Call analysis
What the automatic analysis of calls and emails recognises, where you see the results and how you check them.
The call analysis evaluates calls and incoming emails automatically and records what came up in them: objections, questions, requests for an info email, problems with leads or the CRM, and more. These results are called findings. They end up in reports, on the dashboard and in the review queue.
The AI settings are under AI analyses. This page only describes what you see and can set as a user.
This applies to calls, emails and report texts. The analysis does not run in demo organisations.
When a call is analysed
A call is analysed as soon as its transcript is finished. The system skips:
- projects in which the analysis is switched off,
- calls that are shorter than the set minimum duration (default 30 seconds),
- calls without speech from the other party.
Older transcripts and emails from before the analysis started are not analysed retrospectively. Incoming emails that are assigned to a lead are checked regularly in the background.
The analysis knows the project briefing (see Reports), the relevant part of the findings catalogue and some context about the lead (company, stage, number of earlier calls, own fields without email and phone). Names of people are not part of it.
If a call cannot be analysed, it is skipped and remains pending. The report then shows a warning. You fetch it with Try again in the analysis status (see Reports).
Settings
Per project: in the project centre, open the section Automatic analysis on the Report settings tab (alternatively, for the project under Projects → Edit project → Edit reports). There are three settings:
| Setting | Effect |
|---|---|
| Analyse calls automatically | Switches the call analysis for the project on or off |
| Analyse emails automatically | Switches the email analysis for the project on or off |
| Do not analyse shorter calls (under … seconds) | Minimum duration, whole number from 0 to 3600 |
Only people who may change the report settings of the project can edit these. Without this permission, the fields are locked.
What is recognised
The recognised items are listed in a catalogue of types of feedback. The organisation starts with these types:
| Topic | Type | Source |
|---|---|---|
| Outcome | Outcome of the call (nobody reached, voicemail, gatekeeper, decision maker spoken to, appointment agreed, refusal, callback agreed, other) | Call |
| Objection | Objection or reason for refusal (no need, existing provider, exclusive contract, price, timing, scepticism, not the decision maker, other) | Call, email |
| Question | Question from the other party | Call, email |
| Interest | Info email requested, callback requested, interest (high, medium, low, none) | Call, email |
| Appointment | Appointment (feeds the result of the appointment recognition into the reports) | Call |
| Lead issue | Wrong number, wrong company, wrong contact person, company no longer exists, duplicate, already a customer, not in the target group, other | Call, email, block, day |
| CRM issue | Connection dropped, microphone, recording mixed up, wrongly blocked, system too slow, other | Call, block, day |
| Quote signal | Reaction to the offer (understands the benefit, benefit unclear, price objection, competitor named, wants details) | Call, email |
| Email class | Class of the reply (interest, question, refusal, automatic reply, absence, forwarded, other) | |
| Documentation check | Deviation between documentation and call (hard or soft contradiction) | Call |
| Employee observation | Opening, needs analysis, handling objections, booking appointments, adherence to the script, tone (each as a strength or to be improved) | Call |
Each finding has a value, often a detail; with the AI analysis there is a verbatim quote as evidence and a confidence in percent. The system checks the quote against the text. The model’s word alone is not enough.
The documentation check compares what an employee documented as the outcome with the recognised call outcome. From this, the system forms a hard or soft contradiction according to fixed rules, with both sides as evidence. The AI corrects nothing.
For each call, the system also calculates talk share, longest monologue and call duration from the timestamps of the transcript.
Types of feedback
Here you define what is recognised and who sees it in the report:
- Organisation: Settings → Sales → Reports → Types of feedback. Applies to all projects without their own setting.
- Project: Report settings tab → Types of feedback. The first change creates a copy of the organisation default that applies only to this project. Reset to default removes it again.
The types are sorted into groups by topic; ‘X of Y active’ shows how many are switched on. For each type:
- The switch turns the type on or off. Supplied types can be switched off but not deleted.
- Under Who sees this in the report? you choose the target groups: client, lead quality, CRM operator, internal (project), internal (employee) or the employee themselves. Without a selection, nobody sees the type in the report (‘Nobody’).
Fixed limits always apply: employee observations never go to clients, lead quality or the CRM operator; CRM issues never go to clients or lead quality; lead issues do not go to the CRM operator. The system shows an error if you choose otherwise.
With Add own type you create a new type: name, topic, answer type (‘Yes or no’, ‘Choice from values’ or ‘Free text’), for a choice the values (at least two, one line per value) and the sources (‘Where does the data come from?’: call, email, block, day). A new type is initially visible only to ‘Internal (project)’.
Review queue
Findings where the automatic analysis is uncertain (confidence below the threshold of the report configuration) end up in the Review queue tab of the project centre. A counter on the tab shows the open findings.
Filter by State and Type. Each entry shows type, topic, state (Automatic, Open, Confirmed, Rejected, Changed), value, detail, quote, confidence, lead, date and source. You can:
- Confirm: the finding counts in reports.
- Reject: the finding counts nowhere.
- Edit: correct value and detail; the finding is then marked as ‘Changed’.
Whether open findings still go into a report depends on the setting ‘Also include unreviewed feedback’ in the report configuration. Whether a finding appears in a report depends on the type being active in the project’s effective catalogue, the target group of the report type being set, the finding not being rejected, and it reaching the minimum confidence or having been confirmed.
Where you see the results
| Place | What you see |
|---|---|
| Reports | Building blocks ‘Feedback and anomalies’ (overview and list), observations in the weekly report, lead quality, CRM report, email evaluation, offer validation |
| Dashboard | Chart ‘Objections’ in the sales area (without rejected findings) |
| Project centre → Employees | Calls, talk share, deviations and strengths for each person, internal only |
| Overview → Recordings | Recordings with an overview (recordings, talk time, transcribed, own talk share), display of the transcript and talk share for each recording. See Recordings |
Statements about employees never reach the client, the lead supplier or the CRM operator.
Notes on quality
- The analysis is automated and can be wrong. Check important findings in the review queue before you approve a report.
- The quote in a finding is the passage in the transcript that the statement is based on. If there is no transcript, there is no analysis.
- How precisely names, figures and technical terms are recognised depends on the chosen transcription mode (see AI analyses).