Most arguments about contact-centre performance are not about performance at all. They are about units: the bank counts accounts, the supplier counts cases, the dialler counts attempts, and each side's reach rate is correct in its own unit and wrong in the other's.

We learned to settle units before anything else, because a process measured in mixed units cannot be improved, priced or audited. This piece sets out the units we use, the four kinds of status we attach to them, and how to read large cumulative figures, including our own, without misreading them.

Four units, four different questions#

The units of telephone customer contact
UnitWhat it isWhat it answers
AccountThe customer's relationship with our client, for example one loan contract or one policyHow many relationships did the process touch?
CaseOne assignment of a record to us for handling; the same account can produce several casesHow much work was assigned, and what happened to each piece?
AttemptA single dial attemptHow much effort did reaching people take?
ConversationA connected voice interactionHow often did we actually speak?

Definitions as used in Vee operating data. A case is a record assigned for handling, not a person or a call.

The distinctions matter because the ratios between units vary enormously by process. In early collections, a bank's analysts may assign the same loan contract several times in a month as its delinquency status changes, so one account becomes several cases. Each case may need many attempts before anyone answers, and some cases end without a conversation because the customer paid before we reached them. In an inbound service line, by contrast, a case usually begins with a conversation and the attempt unit barely exists.

One loan contract, one month#

A typical sequence in early collections shows how the units separate. A customer misses an instalment. The bank's analysts, following their own allocation strategy, assign the contract to us: one account, one case. We call within the market's permitted window. The first attempts meet voicemail and a busy line. On a later attempt the customer answers, passes identity verification and gives a date. That case closes with a promise to pay: several attempts, one conversation, one case.

Two weeks later the promised date passes without payment, and the bank assigns the same contract again. Now there are two cases on one account. The second conversation takes the customer's history into account and ends with a new commitment. If the customer pays before a third assignment, the bank's report shows one account cured. Ours shows two cases handled, a set of attempts, two conversations and a few minutes of talk. Every one of those numbers is correct, and only one of them is what the bank is paying for.

Some cases end without any conversation at all. A customer sees a missed call from their bank and pays the same afternoon. In a minute-based report that case is invisible; in a case-based report with a business status from the bank's systems, it appears as a result. This is one of the clearest arguments for counting cases rather than talk time.

Reading 48.6 million cases correctly#

Our cumulative operating figures, to 31 October 2026, illustrate why units need their scope attached.

48,626,959CasesRecords assigned to Vee for handling, cumulative.Cumulative to 31 Oct 2026, Vee operating data. A case is a record assigned for handling, not a person or a call.
38,325,799ConversationsConnected AI conversations, cumulative.Cumulative to 31 Oct 2026, Vee operating data.
298,684,764Call attemptsSingle dial attempts, cumulative.Cumulative to 31 Oct 2026, Vee operating data.
28,569,847AI conversation minutesAI conversation time, cumulative.Cumulative to 31 Oct 2026, Vee operating data; AI conversation time only, human agents not included.

These are four different counts, and none of them is a number of people. A single customer of a bank may appear as several cases over a year, several dozen attempts and a handful of conversations. Equally, one case can involve several conversations, for example a callback the customer asked for. That is why we do not divide one of these aggregates by another and publish the result as a rate: the cumulative mix of processes, sectors and years behind each figure is different, and the ratio would describe nothing in particular.

What the figures do describe is scale of execution. Every one of those cases was executed by Vee's AI; Vee employs no call agents. That is a different claim from a contact centre quoting total interactions in which the automated share sits in a footnote.

Four kinds of status, never mixed#

Units say what was counted. Statuses say what happened. We attach four kinds of status to the work, and keeping them apart removes a second large class of disputes.

  • Technical status: one per attempt, and the most objective. Answered, no answer, busy, voicemail, network failure, call cut off by the customer.
  • Dialogue status: several per conversation. The objective facts of what was said: identity confirmed, payment date given, callback requested, person asked for a human.
  • Processing status: the life cycle of the case, from waiting for launch through scheduled retries to the end of handling.
  • Business status: the client's own version of the truth, in its own language. If a bank wants to see “promise to pay within the horizon” or “hardship, refer”, that is what it receives.

The first three are ours and are consistent across every process we run. The fourth is defined with each client, which makes it the most useful status for the client and the least comparable across projects. A “success” in one bank's taxonomy may be a partial outcome in another's. That is a deliberate choice: the client should not have to translate our vocabulary into its own before acting on a report.

Keeping the axes apart also makes every case explainable. When a compliance team or an auditor asks what happened to a particular customer, the answer is a sequence: each attempt with its technical status and time, each conversation with its dialogue statuses and the script version that was live, the processing steps, and the business status that went back to the client. In a regulated sector that reconstruction is not a nice-to-have. It is the condition for running the process at all.

The statuses are also the raw material of improvement. A rise in one technical status points to a number or time-window problem; a cluster of one dialogue status points to a sentence that needs rewriting. Without the separation, both would disappear into a single “unsuccessful” bucket, and nobody would know which lever to pull.

Where the arguments come from#

Most KPI disputes we have seen fall into a few patterns. Each is resolved by naming the unit and the status axis, not by debating the number.

Reach per attempt versus reach per case#

A supplier reports that a small fraction of attempts became conversations; the client concludes the process is failing. Yet per case, a clear majority may have been reached at least once. Both numbers are true. Only the per-case figure tells you whether the assigned work got done; the per-attempt figure tells you how much effort it took.

Cases versus accounts#

The client's collections report shows fewer accounts reached than the supplier's report shows cases reached. Usually nothing is wrong: accounts reassigned several times in the month appear once in one report and several times in the other. Agree up front which unit the KPI uses, and which unit the fee uses, because they need not be the same.

Dialogue status versus business status#

The supplier counts a promise to pay; the client's system shows no payment. The first is a dialogue status, the second a business outcome measured later in the client's books. A promise is not a payment, and a report that treats them as one unit will be wrong in both directions.

Conversations versus minutes#

A report shows the number of conversations rising and the minutes falling, and someone asks what went wrong. Often nothing did. Shorter conversations that reach the commitment quickly are frequently better ones. In banking, cases ending with a promise to pay averaged 116.5 seconds of handling time (historical). Minutes measure effort and duration; they say nothing about whether the case was resolved. Since 2026 our own minutes have been falling faster than our cases, for exactly this reason, together with a changing campaign mix and growing foreign processes.

Units shape incentives#

The unit a supplier is paid in shapes what it optimises. Paid per minute, it has no reason to keep conversations short. Paid per attempt, it has a reason to dial more than necessary. Paid per case, it has a reason to handle each assignment completely, including all the attempts, SMS messages and short conversations that reaching a person takes. Paid per account, it has a reason to close the relationship on the first assignment rather than see it return.

Each has trade-offs. A fee per case makes the monthly cost predictable regardless of how customers behave, but an account assigned several times a month may be paid several times; how reassigned accounts are treated is set out in the agreement. A fee per account aligns with the client's view of total cost but needs reliable account identifiers in the data exchange. We offer both for new agreements, and we no longer bill new agreements by the minute. The right choice depends on the process, and it should be made deliberately, not inherited from whatever the dialler happens to count.

What to agree before launch#

  1. Which unit each KPI uses, written next to its definition.
  2. Which unit the fee uses, and how reassigned accounts are treated.
  3. The full list of technical, dialogue and processing statuses, and which ones the client's systems will receive.
  4. The business statuses, in the client's language, and who owns their definitions.
  5. How case-level statuses flow back each day, and in what format.
  6. That no published rate will mix units in its numerator and denominator.

None of this is glamorous, and all of it saves weeks. When units and statuses are settled before the first call, the weekly meeting is about what to change next rather than whose spreadsheet is right. That is the meeting worth having.

Follow The First Second

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

Follow via RSS