Your Voice AI Is Cheaper. Congratulations.

There is a particular kind of conversation you have when you work in voice AI.

“Your competitor charges less per minute.”

This is usually presented as the end of the conversation. We have both seen the number. Their number is smaller. Someone should probably tell engineering to stop whatever expensive thing they are doing.

And I understand it. A price per minute is wonderfully concrete. You can put it in a spreadsheet. You can multiply it by a million. You can send that spreadsheet to someone who has never used either product and get a confident purchasing decision back.

The trouble starts when somebody calls the number.

A demo is a remarkably forgiving environment. The person calling knows what the agent does. They ask a sensible question. They wait for the answer. Nobody is driving through a tunnel while arguing with their spouse about which account the booking is under.

Production callers have not read your demo script.

This is where some of the cheapest products fall apart. The customer signs, discovers what they actually bought, and churns within a quarter. I see this happen. The low price gets three months to look like a good decision.

Then somebody has to rebuild the integration.

Of course, poor quality is only one explanation for a suspiciously small number. Another is that the vendor has raised enough money to make arithmetic temporarily optional.

If an investor is paying part of your phone bill, enjoy it. But recognise what you are comparing. A company trying to cover its costs and a company buying market share are solving different equations. You cannot necessarily negotiate your way into having their balance sheet.

Which brings us to sales.

Technical people, myself included, would like the better product to explain itself. We built the feature. We documented the feature. Surely the customer will discover the feature, appreciate its significance, and adjust their procurement spreadsheet accordingly.

Recently, I saw a competitor announce a major feature we had already had for eight months.

Eight months is a substantial engineering lead. It is also commercially useless in a conversation where nobody knows about it.

You need salespeople who understand the product well enough to hold their ground. People who can explain what happens when a transfer fails, how an integration works, and which capabilities the customer will need six weeks after signing.

“Better quality” is easy to say. Walking a customer through the failure their cheaper option cannot handle takes actual knowledge.

There is also a perfectly reasonable way to offer a low starting price: make your money across the wider product.

The voice minutes get the customer through the door. Other capabilities create opportunities to expand the account. Different product lines have different costs and different margins. The whole business does not have to fit inside one per-minute charge.

But now we are back to the spreadsheet problem. Is that cheap minute the whole product? What else will this customer need? What will the bill look like once the system is doing useful work?

All of these businesses can put the same number on a pricing page. One has cut corners. One is spending venture capital. One has a sensible expansion model.

The number does not tell you which one you are buying.

Before you multiply it by a million, make a few phone calls.