AI voice agent ROI: the seven numbers to track before you scale up

Updated 2026-10-22 · 4 min read · By the Convo360 team

Start with a baseline, or ROI is guesswork

Before you switch an AI voice agent on, write down what the same calls cost and achieved over the previous four weeks. You need four numbers:

• how many calls came in or went out;
• how many were missed or abandoned;
• how many led to the outcome you care about, such as a booking, a payment or a qualified lead;
• what the people handling them cost per month.

If you do not track missed calls today, your phone system or telephony provider's call log usually shows them. Our article on the cost of missed calls explains how to value them.

The seven KPIs and how to calculate them

Keep the definitions fixed. If "resolved" means something different each week, the trend is meaningless.

You do not need all seven from day one. For an inbound agent, such as a receptionist or support line, start with missed calls recovered, containment and repeat-call rate. For an outbound agent, such as lead callbacks, renewal reminders or COD confirmation, start with answer rate, conversion and cost per resolved call. Add the rest once those are steady. Whatever you track, write the definition at the top of your report so everyone on the team reads the numbers the same way.

Working out the real cost of an AI call

With a bring-your-own-keys platform, the cost of a call is the sum of your own provider bills plus the platform fee:

• Telephony minutes from Exotel, Plivo, Twilio or your SIP provider;
• Speech to text and text to speech from your voice provider;
• Language model tokens from OpenAI, Claude or Gemini;
• The platform subscription, spread across the month's calls.

Take these from the provider dashboards, not from a price list, because real calls include retries, silence and short calls. Our breakdown of what an AI call costs per minute shows how the four parts add up, and the cost calculator gives a quick estimate.

Then add the human cost that remains: transferred calls, callbacks and the time someone spends reviewing transcripts. Leaving that out flatters the result.

A worked example with assumed numbers

The figures below are assumptions to show the method, not results from any business. Replace each one with your own.

A diagnostics lab in Nagpur handles 1,500 calls a month about report status and home sample bookings.

• Before: two staff on the phones at an assumed ₹40,000 per month in total. They missed 300 calls in peak hours.
• After: the agent handles all 1,500 calls. It resolves 1,050 (70% containment) and transfers 450. Repeat calls on the same issue within 7 days: 60, so 990 resolutions really stuck.
• Costs after: assumed provider bills of ₹9,000, the Pro plan at ₹2,999 ex-GST, and one staff member at ₹20,000 for transfers. Total about ₹32,000.

Cost per real resolution is roughly ₹32,000 ÷ 990, about ₹32. The bigger gain is often the 300 calls no longer missed. If even a fifth of those become home sample bookings, that revenue usually outweighs the cost saving.

Reading the numbers without fooling yourself

• High containment with high repeat calls means the agent is closing calls without solving them. Read those transcripts first.
• A rising transfer rate after a price change or new product usually means the knowledge base is out of date, not that the agent got worse.
• Low answer rates on outbound calls point to the number, timing or list quality. The new caller name display and the right TRAI number series both matter here.
• Conversion should be compared by source. A lead from a Google ad and a lead from an old list behave differently.

Look at the trend over four to six weeks. A single week, especially around a festival or a long weekend, can mislead.

Tracking these KPIs on Convo360

Most of the raw data is already captured when the agent runs on Convo360:

1. Analytics shows call volumes, outcomes and trends over time.
2. Calls stores every call with its full transcript, outcome and AI summary, so you can check whether "resolved" calls really were.
3. Usage & Spend shows what you have used across providers, which feeds your cost per resolved call.
4. Post-call actions can Save CRM lead or move a lead along the CRM pipeline, which gives you conversion.
5. Export from Calls to a sheet and add your baseline for the before-and-after view.

Because Convo360 is BYOK, the provider costs you see are the provider's own rates with no markup. See the analytics docs.

Step by step

1. Record a four-week baselineNote call volumes, missed calls, outcomes and staff cost for the four weeks before launch.
2. Fix your definitionsDecide what counts as resolved, transferred and converted, and keep those definitions unchanged.
3. Collect real costs from provider billsAdd telephony, speech, model tokens and the platform fee from your dashboards each month.
4. Check resolutions against repeat callsCount callers who call back within 7 days on the same issue and subtract them from resolved calls.
5. Review the trend every weekCompare four to six weeks of data against the baseline before deciding to scale up or change the agent.

Example conversation

Frequently asked questions

Related reading

Available across India — see WhatsApp & AI calling automation by state and city.