Skip to content

Applied AI

When an AI voice agent actually makes sense

Voice agents are having a moment, which means they are being sold to businesses that do not need one. Here is the honest test.

By TezCommerce · · 5 min read

We build AI voice agents and we turn down roughly half the enquiries we get for them. Not because the technology does not work (it works far better than most people expect), but because for a lot of businesses it solves a problem they do not have, at a cost they will not recover.

The test

A voice agent pays for itself when three things are true at once:

  1. You genuinely miss calls. Not occasionally: structurally. After hours, during peak, when the one person who answers the phone is with a customer.
  2. A missed call is worth real money. An appointment, a booking, a quote request. If a missed call means someone rings back later, the agent is a convenience rather than an investment.
  3. The conversation is bounded. Taking a name, a reason and a time slot is bounded. Open-ended technical consultation is not.

Clinics, salons, dental practices, service businesses with dispatch, restaurants taking reservations, property offices fielding listing enquiries: these clear the bar easily. A B2B consultancy fielding four calls a week does not, and we will tell them so.

What a good agent does

It answers on the first ring, always. It identifies itself as an automated assistant. We are firm about this, because the trust cost when a caller works it out mid-conversation is far higher than the small friction of saying so up front. It handles English and Urdu, and code-switching between them, because that is how people actually speak here.

It captures the caller's name and their reason for calling, checks a live calendar, offers real available slots, books one, and confirms by WhatsApp before the call ends. Then it sends you a transcript and a one-line summary.

What it does not do

It does not replace your front desk. It handles the calls your front desk was never going to reach: the 8pm enquiry, the third simultaneous caller, the Sunday. Businesses that deploy one expecting to reduce headcount are usually disappointed. Businesses that deploy one expecting to stop losing after-hours bookings are usually surprised by how much volume was going unanswered.

It also does not sell. It can qualify and it can book, but a genuinely consultative sale still needs a person. Trying to push an agent past that line produces exactly the frustrating experience that gives the category a bad reputation.

What it costs to run

Build is the larger number. Running it is telephony plus inference: for a typical clinic handling forty to sixty calls a day, that lands in the low thousands of rupees per month. Against a single recovered appointment, the running cost is trivial. Against the build cost, you want to be confident about point two of the test above.

How to decide in one afternoon

Pull your call logs for the last month. Count the calls that rang out, went to voicemail, or came in outside business hours. Multiply by the value of a booking and by a conservative conversion rate (a third is reasonable). If that number is larger than the build cost, you have your answer. If it is not, spend the money on something else and we will still take the call.

Start a project

Let's work together.

Tell us where you want the business to be in a year. We will tell you how we would get you there, what it would cost and how long it would take, in writing.