Every tele-CRM in India is shipping “AI call summaries” right now. Most of them ship the same model. What’s wildly different is who the summary is for. That choice — invisible from the marketing site — is the difference between a feature your reps quietly love and one your reps quietly resent.
We’ve spent the last six months in real rooms with real reps, watching them use four different AI-summary products. Three patterns showed up. Two are good. One is the failure mode.
Pattern 1: AI types the note while the rep talks
The good version. Call ends; a structured note shows up in the rep’s app within 15 seconds. Outcome, next step, objection raised, any commitments. The rep opens it, edits two lines, hits Save. The note becomes the source of truth.
The rep saves ~90 seconds per call. Multiply by 50 calls a day. Multiply by 250 working days. That’s ~9 weeks of typing per rep per year that the AI is doing.
Pattern 2: AI suggests the next step
Also good. The summary includes a one-line “next step” suggestion: “Send the BHK floor plan; call back Friday morning.” The rep approves or edits. The system creates the reminder.
What makes this work: the suggestion is just a suggestion. The rep is the deciding party. The reminder doesn’t fire until the rep has confirmed. The AI never sends anything on behalf of the rep without consent.
Pattern 3 (the failure mode): AI scores the rep for the manager
Now we’re in surveillance territory. The same call summary, the same sentiment model — but the artifact lands first on the manager’s dashboard. “Rohan’s last call: sentiment 6.2/10. Customer urgency: low.” The rep finds out by being asked about it.
Three things break:
- Trust. The rep stops being candid in notes; performance becomes performative.
- Accuracy. The model is good enough to be useful, not good enough to be a verdict. Used as a verdict, it amplifies false signal.
- Adoption. Within three weeks, reps find ways to game it — short calls, scripted closes, “everything sounded great.”
The line between helpful AI and surveillance AI is one question: who sees the output first, the worker or the manager? Get this wrong and you’ve built a worse product even if the model is the same.
A test for any “AI for sales” feature
Ask: “If we ship this, will my best reps want to use it, or will they want to opt out?” If the honest answer is “opt out,” the feature is structured wrong even if the model is good.
Where this lands at GrowYu
Every AI feature at GrowYu lands in the rep’s app first. No one else sees a call summary until the rep has at least opened it. Managers get aggregates, never per-call sentiment. The rep can override or delete a summary. The system records who edited what.
This isn’t subtle. It’s an architectural decision and we wrote it into our principles. Our AI principles, full version →
Five questions for any vendor’s AI
- Who sees the AI summary first?
- Can the rep edit or delete the summary?
- Does the manager dashboard show per-call sentiment, or only aggregates?
- Can the AI send a message to a customer without the rep’s explicit click?
- Is the audit log available to the rep, not just admin?
If you can ask those five questions and get clear answers, you’ve found a vendor who has thought about this. If the answers are vague — “well, it depends on your settings” — that’s the answer.