Our inbound lines rarely begin as inbound projects. More often, a client whose outbound process has been running well for some time says something like: since this works so well, could we add a small inbound line next to it? That sentence has started some of our most useful work.

The order matters. An outbound process that already works has solved most of the hard problems an inbound line will face: the integration with the client's systems, identity verification, the compliance set-up, the voice, the vocabulary of the product and the agreed rules for handing a customer to a person. In our experience, adding inbound on top is a smaller step than starting from nothing, and the result is usually better. This article describes how that step is taken and what the numbers from two of our inbound configurations do and do not tell you.

Why outbound first is the natural order#

Many clients assume inbound is the easy part. In one sense it is: a customer who calls you wants something and will tolerate a lot to get it. Outbound is harder, because the person you call did not ask to talk to you.

Clients often assume outbound is simpler than inbound. Someone who calls you will cope even with an IVR. Someone you call may not even want to pick up, let alone keep talking. – Dawid Wójcicki, Founder & CEO of Vee

That tolerance is exactly why inbound is so often neglected. Telephone service becomes a second-line function: long queues, an IVR tree that sends people in circles, opening hours that end before most people leave work. Customers cope, so nobody fixes it. A working outbound process changes the picture because it generates inbound traffic of its own. Customers who missed a call ring back. Customers who received a text want to ask a question. Customers who made a promise in a call want to change it. Those calls deserve the same quality as the outbound conversation that prompted them.

What a small inbound actually covers#

Small does not mean trivial. It means a defined scope, agreed in advance, rather than an attempt to replace the client's whole service desk. Typical first scopes look like this:

  • Callbacks from outbound campaigns: the customer rings the number that called them and is recognised, verified and helped in the context of that campaign.
  • Frequent questions with fixed answers: how to pay, where to find a document, what a status means, how to change a date.
  • Data collection: capturing a request, a meter reading, a preferred time or a change of details, and passing it into the client's systems.
  • Routing exceptions: recognising quickly what the process should not handle and passing it to the client's own team with context.

Each item is tuned to the client's base, portfolio, product and, in collections, the bucket the customer sits in. A question about a payment from a customer one week late is a different conversation from the same question asked by a customer three months late.

Every call answered without a queue#

The first promise of an AI-run inbound line is that callers do not queue. That is not a slogan; it is an engineering choice. A human contact centre sizes its staff to an expected volume and accepts queues at peaks. An AI call centre sizes channels, and channels are far easier to over-provision than people. With channels configured for the agreed traffic, calls are picked up without a queue, including at peaks that would overwhelm a staffed desk.

99.99% / 0 sService level and average wait timeShare of calls answered within the target time, and the average time callers waited before being answered.One inbound configuration; not a universal SLA.

We publish this figure with its scope deliberately. It comes from one configuration, with channel capacity set for that client's traffic. It shows what is achievable when telephony and capacity are designed for zero waiting. It is not a contractual promise for every line, because service level depends on how many channels the client chooses to provision and on what happens in the telephony networks between the caller and us.

Every answer approved in advance#

The second promise matters more in regulated sectors. In an inbound line run by Vee, the answers the customer hears come from a closed, versioned library approved by the client. The models recognise what the customer wants; they do not write the reply. If a question falls outside the approved answers, the process does not improvise. It routes the call.

This has practical consequences. Legal and compliance teams review what the line can say before it goes live. Every change is versioned and approved, and applies to every live conversation at once, so there is never a period when some customers hear the old answer and some the new. And every call can be reconstructed afterwards: recording, logs, statuses and the exact version of the script that was in force.

Approved answers also make the line consistent with the outbound process next to it. The customer who was told something in an outbound call hears the same thing when they call back. In a staffed contact centre, that consistency depends on training and memory. Here it depends on a single library.

Resolution on the first call#

Answering immediately is only worth something if the call ends with the customer's question settled. The measure we care about for inbound is first call resolution: the share of calls in which the customer got what they needed without calling again or being passed elsewhere.

73.5%First call resolutionShare of inbound calls in which the customer's request was resolved in that call, without a repeat call or transfer.Inbound FAQ, mobile payments, historical.

The context is a line answering frequent questions for a mobile payments service. Questions in that kind of service are repetitive and well defined, which suits an approved library. The remaining calls are not failures by default: some were questions that should go to a person, and the process sent them there. A resolution rate is only honest if the routing of the rest is honest too.

The route to a person is part of the design#

Customers expect it. Gartner's survey of 3,566 customers, published on 4 August 2026, found that 87% say companies using generative AI for customer service must give them access to a human agent. That is a stated preference, not a legal requirement, but it matches what we hear in calls.

So every inbound line has a defined route out: which cases go to the client's staff, when, with what context, and what happens outside their working hours. Typical triggers are a customer asking for a person, a failed identity check, a complaint, a vulnerable situation or a question outside the approved scope. The rules are agreed before launch and reviewed as the line runs. Vee employs no call agents; the people who take those transfers are the client's own team, and the process passes them the context so the customer does not have to repeat themselves.

How the step is taken#

Because the outbound process already exists, adding inbound follows a shorter path than a new project. It still follows a path, and skipping steps is how small inbound lines turn into big problems.

  1. Look at the traffic that already exists. Before designing anything, we look at what customers do after outbound contact: how many call back, when, and what they ask. Listening to those calls usually defines the first scope better than any workshop.
  2. Reuse what is proven. Identity verification, integration, vocabulary and exception rules come from the outbound process. Only what is new for inbound is specified from scratch: greeting, intent recognition for open questions, the answer library and the routing table.
  3. Agree numbers and routing. Which numbers customers will call, how calls reach us, and how transfers reach the client's team. Telephony is part of our service, so this is one conversation, not three suppliers.
  4. Approve the answers. The client's content owner and, in regulated sectors, legal or compliance sign off every answer before launch.
  5. Start on part of the traffic. One number, one campaign's callbacks or certain hours first. The team listens closely, adjusts and agrees with the client when to widen the scope.

What inbound gives back to the outbound#

The benefit runs both ways. An outbound campaign always leaves people who were not reached, or who heard a message and could not respond at that moment. Without an inbound line, their callbacks land in a general queue, wait and often give up. With one, a callback becomes a completed case: the customer is recognised, the campaign context is known and the outcome is recorded with the same statuses as an outbound conversation.

That changes how the client reads its results. Instead of two disconnected reports, one for outbound attempts and one for the service desk, the client sees one case with everything that happened to it, in whichever direction the call went. When we sit down with the client each week, we discuss the case, not the channel.

We don't manage people, so we control the service far better: one change and every live conversation already reflects it. – Dawid Wójcicki, Founder & CEO of Vee

What to check before adding inbound#

If you already run an outbound process that works, with us or anyone else, these are the questions that decide whether a small inbound line next to it will work too:

  1. Which inbound calls does the outbound process already generate, and where do they land today?
  2. Which questions have fixed, approvable answers, and who in legal or compliance will approve them?
  3. How many channels are needed for zero waiting at your peaks, and who decides that capacity?
  4. Which cases must go to a person, and does your team have the capacity and hours to take them?
  5. Will you measure resolution on the first call, not only the service level?
  6. Can the outbound and inbound share one library of answers and one view of the customer?

Good telephone service is not a second-line function. It is the first line, and when it works customers barely notice it. Starting small, next to something that already works, is the surest way we know to get there.

Follow The First Second

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

Follow via RSS