Per-lead pricing: the only model where your vendor wants what you want

Most voice AI is sold per minute. It is the easiest thing to quote and the easiest thing to compare, and it has one structural problem: your vendor gets paid whether or not the call accomplished anything.

A bot that talks to two hundred people and qualifies none of them bills exactly the same as one that qualifies forty. Every incentive in that arrangement points at minutes, and minutes are not what you are buying.

Per-lead pricing fixes that, and introduces a different problem that decides whether the deal works.

What per-lead actually changes

Under per-minute, the vendor optimises for the thing they bill. Under per-lead, they are suddenly interested in things you care about and previously had to police yourself:

• Connect rate. Dialling numbers nobody answers now costs them, so list quality and calling windows become their problem too.
• Answering-machine detection. A voicemail mistaken for a person is pure cost with no chance of an outcome.
• Conversation quality. A bot that annoys people into hanging up stops being free to them.
• Not padding calls. Under per-minute, a longer conversation is revenue. Under per-lead, it is cost.

That last one is worth sitting with. Per-minute pricing quietly rewards a chattier bot. You will not see it in any proposal, and it is not anyone behaving badly — it is just where the gradient points.

The definition problem

The moment money attaches to the word “qualified”, that word is under pressure. Not through bad faith — through ordinary ambiguity resolved, on each side, in the direction of its own interest.

Questions that look pedantic until the first invoice:

• Does a caller who says “maybe, call me next month” count?
• If someone qualifies and then does not answer the follow-up, is that still a lead?
• If the same person is reached twice in a month, is that one lead or two?
• If your sales team judges a lead junk, who decides — and what is the evidence?
• If the bot mishears and records a qualification that did not happen, who bears it?
• Does a lead from a list you supplied badly still count?

None have obvious answers. All of them will occur. The difference between a per-lead deal that works and one that ends in a dispute is entirely whether they were answered before signing.

How to write the definition

A workable definition has four parts. Anything less and you are relying on goodwill during an argument about money.

1. Observable criteria, not judgement

Every condition should be checkable from the call record by someone who was not on the call. “Expressed genuine interest” is not a criterion. “Confirmed identity, stated income band, and accepted a callback slot” is — each of those is a specific thing that either happened or did not.

2. A named adjudicator and a time limit

Decide up front who rules on a disputed lead and how long you have to dispute it. A forty-eight or seventy-two hour window with a named person on each side resolves almost everything, because both parties still remember the context and the recording is to hand.

3. A rejection allowance, with a ceiling

You will reject some leads. Agree the mechanism and put a cap on it — say, up to a certain percentage rejected without argument, above which you both review the criteria together rather than the individual calls. This stops two failure modes at once: unlimited rejection makes the vendor’s revenue unpredictable, and zero rejection makes the definition meaningless.

4. Deduplication and attribution rules

Same person twice, same household, a number already in your pipeline, a lead that arrived from another channel the same week. Write the rule. It is dull, it takes twenty minutes, and it prevents the argument that actually ends these contracts.

When per-lead is the wrong shape

Per-lead is not automatically better. It fits badly when:

• The work is inbound service. There is no lead. Answering “where’s my order” well has no qualification event, so per-minute or per call is the honest model.
• The outcome is far downstream. If value only appears at a disbursal months later, a lead is a weak proxy and both sides will feel it.
• Your list quality is the constraint. If the data is poor, the vendor prices that risk in — and you pay for it anyway, less transparently.
• Volumes are small. Per-lead economics need enough leads for the average to mean something. On a few hundred calls a month, variance dominates.

The comparison that actually matters

Whatever the model, convert everything to one number before comparing: total monthly spend divided by qualified outcomes. Not per minute, not per call, not the headline rate.

A per-minute vendor at a low rate with a 12% qualification rate can easily be more expensive per outcome than a per-lead vendor whose price looks alarming. The only way to see that is to do the division — and to insist that both quotes include every layer, since a per-minute quote usually covers one layer of six.

Related

Pricing model is recorded per vendor in this directory, and the matcher treats per-lead as a positive for outbound work while flagging the definition risk in the explanation it shows you. The cost calculator converts any model to cost per conversation so the comparison is like for like, and the hard facts page explains why headline rates understate.