Paying for automated customer contact by the minute is an anachronism: it pays the supplier for the length of conversations, while the client is paying for what those conversations achieve. The two goals pull in opposite directions often enough that we stopped offering the minute as the unit for new agreements.

This is an opinion piece, and we will argue the position plainly. We will also be clear about the trade-offs, because per-minute pricing has real virtues, and about what we replaced it with, because criticising a unit is easy and designing a better one is not. We will not discuss prices; the point is the structure, not the rate.

Where the minute came from#

The minute is a telephony unit. Carriers sold minutes, call centres sold agent hours that were ultimately measured in handled minutes, and when automated voice arrived, it was natural to sell it in the same currency. A minute of AI conversation could be compared directly with a minute of a human agent, and the saving was easy to show on a slide.

That comparison made sense when automation was bought as capacity: a cheaper way to handle the same calls. It is the natural unit for a platform model, in which the platform supplies the conversation; the client supplies everything around it, from the database and the dialling strategy to the integration, the reporting and the decision about what to change next. If all you buy is the conversation, paying for its duration is at least coherent.

But capacity is not what large organisations need from customer contact. They need payments before the debtor slides into the next bucket, kept appointments, qualified leads while they are still warm, renewals before the policy lapses. The minute measures none of these.

The role of an external partner is to deliver success, not capacity. – Dawid Wójcicki, Founder & CEO of Vee

Three ways the minute pulls the wrong way#

It rewards long conversations#

The most direct problem is the simplest. A supplier paid per minute earns more when calls run longer. Nobody has to act in bad faith for this to matter; it is enough that nothing in the contract rewards brevity. Yet in many processes the best conversation is a short one. In early collections the aim is a clear, respectful exchange that ends in a specific promise to pay. In banking, cases ending with a promise to pay averaged 116.5 seconds of handling time (historical). Every redesign that shortens that path without losing the commitment is good for the client and, under a per-minute contract, bad for the supplier's revenue.

We have redesigned dialogues many times to reach the point faster without losing the commitment. Under minute billing, each of those improvements would have cut our own invoice. That is not a sustainable incentive structure for either party.

It ignores the work that produces no minutes#

Most of the effort in outbound contact happens before anyone speaks: choosing numbers, scheduling attempts within each market's permitted windows, setting retry intervals by outcome, sending SMS before and after, verifying numbers, handling voicemail and call-screening services. In a database with low reach and frequent hang-ups after the company name, an enormous amount of that work produces very few billable minutes.

This creates a perverse effect. The hardest bases, the ones where skilled reach work matters most, are the least attractive to a supplier paid by the minute. The easy bases, where people answer and talk, are the most lucrative. A pricing unit that makes difficult work unprofitable will, over time, ensure that difficult work is done badly.

Appointment confirmation shows the same pattern from another angle. The business goal is to learn early that a visit will not happen, so the slot can go to another patient. That means calling early enough for the slot to be refilled and following up when the patient was not reached or was unsure. The valuable output is often a ten-second cancellation, or a confirmation obtained only after several attempts. A clinic network paying by the minute is paying for the least important part of that work and nothing for the part that fills its calendar.

It cannot see results that happen without a conversation#

Contact has value even when no conversation takes place. Some debtors pay simply because their bank called: they see the missed call, understand why, and settle the same day. In appointment processes, an SMS fallback after an unanswered call can produce a cancellation that frees a slot for another patient. These outcomes generate no minutes, so a per-minute model neither pays for them nor reports them. The client gets a result that its supplier's invoice and reporting cannot see.

If we are paid by the minute, we are not playing towards the same goal as our client. – Dawid Wójcicki, Founder & CEO of Vee

The case for the minute, fairly stated#

Per-minute pricing survives for good reasons, and any replacement has to answer them.

  • It is simple. Everyone understands it, and it needs no negotiation about what counts as success.
  • It is auditable. Call detail records are objective; nobody argues about whether a minute happened.
  • It scales with use. If the client sends fewer cases, it pays less, with no fixed commitment.
  • It is easy to compare. Procurement can line up three offers in one column.

All four are real. The last is also the trap. A low price per minute is the easiest number to compare and the least predictive of what a result will cost. Back in December 2022, in an interview with NowyMarketing, our CEO made the point with simple arithmetic: a solution that converts a small fraction of contacts at a low unit price can cost more than twice as much per success as one that converts many times as often at a higher unit price. Comparing unit prices without comparing effectiveness measures the wrong thing precisely.

The other virtues can be kept without the minute. A fee per case is as simple to understand and, if anything, more predictable, because it does not depend on how talkative customers are. Case-level statuses are as auditable as call detail records, and they say what happened, not only how long it took. A per-case or per-account fee scales with the volume the client assigns, exactly as minutes do. What is lost is only the easy column in the procurement spreadsheet, and that column was measuring the wrong thing.

What we use instead: three layers#

For new agreements we use a structure with three layers. Each answers a different question, and together they keep the virtues of simplicity and predictability while aligning our incentive with the client's.

The three-layer commercial model
LayerWhat it coversHow it behaves
SetupPreparing and launching the process; scope set out in the agreementOne-off
Base operationRunning the process; scope set out in the agreementA fee per case or per account, independent of outcome
Performance layerImprovement on agreed KPIs, or a verified result where it can be measured reliablyOptional; rises or falls with the agreed indicators

Structure only. Existing contracts may still be billed by the minute; new agreements are not.

Setup#

The setup layer pays for the work that makes a process ready to run, from design and integration to testing before launch; its exact scope is set out in the agreement. Keeping it separate makes the cost of starting visible and stops it being buried in a unit rate that the client then pays for years.

Base operation: per case or per account#

The base fee is paid for each case or each account, independent of the outcome. It covers running the process for that unit, as set out in the agreement. It is the layer that covers the cost of running the service, and it makes the monthly cost predictable regardless of how customers behave on the phone.

The choice between case and account is a genuine trade-off, and we offer both.

  • Per case: simple and predictable, aligned with how assignments flow from the client's systems. The drawback is that an account assigned several times in a month is paid several times.
  • Per account: aligned with the client's view of the total cost of handling a relationship, and it motivates us to close the matter on the first assignment. It requires reliable account identifiers in the data exchange and a clear rule for what counts as one account over a period.

In practice the right answer often sits between the two, and it depends on how the client's analysts allocate work. The point is that both units describe the work and its purpose. Neither depends on how long anyone talked.

Performance layer#

The optional third layer links part of our pay to progress the client can see. One form is a KPI incentive: the base invoice is adjusted by a multiplier that moves with changes in agreed indicators, typically one on each rung of the KPI ladder, from reach to continuity to key information to business outcome. Each indicator can raise or lower the multiplier. The effect is that we invest work in improvement without sending a separate bill for it, and the client sees the result in subsequent invoices. We are introducing a model of this kind with clients outside Poland.

Another form is a fee for a verified result: a payment received, a completed survey, a patient who attended. That works where the result can be measured reliably in the client's systems, and only there.

Why not “only for results”#

Outcome-only pricing sounds like the logical end point, and some buyers ask for it. We do not offer it as a universal model, for three reasons.

First, the base layer exists because running a process has a fixed cost: telephony, infrastructure, quality assurance, people who listen to calls every day. If all of that is paid only on success, the supplier has to price in the risk of a bad base or a change in the client's policy, and the client pays for that risk one way or another. Second, many outcomes depend on things no contact supplier controls. When banks tighten credit criteria, a confirmed interest in a loan may not become a loan. Paying the contact supplier only when the bank's risk policy says yes would make the supplier underwrite the bank's credit decisions. Third, outcome-only models tempt everyone to reward easy declarations: promises to pay that look good on the day and never turn into money.

So we say it precisely. Outcome-linked pay is an option where the result is measurable in the client's systems. It sits on top of a base fee that covers the work. It is not a promise that the client pays nothing when nothing happens.

Defining success before the first call#

A performance layer is only as good as its definitions. Before launch, every process needs written answers, set out in the agreement, to six questions:

  1. What counts as success: a promise, a payment, a payment of the promised amount, an attended visit, a fully completed survey.
  2. When it must happen.
  3. Measurement window: how long after the contact the result is counted.
  4. Source of truth: which system and which report decide.
  5. Attribution: how cases also worked by other channels or teams are treated, and what the comparator is.
  6. Disputes: how disagreements over data are raised and settled.

The comparator deserves emphasis. A process that collects from a large share of a bucket may be adding little if most of those customers would have paid anyway. Choosing a base that pays anyway inflates any success fee. Honest measurement compares against a comparator on the same buckets and portfolios. Our published early-collections result is framed that way: +7–13% on-time payment frequency, historical, versus a human-agent comparator. That is an increment, and it is the only kind of number a performance layer should pay on.

Moving from minutes without a cliff edge#

Changing the billing unit of a running process is not trivial, and we do not pretend otherwise. Existing contracts may still be billed by the minute, and some will stay that way until there is a good reason to change. The transition usually follows the delivery path rather than a negotiation calendar.

It starts with a controlled pilot on a sample of the database, never the whole of it. The pilot establishes the baseline on each rung: how many cases are reached, how many people stay past the opening, how much key information is obtained, and what happens in the client's systems afterwards. That baseline, measured against a comparator, is what a performance layer can later be built on. Without it, any KPI incentive is a guess.

The base fee is then set per case or per account so that it covers the cost of running the process at the intensity the client wants. Only once both sides trust the measurement does the performance layer come in, often on a single indicator first. It is slower than signing a minute rate, and it produces a contract that both sides can live with for years.

What changes in the relationship#

The unit of billing changes the conversation between client and supplier more than any clause. Under minute billing, the weekly meeting tends to be about volumes: how many minutes, how many lines, how much capacity next month. Under a per-case model with a performance layer, it is about the ladder: which rung moved, why, and what to change next. The supplier stops defending its invoice and starts arguing for the next improvement, because the improvement is what it is paid for.

It also clarifies responsibility. When something on our side needs correcting, a missed rule or a badly worded sentence, we correct it straight away. When the client wants a change, it is approved on the client's side and, in regulated set-ups, versioned. How changes are scoped and settled is set out in the agreement. Neither rule depends on minutes, and both are easier to keep when the commercial model rewards the same thing the client measures. We sit on the client's side of the table; the billing unit should put us there too.

The market is moving the same way#

The unit is shifting across the industry, not only at Vee. In January 2026 Gartner forecast that the cost per resolution of generative AI in customer service will exceed $3 by 2030, above the cost of many offshore human agents. It is a forecast and should be read as one, but it frames the economics around resolutions, not minutes or tokens. When automation is cheap to start and not free to run, the risk of paying for activity that does not resolve anything grows, and someone has to carry it.

Our own reporting followed the same logic. On 15 September 2026, in current report ESPI 11/2026, we ended regular publication of case volumes and conversation minutes, because in a business where billing increasingly depends on effectiveness, equal minutes no longer carry equal value. A supplier that still asks you to judge it by minutes is asking you to judge it by its capacity, not by your results.

What to ask before you sign#

  1. What exactly is the billing unit, and does it rise when conversations get longer?
  2. Is the work that produces no minutes, such as attempts, SMS and number strategy, paid for, and how?
  3. Is the base fee per case or per account, and how are reassigned accounts treated?
  4. Which KPIs, if any, move the supplier's pay, and can they move it down as well as up?
  5. How is success defined, measured, attributed and disputed?
  6. What is the comparator, and who chose the base it is measured on?

If the answers show a supplier whose revenue rises when your customers stay on the phone longer, the incentive is wrong, however good the technology. If they show a supplier paid for running the whole process, with part of its pay tied to the indicators you manage, you are buying what you actually need. That is the model we work to, and the reason we no longer put the minute at the centre of new agreements.

Follow The First Second

Occasional issues on the mechanics and economics of AI-powered customer contact. No noise, no vendor news.

Follow via RSS