An AI call summary is a short recap of a phone call, written by AI from the call transcript, so the agent doesn't have to type notes from memory. In Edesy's cloud contact center, each recorded call can get a transcript, a summary, key points, the caller's sentiment and a suggested disposition. The agent reviews it, confirms the disposition and moves on.
Try AI call insights: Open the Edesy contact center. Turn on recording, take a call and open its insights.
Why summaries matter
After every call, an agent has to record what happened: the reason for the call, what was promised, what happens next, and a disposition code. Doing that from memory is slow and uneven. Two agents describe the same call differently, notes get skipped when the queue is busy, and the next agent who picks up the customer starts from scratch.
An AI summary gives every call the same kind of record, written the same way, whether the queue was quiet or overflowing.
What you get for each call
| Output | What it is | Who uses it |
|---|---|---|
| Transcript | The full call as text | Supervisors reviewing a call; disputes |
| Summary | A few sentences on what the call was about and how it ended | The next agent; CRM notes |
| Key points | A short list of facts, requests and commitments | Follow-up tasks |
| Sentiment | The caller's overall tone | Spotting unhappy customers |
| Suggested disposition | The wrap-up code the call most likely fits | The agent, to confirm or change |
How it's produced
- The call is recorded. Recording can be turned on per number. Recordings are stored in encrypted cloud storage with signed links.
- After the call, the recording is transcribed. Edesy uses Deepgram for speech-to-text.
- The transcript is summarised. An AI model (Anthropic Claude) writes the summary, key points, sentiment and suggested disposition.
- The result is saved with the call. Opening it again doesn't regenerate it.
Insights are produced on demand when someone opens them. If the recording is still uploading, the insight shows as
not ready yet; open it again shortly. Developers can fetch the same data with GET /v1/ccaas/calls/:callSid/insights.
Using it in wrap-up
A simple routine for agents:
- The call ends and the agent goes into wrap-up.
- Open the call's insights.
- Read the summary and key points. Fix anything the AI got wrong in your own notes.
- Check the suggested disposition. Confirm it, or pick the right one.
- Submit the disposition. If your setup requires notes, add them.
For leads called through an agent-dialer campaign, the submitted disposition is also sent to your systems with
the disposition.submitted webhook, so the
outcome lands in your CRM or order system.
Using it as a supervisor
- Find the calls worth a listen. Negative sentiment or an unexpected disposition is a good reason to open the recording.
- Check script changes. Read summaries from the first calls after a change to see how customers respond.
- Coach with examples. Summaries and transcripts make one-on-one feedback specific.
For live coaching during the call, use listen and whisper; see listen, whisper and barge.
What a useful summary looks like
A good summary answers three questions in a few lines: why the customer called, what was agreed, and what happens next. For example: "Customer called about a delayed refund for order 58213. Agent confirmed the refund was issued on the 20th and shared the reference. Customer will check with their bank and call back if it hasn't arrived by Friday." If the AI summary is missing one of those three, the agent should add it in their notes.
Limits to be honest about
| Limit | What it means in practice |
|---|---|
| Needs a recording | Unrecorded calls get no insights |
| After the call, not live | It helps wrap-up and review, not mid-call prompts |
| Transcription isn't perfect | Names, numbers and mixed-language speech can be misheard; check key facts |
| The disposition is a suggestion | The agent confirms it |
| Not QA scoring | It doesn't grade the agent against a scorecard |
Indian calls often switch between English and Hindi or another language mid-sentence. Transcripts of those calls can miss words, so agents should check amounts, dates and names before relying on them.
Privacy
AI summaries are built from recordings, so they inherit the same care:
- Tell callers calls may be recorded, in whatever way your business has decided is right.
- Turn recording on only for numbers where you need it.
- Limit who on your team can open recordings and insights.
- Don't paste summaries containing personal data into tools that don't need it.
This isn't legal advice. If you handle sensitive data, take advice on your obligations, including under India's Digital Personal Data Protection Act. See DPDP and voice AI for a general overview.
Pricing
AI call insights are part of the Edesy cloud contact center. Seats are Rs 1,500 (Starter, up to 20 agents), Rs 2,500 (Professional, up to 100) or Rs 3,500 (Enterprise, unlimited) per agent per month. Call minutes are extra at Rs 0.50/min on a 60-second pulse. Prices exclude 18% GST.
| Plan | Per agent per month | Agents |
|---|---|---|
| Starter | Rs 1,500 | Up to 20 |
| Professional | Rs 2,500 | Up to 100 |
| Enterprise | Rs 3,500 | Unlimited |
See contact center pricing, the cloud contact center page and the contact center docs.