Most return-on-investment arguments for AI calling compare the per-minute rate against a telecaller's salary. That comparison flatters the technology and misses the actual reason it pays: the calls it makes are calls nobody was making at all.
Below is the arithmetic per industry, using real rates — ₹4–5 per call-minute for two-way conversation, ₹1.3–1.5 for one-way announcements — and honest call lengths.
The general shape
Every case in this article has the same structure:
- A call costs ₹2 to ₹12 depending on length and type.
- It prevents a failure worth ₹60 to several thousand.
- It succeeds some fraction of the time — usually 20% to 50%.
So the question is never whether one call pays back. It is whether the failure it prevents is expensive enough that a 30% success rate still clears the cost. For most operational calls it clears it by an order of magnitude.
Worked numbers
| Industry | Call | Length | Cost | Failure prevented |
|---|---|---|---|---|
| E-commerce | COD confirmation | 40s | ~₹3.50 | RTO: shipping both ways plus handling |
| E-commerce | Cart recovery | 90s | ~₹7 | Margin on the basket |
| Healthcare | Appointment reminder | 50s | ~₹4 | ₹500–1,500 consultation slot |
| Real estate | Site visit confirm | 75s | ~₹6 | An executive's Saturday |
| Real estate | Lead qualification | 2 min | ~₹10 | Agent hours on unqualified leads |
| Education | Counselling follow-up | 90s | ~₹7 | A seat worth ₹30,000+ |
| Lending | Pre-due EMI reminder | 45s | ~₹4 | ₹40–90 collections call, ₹300+ field visit |
| SaaS | Trial activation | 2 min | ~₹10 | A full subscription |
| Logistics | Pickup confirmation | 30s | ~₹2.50 | ₹80–150 wasted rider trip |
| Logistics | Delivery reattempt | 60s | ~₹4 | ₹60–120 per attempt plus RTO risk |
| Insurance | Renewal reminder | 60s | ~₹5 | Whole policy plus re-acquisition |
| Automotive | Service reminder | 60s | ~₹5 | Years of service revenue |
| Local services | Appointment confirm | 25s | ~₹2 | A blocked chair or technician slot |
| B2B | Payment follow-up | 60s | ~₹5 | Working capital on the invoice |
Where the payback is obvious
Anywhere the prevented failure costs more than about ₹100, and where the success rate is materially above zero.
Healthcare appointment reminders are the clearest case in the table. A ₹4 call protecting a ₹500–1,500 slot needs to work roughly once in every hundred and fifty attempts to break even. It works about a third of the time.
Logistics pickup confirmation is similar. ₹2.50 against a ₹80–150 wasted trip, and the call also catches parcel-count mismatches that would have failed anyway.
Pre-due EMI reminders pay back not through recovery but by shrinking the population that reaches a ₹40–90 collections call or a ₹300+ field visit.
Where it is marginal
Being straight about this is the point of the article.
- Low-value cart recovery. Below about ₹500 of basket value, the margin at stake does not justify a call. Use email.
- Any call under 200 a month. The setup and prompt-tuning cost more than the calling. Below that volume, a person with a phone is genuinely the right answer.
- Announcements to a list you did not build. Poor response and a compliance problem.
- Anything requiring negotiation. Not a cost question — the call simply cannot do the job.
The cost that is not per-minute
Three real costs sit outside the rate card and get left out of most estimates:
Prompt iteration. The first version of a script is never the last. Expect to read fifty transcripts and rewrite twice. That is a few hours of a good person's attention, and skipping it is the main reason projects underperform.
Integration. Outcomes have to reach the system where the work happens. A confirmed appointment that nobody wrote back to the scheduling system is a call you paid for and did not use. Getting outcomes into your system is not optional.
The number itself. A dedicated caller ID that customers can recognise and call back is worth having. Why that matters more than it sounds.
A sensible way to size the first month
Take your single highest-volume operational call. Multiply monthly volume by the per-call cost from the table. That figure is almost always smaller than expected — a clinic calling 1,200 appointments a month spends about ₹4,800, which is less than a part-time receptionist.
Then ask what fraction of the failures it needs to prevent to cover itself. For most operational calls in the table, the answer is somewhere between one and five percent, and realistic performance is many times that.
If that ratio does not work for your call, the honest conclusion is that this particular call should not be automated — which is a useful thing to learn before spending anything. The four-question test gets you there faster, and the pricing page walkthrough covers plans and wallets.