spoken on a call — handy for compliance breaches or hot sales leads.
How the Analytics Score is calculated
The AI reviews the conversation and rates several aspects of it — for example:
- Customer satisfaction
- Agent clarity
- Professionalism and tone
- How effectively the issue was resolved
- Sales opportunity handling (for sales teams) or empathy and active listening (for support teams)
- An overall call score
Each aspect is rated on a small scale (roughly 1 to 5). The overall call score is the one shown on the call as its Analytics Score, converted to a 0–100 percentage. The individual aspect ratings feed the agent scorecards and radar charts described below.
Account reference points
Each account has a few reference values, returned with analytics results and used when reporting on scores:
- Target Score — the score you're aiming for (default 50).
- Silence Target — the silence level you consider acceptable (default 30%).
- Answer Score — a baseline setting for answered calls (default 0).
Features such as keyword matching, AI summaries and scope filtering are switched on per account.
Agent results
Scores also roll up per agent in Intelligence: average score, customer satisfaction, risk and positive-sentiment percentages, and whether the agent's recent calls score better than their earlier ones.
Quick reference
| Question | Look at |
|---|---|
| Did agents say the required phrases? | Keyword Score (score) |
| How good was the conversation overall? | Analytics Score (analyticsScore) |
| Was the call relevant to us? | In-scope flag (inScope) |
| Too much dead air? | Silence measurements per channel |
| Poor audio? | Noise measurements per channel |
| Which agents are improving? | Agent results in Intelligence |