Most outbound calls do not become conversations, and what an operation does with those calls decides a large part of its result. A customer who asked to be called back, a phone that went to voicemail, a line that was busy and a call that was cut off are four different situations. Treating them as one generic "not reached" is cheap to configure and expensive to live with.
This article is about why the difference matters, what it means for the customer on the other end, and which controls a buyer should expect from whoever runs the process. It is deliberately not a manual for retry timing. That part is tuned per process and per market, and it belongs in the specification you agree with your supplier, not in a blog post.
Every attempt ends with a fact#
Each dial attempt returns a technical status: answered, no answer, busy, voicemail, dropped, failed. It is the most objective thing a contact operation records about a call. It is not an interpretation or a guess at intent. It is simply what the network and the handset reported.
At scale, these statuses are most of what an outbound process produces. The figures below are cumulative across every process, sector and market we have run, so they describe a mix rather than anyone's performance. They do show where the volume sits.
The gap between the two is not wasted effort. It is where reachability is won or lost, and every attempt in it carries information. An operation that collapses those statuses into one bucket throws that information away and then wonders why reach stalls.
Four outcomes, four different messages#
A callback request is a promise from the customer#
When a customer says "call me later", they have been reached, they have engaged and they have told you when they can talk. That is the most valuable outcome short of a completed conversation. It is also a commitment on both sides. A callback that arrives late, or not at all, tells the customer the conversation did not matter, and it wastes the best signal the process produced. In our view, an agreed callback should be treated as an appointment that the system keeps, not as one more entry in a queue.
Voicemail says the phone is there and the person is not#
A mailbox can mean a phone switched off, a call declined, a call ignored or a call diverted, and increasingly a smartphone assistant screening the call. What it rarely means is that the person is waiting for you to try again straight away. There is also a precondition: the platform must recognise the mailbox correctly. Mistake a recording for a person and you speak to a machine; mistake a person for a recording and you hang up on a customer.
Busy says the line is alive#
A busy signal is, in a sense, good news. The number works and someone is using it. It says something quite different from a phone that rings out, and an operation that cannot tell the two apart is guessing where it should be reading.
A cut-off call can mean two opposite things#
If the network dropped the call mid-conversation, the customer was engaged and may well welcome being reconnected. If the customer ended the call, especially in the first seconds after hearing the company's name, they have made a choice, and the response should respect it. The platform needs to distinguish a connection ended by the person from one ended by the system. Without that distinction, the most sensitive outcome of the four is handled blind.
| Outcome | What it tells you | Risk if handled like the others |
|---|---|---|
| Callback requested | The customer engaged and chose when to talk | A broken promise; the best signal in the process is lost |
| Voicemail | The phone is present; the person is not available | Repeated, pointless attempts that irritate and add nothing |
| Busy | The line works and someone is near it | A reachable customer treated as unreachable |
| Cut off by the network | A conversation was interrupted | An engaged customer left without an answer |
| Cut off by the customer | The customer chose to end the call | Pressure that turns a refusal into a complaint |
Descriptive only. The response to each outcome is set per process and market in the agreed specification.
What it costs to treat them alike#
A single generic retry rule is attractive because it is easy to explain. Its costs are spread across four places, and none of them shows up on the dialler's own dashboard.
- The customer. People notice when they are called without regard to what they said last time. A customer who asked for a callback and got a random redial, or who hung up and was rung again, is less likely to stay on the line when they finally answer.
- The base. Attempts spent at the wrong moments are attempts not spent on cases that could have been reached. A base also cools: in early collections and in lead handling, a contact that comes too late is worth less than one that comes on time.
- The number. Handsets and networks increasingly label numbers as unwanted. Calling patterns that customers experience as harassment feed that labelling, and a damaged number hurts every later campaign that uses it.
- The regulator. Contact limits, permitted hours, consent and AI disclosure are not preferences. A retry after a customer has asked to stop is not a reach problem. It is a conduct problem, and in regulated sectors it costs far more than the call.
Without reaching the customer, there is nothing else. – Dawid Wójcicki, Founder & CEO of Vee
We agree, and we would add one qualification. Reach bought at the cost of the customer's patience, or of the number's reputation, is borrowed from the next conversation.
Limits the system enforces, not the script#
Outcome rules decide what happens next. They operate inside limits that decide whether anything may happen at all. Those limits differ between markets and between clients, and in our view they should never depend on someone remembering them on a busy day.
- Contact caps. A ceiling on how often a person may be contacted, configured per market and per client.
- Permitted hours and consent. The windows in which contact is allowed, and how consent, including consent to recording, is captured and withdrawn.
- Agreed callbacks. A callback the customer asked for is kept by the system, not left to chance.
- Stopping when asked. When a customer says stop, the case leaves the contact cycle and the reason is recorded.
- AI disclosure. We always introduce ourselves as AI. The customer is never left wondering who is calling.
This is a different operating model from one in which limits are a policy that supervisors are expected to apply. Policies bend under pressure. Platform rules do not, and that is what makes them auditable. The transparency duties in Article 50 of the AI Act have applied since 2 August 2026, and the European Commission published guidelines on them on 20 July 2026. We help clients assess their own configuration against them; disclosure itself is not something we leave to configuration.
Knowing when the right next step is no call#
Outcome handling also includes knowing when to stop. In our collections processes, a customer may ask for a person, fail identity verification, withdraw consent, report a dispute or a bereavement, or say they are in financial difficulty. Each of these has a defined destination agreed with the client: a transfer to the client's own staff, a structured stop or a referral. None of them should send the case back into the contact cycle as if nothing had happened.
Debt details are never disclosed before identity is verified. That is another rule the platform enforces rather than a line in a script, and it is the reason a call answered by someone other than the customer ends without anything being said about the account.
Every attempt on the record#
A contact process will eventually be asked to explain itself: by a customer who complains, by the client's compliance team or by a regulator. The question is always the same. What happened on this case, when, and why?
In the processes we run, every attempt is logged with its technical status, and every conversation can be reconstructed from its recording, logs, statuses and the version of the dialogue that was live at the time. That means we can show, for any case, when each attempt was made, what it returned and which rule governed what followed. A complaint becomes a fact-finding exercise of minutes rather than a debate about what probably happened.
Judging outcome rules on the right rung#
We measure our work on four rungs, in order. Reach: did we get through to the right person? Continuity: did they stay on the line after the opening? Information: did we obtain what the process needs, such as a promise to pay, a confirmation or a completed survey? Outcome: did the business effect follow, such as a payment, an attendance or a contract, measured in the client's own systems?
Outcome rules are usually discussed only on the first rung. In our experience that is a mistake. Rules that lift connections can lower continuity and the quality of what customers commit to, because the customer who answers after being pressed is not in the mood to agree to anything. We therefore judge changes to outcome handling on the rungs above reach, and against a comparator on comparable cases, not on reach alone.
This is also why the rules are not set once. We tune them in production, per process and per market, and every change is agreed, versioned and traceable. What we publish is the principle. The settings stay in the specification.
What to ask a supplier#
- Does each call outcome have its own rule, or do several outcomes share one? Which ones, and why?
- Who sets those rules, who approves a change, and where is the history of changes kept?
- Is an agreed callback kept by the system, and can you show how often it is kept on time?
- Can your platform tell a network drop from a customer hang-up, and voicemail from a person?
- Which limits are enforced by the platform itself, rather than by policy or by agents remembering them?
- Which statements from a customer take a case out of the contact cycle, and where does each one go?
- For any single case, can you reconstruct every attempt, its status and the rule that applied next?
- On which rung do you judge a change to outcome handling, and against what comparator?
The answers will tell you more about a supplier than a demo will. A demo shows the conversation that worked. These questions show what happens to the hundreds of attempts that did not, and that is where your customers form their view of you.
Designing, running and answering for that part of the process is what we do at Vee. If the answers you are getting today are vague, that is a conversation 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