AI speech analytics turns your recorded sales and service calls into searchable transcripts, short summaries and structured fields on the contact, so you can see what was said without listening to hours of audio. On the levelpartner.ai platform the calls already sit inside Conversations, which means the analysis lands on the same record as the lead's source, pipeline stage and bookings. This page explains what you get, how to switch it on and how to use it without creating a compliance problem.
What the platform does with a call
When call recording is enabled, each inbound and outbound call made through a platform number is recorded and attached to the contact's conversation. Speech analytics AI then works on that recording in three layers. It transcribes the audio into text. It summarises the call into a few lines covering the reason for the call, the outcome and any next step. And, through a workflow, it can write specific answers into contact fields, such as the budget mentioned or whether a meeting was agreed.
Calls handled by an AI voice agent are transcribed and summarised by default, because the agent needs the text to do its job. The same reporting applies to both.
Before you start
- Phone numbers on the platform for the lines you want analysed, with calls routed through them.
- Call recording switched on in the phone settings for your account.
- A recording notice played at the start of calls, worded to meet the consent rules in each country, and each US state, you call into.
- A short list of facts you want extracted from every call, such as budget, timeline and decision maker.
- Custom contact fields created for each of those facts.
How to turn it on
- Confirm call recording is on for each number you will analyse.
- Add a recorded greeting telling callers the call is recorded.
- Go to Settings, then Phone System, and open the Voice tab.
- Open Call Transcription and check Enable Call Transcription.
- Place a test call, then find the contact in Conversations.
- Click View Transcript under the call's playback controls.
- Go to Settings, then Custom Fields, and create fields for budget and timeline.
- Go to Automation, then Workflows, and click + Create Workflow.
- Add the Transcript Generated trigger, set Call Type, then Save Trigger.
- Add AI Extract Data with the full transcript in Extract From.
- Add Update Contact Field and insert each extracted value into its field.
- Add an If/Else branch that tags the contact when a value is empty.
- Switch the workflow from Draft to Publish.
- Add Incoming Calls by Status to your sales dashboard.
- Review ten transcripts against the recordings before managers rely on them.
What it is for in a 60-person firm
When revenue is flat and the team is busy, the sales manager usually has no idea what happens on calls. Coaching relies on the rep's own account of the conversation, and the CRM notes are written from memory, if at all. Speech analytics replaces those notes with an accurate record and replaces random call sampling with targeted review.
The practical win is in qualification. If every discovery call writes budget and timeline to the contact, your pipeline report can filter out deals that were never real. That alone tends to make forecasts more honest within a month. For the dashboard that shows those fields, see laying out a sales and marketing dashboard.
Suppose advisers spend about four minutes writing notes after each call. At 600 calls a month that is 2,400 minutes, or 40 hours across the team. If transcripts and an AI step fill loan amount, timeline and property status, most of that time becomes a quick check of the summary. And if 30 calls a month are tagged for review, the manager would read those rather than sampling at random. The numbers are hypothetical.
Weekly time to run it
After the build, which takes a day, the sales manager should spend about an hour a week. Half of that goes on calls tagged for review and half on reading a sample of AI-filled fields to check the extraction is still accurate. Expect to rewrite the AI instruction once or twice in the first month as you learn which phrases it misreads.
Firms switch on recording and extraction across every line at once, including HR, finance and personal lines. You end up storing conversations you had no reason to record. Start with sales and enquiry lines only, and exclude any number used for internal or sensitive calls.
When not to use it
Do not use it where callers discuss health, financial hardship or other special category data unless your data protection officer has signed off the lawful basis and retention period. And if most of your sales conversations happen on video calls outside the platform, call analytics here will only see a small slice of the picture, so start by moving first-contact calls onto platform numbers.
Troubleshooting
No transcript appears after a call. Recording is off for that number or the call was too short to transcribe. Check the number's settings and test with a call of at least thirty seconds.
The summary misses the key point. Long calls with several topics confuse a generic summary. In AI Extract Data, define exactly which three things to capture, each with a short Description.
Fields are filled with guesses. The instruction allows the model to infer. Add a rule that it must leave the field blank unless the caller stated the answer directly.
Reps object to being recorded. Explain that the main use is removing note-taking, share the review criteria openly and keep coaching sessions private. Resistance usually fades when reps see the admin time they get back.
Frequently asked questions
Is speech analytics AI legal?
Recording business calls is generally permitted in the UK, the US and most other markets with the right notice, but rules vary by country and by US state, and some require every party to consent. Play a clear notice, record the lawful basis, and set a retention period. Take legal advice for regulated conversations.
Does it work on calls made from personal mobiles?
No. Only calls routed through platform numbers are recorded and analysed. If reps use personal phones, give them the platform's mobile app so calls go through the business number.
Can I search across all transcripts for a phrase?
You can search conversations, and you can use workflow conditions on extracted fields to group contacts. For trend reporting, extract the topic into a field and report on that field rather than searching raw text. For how calls fit the wider picture, read tracking the customer journey.
How is this different from the AI voice agent?
The voice agent holds the conversation. Speech analytics reviews conversations your people held. Both write to the same contact, so you can report on calls handled by either in one dashboard, as described in what a marketing KPI dashboard should show.