Calls Analytics
This page analyzes all calls recorded in the CRM and shows trends, distributions, and outstanding callbacks at a glance — useful for tracking team performance and conversation quality over time.
What can I do here?
- View metrics such as total calls, pickup rate, conversion rate, and average call duration
- View trends over time (calls per day, sentiment over time)
- View distributions: outcomes, intents, call duration
- Identify top topics and the most common buying signals and objections
- View the callback backlog (open callbacks total, overdue, trend)
- Jump straight from any metric/tile into the filtered call list (drill-down)
- With an additional permission scoped to "own": view your own numbers in the callback backlog (see below)
- With an additional permission scoped to "all": view "Top Agents" and the callback backlog for every owner (personal performance data, see below)
Step by step
- Open CRM → Calls → Analytics or click "Analytics" from the call list.
- Check the KPI tiles at the top (total calls, pickup rate, conversion rate, avg. quality, avg. sentiment, open follow-ups).
- Click a bar, a segment, or a tile to jump directly to the underlying calls.
- Check the "Callback Backlog" to see how many callbacks are open or overdue and how they're distributed across owners.
Fields explained
| Field | Meaning | Notes/Effect |
|---|---|---|
| Calls Total / Calls Answered | Total number of calls, or number successfully connected, in the selected time range | |
| Avg. Call Duration | Average duration of the calls | |
| Pickup Rate | Share of answered calls out of all calls | |
| Conversion Rate | Share of calls that led to a positive outcome | |
| Avg. Quality Score / Avg. Sentiment Score | Average AI-rated conversation quality/sentiment | Based only on analyzed calls |
| Open Follow-ups | Number of calls with a callback that hasn't been completed yet | |
| Calls Per Day | Call volume over time | |
| Outcome Distribution / Intent Distribution | Breakdown of calls by outcome or detected customer intent | |
| Top Agents | Owners with the most or best calls | Only visible with an additional permission at scope "all" (see below) |
| Top Topics / Top Buying Signals / Top Objections | Most frequent conversation content detected by the AI | |
| Callback Backlog | Open and overdue callbacks, 7-day trend | Overdue threshold can be set to 24/48/72 hours |
| Callback Backlog by Owner | Backlog broken down per owner | Only visible with an additional permission; at scope "own", your own row only (see below) |
Values & statuses
This page doesn't define its own value lists — it aggregates the fields explained on the Calls page (direction, outcome, status, callback status).
Personal performance data ("Top Agents", callback backlog by owner)
The "Top Agents" and "Callback Backlog by Owner" tiles list performance numbers for individual, named employees (call count, conversions, open callbacks per person). That is personal performance evaluation, and it is gated on its own permission, not just the general read permission for calls.
The permission carries a scope. It decides whose numbers you see:
| Scope | What you see |
|---|---|
| own | Your own numbers in the callback backlog, and nothing else. The "Top Agents" tile does not appear. |
| all | Every owner and the "Top Agents" ranking - unchanged. |
| (permission not granted) | Neither tile, not even your own numbers. |
- Seeing your own numbers is not a special case of judging other people. Someone who may not look at their own performance can neither place themselves nor object - and the number exists regardless. That is why scope "own" lets you see yourself without showing you your colleagues' numbers.
- "Top Agents" stays locked at scope "own". A ranking is only meaningful with other people's numbers, and a one-row leaderboard would give away your own position in the field. The tile is then absent: not empty, no placeholder.
- The scope is enforced on the server, not hidden in the display. Calling the underlying endpoint directly at scope "own" returns your own row only - no other names are in the response. Without the permission, it returns a rejection, never filtered or empty data.
- Every other tile (total calls, pickup rate, conversion rate, callback backlog total/overdue/trend, and so on) stays visible to anyone with read access.
- Who gets which scope is decided by your organization's role and permission administration. Roles that may read calls receive the self-service scope ("own") automatically; scope "all" is typically reserved for team or department leads.
Frequently asked questions
What does "overdue" mean in the callback backlog?
A callback is considered overdue once it has been open longer than the configured threshold (24, 48, or 72 hours).
Why doesn't the call count here exactly match the call list?
The analytics view is based on a selected time range and possibly other active filters than the call list.
Can I jump from a metric directly to the underlying calls?
Yes, practically every tile, bar, and segment is clickable and opens the correspondingly filtered call list.
I don't see "Top Agents" or the callback backlog by owner - is that a bug?
No. Both tiles show performance numbers tied to named individuals and are gated on a separate permission (see "Personal performance data" above). If you don't hold that permission, these two tiles are intentionally not shown; every other tile keeps working normally.
The callback backlog only shows me, and "Top Agents" is missing - is that a bug?
No. Your permission carries the scope "own": you see your own numbers, no one else's, and the "Top Agents" ranking is not shown at all. If you need the view across the team, ask your administration for scope "all".
I have scope "own" but see nothing at all - why?
Most likely your user account is not linked to an employee record, so there is no way to tell which calls are yours and the breakdown stays empty. Contact your administration - the case is logged and traceable there.
How the metrics are calculated
The metrics summarise all calls in the selected period (and active filters). They are recomputed fresh on every load; only the three "top" lists (topics, buying signals, objections) are cached for up to 5 minutes for performance.
| Metric | Calculation |
|---|---|
| Pickup rate | answered calls (outcome "Connected") divided by all calls |
| Conversion rate | calls with outcome "Interested" divided by the answered calls (not by all) |
| Ø talk duration | average of the recorded talk durations (calls without a duration are not counted) |
| Ø quality score | average of the AI quality values — only over calls already analysed; calls without AI analysis are excluded |
| Ø sentiment score | average of the AI sentiment values, likewise only over analysed calls |
| Open follow-ups | number of calls where the AI flagged a callback / follow-up need |
How the per-call AI values (quality, sentiment, topics …) are produced is explained on the Calls page.
Callback backlog: an open callback counts as overdue once it has been open longer than the configured threshold. You choose the threshold in the view (24, 48 or 72 hours); the default is 24 hours.
Reviewing and correcting speaker roles in existing transcripts
Transcripts created before direction-based speaker assignment carry a mapping that stems solely from the earlier AI guess - "Employee" and "Contact" may be swapped in them. How this happened and what determines the mapping today is explained in the "Who is speaking in the transcript" section on the Calls page.
The "Speaker roles in existing transcripts" block shows how many of your existing transcripts still carry an unverified mapping, grouped by category:
| Category | Meaning |
|---|---|
| Direction present and contradictory | The call direction contradicts the current mapping - the backfill flips the roles. |
| Direction missing | No direction is recorded on the call - the mapping stays a genuine doubt case ("speaker mapping unverified"). |
| No speaker labels | The transcript is not a two-speaker dialog. |
| Direction present, already correct | The mapping is already right - the backfill only stamps the confidence without flipping anything. |
Dry run and correction
- Dry run changes nothing. It only reports how many transcripts a sharp run would correct - useful to gauge the scope up front.
- Apply correction straightens the affected existing transcripts: roles are re-set based on the call direction (including the formatted transcript, segments, and any cached translation). Nothing is re-transcribed, so there are no AI costs. The action is logged (who, when) and requires write permission for calls.
The backfill only touches transcripts without a verified mapping. A second run therefore does not flip already-corrected transcripts again. After the run, every resolvable transcript carries a verified mapping; only the genuine doubt cases (missing direction, no speaker labels) stay flagged as "speaker mapping unverified" and can be corrected individually in the call detail via "Swap speakers".
How to improve the metrics
- Record outcomes consistently. Pickup and conversion rates depend directly on the outcome values "Connected" and "Interested" — only cleanly recorded calls are counted correctly.
- Enable recordings. Ø quality and Ø sentiment arise only from analysed calls; without a recording/analysis these averages stay thin.
- Work off callbacks promptly. This lowers the "open follow-ups" and the overdue share in the callback backlog.