Quality & AI: interpret analyzed conversations
This workspace aggregates saved recording analyses. Establish which calls have usable analysis before interpreting scores or trends. Reading existing statistics does not require a new audio analysis or a new report-generation request.
Select calls and check coverage
Choose the period and available filters, then inspect the coverage indicators. The period follows call date, not the date when AI processing happened. Unanalyzed calls, errors and partial results can reduce the usable population. Return to Recordings for a specific conversation and open its sentiment icon to examine the saved response.
Read scores and business signals
Quality is assessed on 0–100; higher is better. Churn risk uses 0–10; higher means greater risk. Read sentiment, resolution, categories, empathy, privacy, upsell opportunities, competitors, objections and pain points with their available explanations. Missing fields are unavailable, not zero. These are model assessments of recorded content, not independently confirmed customer outcomes.
Compare two periods fairly
Select periods A and B in the comparison panel. Each percentage uses only analyses containing that field; read the numerator and denominator under the value. Differences are A minus B: quality-score points, percentage points for shares and absolute counts. An unavailable measurement stays unavailable. This comparison reads saved analyses without sending a new Gemini request.
Generate and receive reports
Use Generate AI Report when you need a narrative interpretation. It sends a new request to the configured AI service and uses its quota. Check the stated period, usable-analysis count and text-summary sample limit; narrative samples are distinct from aggregate indicators. Weekly and monthly email reports require the administrator’s AI, mail and scheduling configuration. Verify a delivered report rather than assuming configuration alone proves delivery.
Read coverage before comparing quality
The sample is the selected recording analyses, not every call handled by the PBX. Read the available counts before interpreting percentages.
| Indicator or control | How to read and use it |
|---|---|
| Analysis coverage | Separate available analyses from recordings without an analysis; a missing assessment is not a low score. |
| Sentiment and resolution | Read category shares together with valid-value counts. Verify examples before interpreting a shift as a change across the whole contact center. |
| Quality, categories and daily trends | Compare score distribution, conversation categories and daily patterns. A different mix of call reasons can change an average even when the team’s work is unchanged. |
Generate a report and verify its evidence
Choose and compare periods
Set dates and the available search field in Quality & AI. Use the comparison control to choose the comparison period, then verify the displayed context and sample sizes. The comparison is meaningful only when the selected recordings are comparable.
Generate from the selected evidence
When AI is configured and permitted, click Generate AI report and wait for completion. Read the generated findings and proposed actions beside the charts; do not confuse an in-progress message with a completed report.
Return to the recordings
Open representative recordings to check transcript, audio, score and conversation signals. Use the recording-analysis field guide for churn, empathy, privacy, opportunities and objections. Record a team decision only after checking the underlying examples.
Arrange weekly and monthly email reports
Ask the administrator to configure AI, recipients, language, time zone and email delivery in System and utilities, and to verify the scheduled jobs on the server. Each Team Leader is not automatically subscribed. Verify a completed run and receipt of the report before relying on recurring delivery. If a report is absent, check processing, credentials and mail configuration rather than concluding that no calls were analyzed.
Example: quality falls while volume rises
Compare equivalent periods and coverage first. Check whether call categories changed, inspect low-scoring recordings and verify the transcript against audio. Agree on coaching or process changes with the team and review the same indicators later. A score or generated recommendation supports investigation; it does not establish an individual performance verdict.
Use the controls, understand what changes
Read coverage before percentages: each indicator uses analyses with that field available. Changing comparison periods reads saved analysis; Generate AI report requests a new narrative report. The recording date, analysis date and queue-event date can differ, so identical date labels do not imply identical samples.
The product’s information button opens on hover or focus. A click keeps the help visible; click again, click outside or press Escape to close it. The examples below reproduce selected product help text; they are teaching controls, not screenshots.
The period follows the date in the FreePBX recording filename, not the analysis date; it may differ from queue reports, which use the outcome date. Complete or partial AI results contribute only to indicators with available values; missing data is not treated as zero.
Complete product screenshots
Illustrative data. Open the screenshot to enlarge it and match the controls to the instructions in this chapter.
Open complete screenshot
Read the workspace section by section
Match the numbers and panels below to the complete workspace. These screenshots use illustrative data; the procedures above explain how to work with your own installation.
Choose the period and comparison
Start with call dates and a search term, then choose a preset or custom period. Compare it with another period using saved analyses and the same search criteria.
Understand the available evidence
Read complete, partial, processing and failed analyses alongside the available values for each indicator. A missing assessment is not a zero: coverage helps you interpret differences between periods.
Volume, sentiment and average quality
The first cards summarize the usable analyses, the three sentiment groups and the average quality score. Sentiment percentages use the available sentiment assessments; the average uses the saved quality scores.
Analysed volume
The number of usable saved analyses in the selected period and search. It describes the analysed sample, rather than every call handled by the PBX.
Sentiment
Positive, neutral and negative counts describe the assessed tone. Percentages use the calls with an available sentiment value. Return to the recording to understand the words and context behind a negative signal.
Average quality
The mean of available scores from 0 to 100. The per-call score considers politeness, resolution, empathy and clarity; its explanation remains in the recording analysis.
Explore quality, resolution and categories
Compare the distribution of sentiment, call outcomes and quality bands. Categories show the main subjects found in the saved analyses, helping you choose which conversations deserve a closer review.
Resolution
Counts distinguish resolved requests, unresolved issues and escalation. They describe the outcomes identified in available analyses, helping you choose conversations that need follow-up.
Quality distribution
Read how many available scores fall into each band. The distribution makes weaker calls visible even when the average looks positive; a real score of zero remains a valid assessment.
Leading categories
The chart shows the eight most frequent saved categories. Categories describe subjects such as technical support, administration or sales; the daily trend follows the five leading categories.
Follow the daily pattern
Read daily sentiment volumes, the average quality score and the five leading categories. A peak identifies a day to investigate; compare the available sample and return to the conversations for context.
Connect risk, opportunities and service
Bring churn risk, sales opportunities, competitor mentions and customer feedback together with empathy, privacy mentions and next actions. Read these as indicators from the analysed conversations, with their own available data.
Churn risk
Read the average risk index out of ten and the number of high-risk calls. The reasons concern medium and high risk. This is an AI assessment to investigate, not a measured probability of losing a customer.
Upsell opportunities
The count and progress bar show calls marked as potential opportunities among those with an available upsell assessment. They identify conversations to review, rather than completed sales or revenue.
Competitor mentions
The bars count competitor names found in saved analyses. Use recurring mentions to understand the alternatives discussed by customers; the count does not establish market share or a lost sale.
Objections
The leading objections summarize what customers question or resist, with occurrence counts. Compare price, contract or service concerns and review examples before changing a script or an offer.
Recurring problems
Pain points collect difficulties and frustrations identified in conversations. Their recurrence helps prioritize a process or service review, while the recording provides the detail behind each issue.
Agent empathy
The average uses available empathy assessments on a one-to-five scale. Read it as a coaching signal and compare it with the conversation context, tone and words.
Privacy mentions
The percentage uses available yes/no assessments of privacy or terms being mentioned. It supports script review; it does not certify consent or legal compliance.
Calls with next actions
The indicator counts calls whose saved analysis contains at least one action item. Open the recording analysis to read the commitments and decide how to follow them up.
Recognize recurring subjects
The keyword cloud highlights frequent terms saved by the analyses. Larger words occur more often; use the leading topics to choose where to investigate, then check the summaries and recordings.
Turn the selected evidence into a report
Generate an additional AI reading of recent summaries from the selected period and search, with the statistical context of the usable analyses. The result contains recurring themes, their frequency labels, explanations and strategic advice.
Open an illustrative example in the product’s real report layout. It shows recurring themes, frequency labels, strategic advice and the size of the analysed sample.
In Obelisk, the report is generated on request from saved summaries and indicators. The configured sample limit and available evidence remain visible with the result.
Return to concrete conversations
The five highest and five lowest available scores provide examples for review, with call date, score and summary. Use these calls as coaching material and compare the analysis with the original recording when available.