Call Center Report
Analysis results answer questions about one call. The Call Center Report answers questions about your whole operation: how compliance is trending, which criterion agents fail most often, why customers call, and how the night shift compares with the day shift.
Open it from Call Center Report in the left navigation.

The report covers every analyzed call, no matter which analysis run produced it — a manual run over 500 recordings and a batch analyzed automatically on arrival feed the same dashboard.
Agents can have their own login, where this report and its PDF cover only their own calls. See Agent Access.
Filters
Section titled “Filters”| Filter | Options |
|---|---|
| Project | All call projects, or a single one |
| Date range | Month to date, last month, last 30 days, last 90 days, year to date, all time, or a custom range |

Dates filter on the call date/time, which comes from your call metadata. Calls without metadata fall back to their upload time. Days and months start at midnight in the time zone set in your profile (UTC if none is set).
While calls are being analyzed automatically, the header shows Analyzing new calls… and the dashboard refreshes as results land. Otherwise it shows how many calls have been analyzed in total.
A second figure — N duplicate uploads ignored — appears when a file was uploaded again under a name the project already had. Those count as the same call rather than inflating the totals.
Calls not analyzed yet
Section titled “Calls not analyzed yet”The report covers analyzed calls only. A call that has been transcribed but not analyzed — for example one uploaded through the API with its agent and date while automatic analysis was off — is left out of every figure, the agent table and the calls list, so it never shows up as an agent with calls but no scores.
When the selected period has such calls, a warning above the metrics says how many: “N calls in this period have not been analyzed and are not included in this report.” Analyze them once with New Analysis on the project, and turn on automatic analysis so later calls are analyzed as they arrive. Calls already queued for automatic analysis are not counted; the header shows Analyzing new calls… for those instead.
Headline metrics
Section titled “Headline metrics”| Metric | How it is calculated |
|---|---|
| Calls analyzed | Number of analyzed calls in the range |
| Hours of audio | Total duration of those calls |
| Average compliance | Mean COPC score across the calls that were scored |
| Negative sentiment | Share of calls with negative sentiment, among calls where sentiment was determined |
A note appears when some calls have no client-provided date/time, telling you how many — those calls are placed by upload time in the trend and are excluded from the shift comparison.
Quality scorecard by criterion
Section titled “Quality scorecard by criterion”Average result per rubric criterion across all scored calls.
| Column | Meaning |
|---|---|
| Evaluation criterion | Criterion name from the rubric |
| Weight | The criterion’s share of the total available points |
| Result | Average score, as a percentage of the criterion’s maximum |
| Applied to | Calls the criterion applied to — 25 of 25, or 12 of 25 · 13 N/A when it did not apply to some. Calls analyzed before Uspeech recorded this are listed apart as no data. See How Scoring Works |
| Trend vs previous period | Change in percentage points against the previous period, whose dates are shown under the column title |
| Critical errors | Number of calls that triggered a critical error on this criterion |
When the project has an evaluation rubric attached, criteria are listed in rubric order.
The previous period follows the calendar:
- Whole months are compared with as many months before: Last month or a custom September 1–30 with all of August, a quarter with the previous quarter.
- A range from the 1st inside one month, such as Month to date, is compared with the same days of the previous month: October 1–15 with September 1–15.
- Any other range is compared with a window of the same length just before it: Last 30 days with the 30 days before.
The trend column stays empty for the all time range, and for a criterion with no scored calls in the previous period.
Compliance trend
Section titled “Compliance trend”Average compliance over the selected period, so a drift shows up before it becomes a quarterly problem. Choose By day, By week (the default, weeks start on Monday) or By month in the corner of the chart.
Contact reasons
Section titled “Contact reasons”Why customers called, with each reason’s share of volume. The top eight reasons are listed; everything else is grouped into Other.

Reasons are categories, not raw text. Uspeech groups the free-text topic extracted from each call into a manageable set of contact reasons — otherwise 600 calls would produce 600 unique topics.
- Projects using frozen predefined codes classify every call against those codes as it is analyzed
- Projects with open-ended categories are re-grouped on a daily sweep, or immediately when you press Refresh contact reasons
Between re-groupings, calls that have not been classified yet appear under their raw topic, so nothing goes missing from the report.
Shift comparison
Section titled “Shift comparison”Day shift versus night shift, each with calls, compliance, critical errors and negative sentiment. The day shift is 08:00–20:00 by default.
Only calls with a real call date/time are classified — a call placed by upload time could land in the wrong shift, so it is left out rather than counted incorrectly. This is another reason to upload a CSV manifest.
Agent performance
Section titled “Agent performance”Per-agent rollup: calls, compliance, critical errors, negative sentiment and goal-reached rate. Every column is sortable, so “who needs coaching first” is one click away.

Agents come from the agent_id / agent_name in your call metadata, falling back to the name the AI extracted from the conversation.
Personal report downloads one agent’s personal PDF for the projects and dates you have selected, written to the agent from their own figures. Send it to them as it is: the agent doesn’t need a login. It has no download icon for an agent known only by the name the AI extracted.
This table is not shown to agents.
The goal-reached rate comes from the rubric’s goal of the call criterion, or from Calls Analysis when the rubric has none. It stays empty when neither is available.
Exemplary calls
Section titled “Exemplary calls”The rest of the report points at problems. This section points at the calls worth copying: the period’s best calls, ready to be used as reference material when training other agents.

A call qualifies when:
- It has no critical error
- Most of the rubric applied to it — at least 3 criteria, and at least 60% of the rubric’s criteria
- It scored 85% or more on the criteria that applied
The score that matters here is the score on applied criteria. A criterion that did not apply to a call is awarded its full points (see How Scoring Works), so a short call where little of the rubric applied can read 100% overall without showing much. The N of M criteria line under each score shows how much of the rubric the call actually exercised.
Qualifying calls are ranked:
- Goal reached first, then calls where it is unknown, then calls that did not reach their goal
- Then by score on applied criteria, fewer mistakes, more criteria applied, and the most recent call
| Column | Meaning |
|---|---|
| Score on applied criteria | Points over possible points, counting only the criteria that applied |
| Goal reached | Yes, No, or — when unknown. See The goal of the call |
| Strengths | Up to three of the heaviest criteria the agent met in full |
Click a row to open the full call detail, with every criterion’s justification and quote.
Show picks how long the list is: 5, 10, 20 or 50 calls. It is a ceiling, never a quota. When fewer calls meet the bar, the list is shorter — “Only N calls meet the exemplary bar in this period” — and it may be empty. To keep the list varied, no agent appears more than twice.
Best practices to replicate
Section titled “Best practices to replicate”Below the list, Generate best practices turns the exemplary calls into training material: three to five behaviors other agents should copy, each with a short explanation and a quote from the call it came from.

- The AI reads what the evaluation recorded for up to ten of the best exemplary calls: the criteria met in full, why, and the supporting quotes. It does not read whole transcripts.
- It prefers behaviors that recur across several calls, and behaviors from calls that reached their goal.
- Every quote is checked against the call it names. A practice whose quote cannot be found word for word in that call is left out, so what you see was really said.
- Click the agent and file under a practice to open that call.
Generation takes a few minutes and runs in the background — you can leave the page and come back. The result is kept: anyone who opens the report for the same calls, in the same language, sees it straight away without generating again. When the list changes — a different period or project, a different Show value, or a call re-analyzed — the button appears again for the new set. If generation fails, Try again starts it over.
Best practices are written in your interface language; the quotes stay in the language of the call.
A browser over every analyzed call, independent of which analysis run produced it.

- Filters — agent (including (unassigned)), sentiment, plus the report’s project and date filters
- Search — file name, topic, caller number or campaign
- Exemplary only — shows only the calls that meet the exemplary bar, best first. Unlike the section above, it is not limited by Show or by the two-calls-per-agent rule
- Sorting — date, agent, compliance and mistakes. Calls without a score are always listed last, whichever direction you sort
- Pagination — 25 calls per page

Click any row to open the full detail of that call: criterion-by-criterion scores with the AI’s justification and the supporting quote, mistakes and missing protocol steps, custom-question answers, and links from each quote into the transcript.
Custom question answers
Section titled “Custom question answers”If the project answers custom questions on each call, an extra panel lists the answers per call, with one column per question. It appears only when answers exist.
Export PDF
Section titled “Export PDF”Export PDF produces a client-ready report for the current filters — the same project, period and comparison period as the screen — containing:
- Cover metrics — calls analyzed, coverage, hours processed, average compliance. Coverage is the share of every call received in the period that was scored, so calls not analyzed yet lower it, and a note under the metrics says how many there are
- Executive summary, key findings and prioritized recommendations
- Quality scorecard with a weighted overall compliance row and a short observation per criterion
- Contact reasons with a signal note per reason
- Agent performance with a coaching priority per agent
- Exemplary calls — training reference, the same calls the dashboard lists for its current Show value
- Best practices to replicate, when they have been generated on the dashboard for those same calls. Otherwise the section says “Generate best practices on the dashboard to include them in this report.”
- A methodology and confidentiality note
The narrative sections are generated by AI from the aggregated numbers only — it has no access to transcripts and never invents quotes or causes. If narrative generation fails, the PDF is still produced with the tables and a note that insights are unavailable.
An agent who exports the report gets a personal PDF instead, covering only their own calls.