Between a dial attempt and a conversation sits most of the work in outbound customer contact, and almost none of it is visible in a report that counts minutes. Cumulative to 31 October 2026, Vee's AI made 298,684,764 single dial attempts and held 38,325,799 connected conversations; the space between those two numbers is where reach is won or lost.

This article is about that space. It is not about a ratio. We deliberately do not publish one, and we explain why below. It is about what actually happens to the attempts that do not turn into a conversation, what each of those outcomes tells you, and which levers move them. Everything here comes from running the service ourselves: Vee employs no call agents, so every one of those attempts was placed, timed, retried and classified by our own platform, and every pattern we describe is one our team has had to answer for.

Two numbers, two different units#

298,684,764Call attemptsSingle dial attempts placed by Vee's platform. One attempt is one dial, whatever its outcome.Cumulative to 31 October 2026, Vee operating data.
38,325,799ConversationsConnected conversations held by Vee's AI.Cumulative to 31 October 2026, Vee operating data. Vee employs no call agents.

An attempt and a conversation are different units, and they sit alongside two others we report: cases and minutes. A case is a record assigned to us for handling. One customer account can produce several cases in a month, because a bank may assign the same loan contract again after a broken promise or a new missed instalment. A minute is time spent in conversation. None of these is a person, and none of them can be divided by another to produce a performance figure without a great deal of context.

The four units behind Vee's operating aggregates
UnitWhat it countsWhat it is not
AttemptOne dial, whatever happened nextA contact, a person or a case
ConversationA connected AI conversationA right-party contact or a result
CaseA record assigned for handlingA person, an account or a call
MinuteAI conversation timeEffort, value or outcome

Cases 48,626,959; minutes 28,569,847; conversations 38,325,799; attempts 298,684,764. All cumulative to 31 October 2026, Vee operating data.

That last point matters more than it sounds. A fresh lead that filled in a form ten minutes ago and a record that has been passed between several lenders are both one case. The first may connect on the first or second attempt. The second may take many attempts and still produce nothing worth having. A blended ratio across both tells you about the blend. What is useful is the anatomy of the gap, and that anatomy is remarkably consistent across sectors.

What happens to an attempt that does not become a conversation#

Every attempt ends with exactly one technical status. It is the most objective thing we record about a call, because it comes from the telephone network and the platform rather than from anyone's interpretation. The outcomes are few, but each one means something different, and treating them as one bucket called “no contact” is, in our view, a common and costly first mistake.

Common outcomes of an attempt and what they tell you
OutcomeWhat it usually meansWhat it should trigger
No answerThe phone rang; nobody picked upA further attempt, within the agreed limits
BusyThe line is alive and someone is on itA further attempt, handled separately from no answer
VoicemailPhone off, call declined, ignored or divertedIts own handling; sometimes another channel
Cut offConnection dropped, by the network or by a personDifferent handling depending on who ended it
Failed or unavailableNumber not reachable on the networkData review rather than blind retries
Another person answersReached a household or shared lineClose without disclosure, retry later
Wrong numberThe number does not belong to the customerDiscard and flag for data review

Typical classification used in Vee's outbound processes; exact rules are configured per client and market.

Two of these deserve more attention than they get. “Another person answers” is a connected call, sometimes a long one, and it is still not reach in any useful sense. In early collections, the AI must close that call politely without disclosing anything about the debt, because debt details are never disclosed before identity is verified. That rule is enforced by the platform, not left to a script. The case goes back into the queue. A report that counts that call as a conversation is technically right and commercially misleading.

“Wrong number” is the other. It is not a contact failure at all. It is a data finding, and its value lies in going back to the client so that the record can be corrected. Dialling it again tomorrow is waste. Dialling it again for a year, which happens more often than anyone admits in organisations without case-level statuses, is waste that also irritates a stranger.

The first fraction of a second#

When a call is picked up, there is a fraction of a second in which the platform has to decide what has answered it. It could be a person. It could also be a voicemail box, a smartphone's call-screening assistant asking the caller to state their business, or another machine. Internally we call this moment the superposition, because until it is resolved the call is several things at once, and it has to be resolved with a level of confidence attached.

Both kinds of error are expensive. If the platform takes a mailbox for a person, it starts a conversation with a recording and logs a contact that never happened. If it takes a person for a mailbox, it hangs up on a customer who has just answered, which is the worst possible first impression. Call screening has made this harder over the last few years, because the first voice on the line is now often a machine acting on behalf of a real person who may well pick up if the caller handles the screen correctly.

This is one reason we built and run our own telephony layer rather than renting a generic one, and why voicemail detection is a quality topic for us rather than a setting. Our quality team listens to selected calls every day in every project, and unusual endings are exactly the calls they pick.

Every day, in every project, our people listen to selected calls, especially unusual ones, to hear with their own ears what technology still cannot. – Dawid Wójcicki, Founder & CEO of Vee

Numbers: the line you call from and the line you call to#

Two different telephone numbers are involved in every attempt, and both shape whether it connects.

The outbound number#

The number that appears on the customer's screen carries a reputation. People decide in a second whether to pick up, and they decide partly on whether the number looks familiar, local, legitimate or already labelled as unwanted by their handset or network. Number presentation and spam labelling are part of our daily operating work, not an occasional incident. Rotating numbers helps where a customer has learned to ignore one of them, but rotation on its own is not a contact strategy.

The destination number#

Customers often have several numbers on file. We work them in rotation, and over time the records show which number actually gets answered. Something less obvious shows up as well: the number that answers most is not always the number that produces the result. A household landline may be picked up readily by someone who is not the customer, while the customer's own mobile answers less often but leads to a verified conversation and a commitment. Optimising for pickup alone would steer you towards the wrong line.

Number quality is also where our own history began. The first outbound campaign Vee designed, tested with a single call centre between 2016 and 2018, started by verifying numbers before dialling. It was an unglamorous step, and it was one of the reasons the campaign worked.

Time: windows, preferences and a base that cools#

When you call decides a large part of whether anyone answers. There are three layers to it.

  • Permitted windows. Each market has its own calling hours and contact rules. We configure them per client and per market, and the platform enforces them. Nobody can push a campaign outside the window to catch up on volume.
  • Individual timing. Within the window, people answer at different times. When a customer tells us to call after five, that is a signal worth more than any average, and honouring it is part of the reachability strategy.
  • Speed relative to the event. A lead cools by the hour. A debtor in early delinquency slides into the next bucket day by day. An appointment confirmation is only useful while the slot can still be given to someone else.

The third layer is where automation changes the economics most. A call centre has a fixed number of agents and a base that fluctuates; on a heavy day, part of the assignment is simply not touched until it is too late. Our capacity is infrastructure, not headcount. In March 2020, when demand from large clients rose roughly fivefold, channel capacity at one mobile operator was raised from 250 to 900. A whole confirmation campaign for the coming days can be completed within hours, with repeat attempts for everyone who did not answer.

Persistence, and where it must stop#

Most of the distance between attempts and conversations is covered by repeat attempts. This is where people and machines differ most sharply. A clinic receptionist will not make three or more attempts to reach a patient who does not pick up; there is a queue at the desk. A sales team's dialler stops when the agents are busy with the interested customers. Vee keeps calling, because it has the lines and it does not tire. In lead qualification, that persistence is often the main reason a base yields more interested customers than it did with a traditional dialler.

Persistence has a ceiling, and the ceiling is a design decision, not an accident. Each market has a maximum number of contact attempts, and we configure and enforce it per market. Beyond the legal limits there is a cultural one: a brand that calls too often stops being answered, and the customer who finally picks up starts the conversation annoyed, which shows up immediately in the next rung of the ladder, continuity. We tune intensity on evidence. Whether a second attempt on the same day raises the chance of an answer, for this base and this segment, is a question we test in production rather than a rule we copy from the last project.

The other half of persistence is spacing. When the next attempt is made should depend on how the last one ended. That subject has its own article in this series, because treating every outcome the same way is, in our experience, a costly simplification in outbound contact.

The base sets the ceiling#

No configuration overcomes the quality of the base. We say this to clients early, usually in the first diagnostic meeting, because it determines what a realistic result looks like. A fresh lead that has just asked for a loan answers quickly and wants to talk. A base of existing customers with a known history answers in patterns we can learn. A base that has been worked over by several lenders before it reaches our client behaves differently: the people who are still interested there are more likely to be people others refused than people others failed to reach.

Knowing which base you are calling does not make it better, but it changes how you run it and how you measure it. The attempts needed per conversation, the right intensity, the voice that keeps people on the line and the definition of success all differ by base type. Comparing reach across them, or across suppliers working on different bases, produces numbers that look precise and mean little.

The same gap looks different in every process#

In early collections, the gap is dominated by verification. Reaching a phone is not enough; the conversation counts only once the right person has confirmed who they are, and a household line can connect many times without producing a single right-party contact. The schedule is shaped by the bucket: the earlier the delinquency, the more the speed of the first verified contact matters.

In lead qualification, the gap is dominated by time. A person who asked for a quick loan is often comparing offers and will talk to whoever calls first. Attempts made in the first minutes after the form is submitted are worth far more than the same number of attempts made the next day, and the voice matters too: in cold contact, studio-recorded voices often keep people on the line where synthesis loses them.

In appointment confirmation, the gap is dominated by the deadline. The goal is to learn early that a visit will not happen, so the slot can be offered to someone else. The contact plan is built backwards from that deadline, with another channel stepping in when calls fail. An attempt that connects after the slot could have been refilled has lost most of its value.

In satisfaction surveys, the gap is dominated by willingness. People can be reached easily and still decline. Here, reach is rarely the bottleneck; consistency is. The thirty-first call has to sound exactly like the first.

Why the gap is invisible in a minute-based report#

Here is the commercial consequence. Almost everything described above, the number checks, the windows, the retries, the detection of mailboxes and screens, the rotation across a customer's numbers, generates attempts and very few minutes. In a base with low reach and frequent hang-ups, a very large amount of work produces a small number of conversation minutes. A supplier paid by the minute is paid nothing for the work that made the conversations possible, and is rewarded for making the conversations it does get longer.

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

This is why new Vee agreements are no longer billed by the minute, and why our public reporting moved in the same direction. Since ESPI report 11/2026 of 15 September 2026, we no longer publish case volumes and conversation minutes as regular metrics. The work between an attempt and a conversation is real work. The value it creates is visible only further up the ladder: in verified contacts, commitments and, eventually, payments, visits and contracts.

There is one more reason not to treat unconnected attempts as wasted. Contact itself has value. Some customers pay because their bank called, even when they never picked up, and an honest measurement looks for results among cases where nobody answered, not only among those where a conversation took place.

What to ask about the gap in your own operation#

Whether you run outbound contact in-house, through a call centre or through a provider like us, the gap between attempts and conversations is measurable and manageable. These are the questions we would ask first.

  1. Do you record one technical status per attempt, and can you separate no answer, busy, voicemail, cut off, unreachable and wrong number?
  2. Do wrong-number and unreachable outcomes flow back into your customer data, or are those records dialled again next month?
  3. Which of each customer's numbers produces verified conversations, as opposed to pickups?
  4. Are calling windows and maximum attempts enforced by the system, per market, or by a scheduler someone can override under volume pressure?
  5. Does each outcome have its own retry interval, and when were those intervals last tested?
  6. How is a voicemail or a call-screening assistant told apart from a person, and who listens to the cases where the platform got it wrong?
  7. Are you measuring reach separately for each base type, rather than as one blended figure?
  8. Does your billing model pay for the work that produces conversations, or only for the length of the ones that happen?

None of these questions needs a new platform to answer. They need case-level data and someone whose job is to read it every day. That is the part of the service we have run ourselves since production at scale began in 2017, and it is the part we would look at first in any diagnostic.

Follow The First Second

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

Follow via RSS