Every telephone process changes after launch, and the way those changes are made says more about a supplier than its proposal does. Ours rests on three habits. If something on our side needs correcting, we correct it straight away. If a change starts on the client's side, it is approved there before it goes live. And in regulated configurations, every change to content is versioned and approved before it reaches a single customer.

Those habits settle most questions before they become arguments. The rest of this article explains why they work that way, and what they mean in practice for a bank, an insurer or a healthcare provider running a process with us. The commercial side, how changes are scoped and settled, belongs in the agreement, and we leave it there.

Not all changes are the same#

The word “change” covers very different situations, and treating them alike is where disputes come from. We distinguish four kinds.

Four kinds of change and how each is handled
Kind of changeExampleWho approves
Our omissionA dialogue path we did not anticipate, a status mapped wrongly, a rule that does not match what was agreedWe decide and inform the client; content still follows the approval route
Client-side changeA system on the client's side works differently than described, a requirement was forgotten, a new product ruleThe client
OptimisationNew wording, an alternative voice or another adjustment, tested on part of the baseClient, for content and anything affecting customers; tests by agreement
New scopeA new bucket, portfolio, lead source, process or marketThe client, after a workshop

Typical handling in Vee's delivery model. How changes are scoped and settled is set out in each agreement.

If something on our side needs correcting, we correct it#

The first habit is the simplest and the one we hold to most strictly. When we find that something in the process is wrong because we overlooked it, and it matters for the results of the campaign, there is no queue of approvals to wait for on our side. We implement the correction as fast as possible, and any content involved still follows the client's approval route.

This follows from where we sit. Vee runs the process; it does not sell a platform and leave the client to configure it. When we design the dialogue, the schedule, the integrations and the statuses, the quality of that design is ours to answer for. Customers keep being called while a problem waits, so speed matters more than ceremony.

We play on the client's side of the table. – Dawid Wójcicki, Founder & CEO of Vee

There is a practical side too. Our quality team listens to selected recordings every day in every project, and our analysts read statuses across the whole base. We usually find our own omissions before the client does, and a shared, versioned specification, approved by the client at each stage, makes it clear what was agreed and when.

Client-side changes are approved on the client's side#

The second kind of change starts on the client's side. A core system returns data in a slightly different format than the specification said. A transfer number changes. A business unit remembers an exception that was never mentioned in the workshop. These are normal; no specification survives first contact with a large organisation's systems intact.

On the client's side, several people shape a process: the process owner who sets its business goals, the people who approve content, the analysts who decide allocation, and the person who gives final approval. Knowing who holds which approval before launch is what keeps a change from stalling in someone's inbox, and it prevents the slow accumulation of small, unapproved extras that nobody remembers agreeing to.

Optimisation is not a change request#

Improvement work deserves its own category. When our analysts see that something might work better, we propose a test. Content and anything that affects how customers are treated is approved by the client first; tests run by agreement, often on part of a larger base so that the comparison means something. If the test wins, the new version becomes the standard. This work is part of running the process, not an extra the client must remember to order.

In regulated processes, every content change is versioned#

Banks, insurers and healthcare providers need more than clear change rules. They need to know, for any conversation on any date, exactly what the customer heard and why. That is why our regulated reference configuration works differently from a general-purpose voice agent.

  • A closed library. What the customer hears comes from a closed, approved and versioned library of utterances. Statistical models may recognise intent; they do not write new sentences to customers.
  • An approval owner and an effective date. Each version of content and rules has a named approver, a date from which it applies and a history of what it replaced.
  • No silent learning. Nothing learned from production data changes the customer-facing layer without passing through controlled change management.
  • Reconstruction. Recordings where agreed, connection logs, status history and the script version in force make it possible to reconstruct a conversation step by step.

In practice, each entry in the change log records the date, the owner, what changed, which scenarios or assets it affects and its approval status. It sounds bureaucratic, and in a sense it is. But it is the same discipline a bank already applies to its own customer communications, now applied to a conversation that may run many thousands of times in a single campaign. Under the AI Act's transparency rules, which have applied since 2 August 2026, the disclosure that the customer is talking to AI is part of that versioned content too.

One change, every live conversation#

There is one more reason change discipline matters more with us than with a traditional call centre. Vee employs no call agents; the conversations are executed by our AI. In a call centre, a change in wording spreads unevenly through training, shift by shift and person by person, and some agents never quite adopt it. With us, an approved change reaches every live conversation at once.

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

That is a strength when the change is right, and a reason for care before it goes live. So we test changes before release, roll them out on part of the base when there is any doubt about their effect, and keep the previous version ready. Speed and control are not opposites here; the version history is what lets us move quickly without losing track of what changed.

Incidents are a different conversation#

A change is a decision. An incident is something that happens to the process: a carrier problem, a failed file delivery, an integration timing out. We monitor most failures that can occur and work under SLAs, basic or extended, that set times for reaction, repair and communication. An incident team with both technical and business specialisms handles them. When the cause is on our side, we fix it, with the same urgency as any other correction of ours.

What to agree with any supplier before launch#

Change rules belong in the contract and the operating documentation, not in goodwill. Before you start, ask for clear answers to these questions.

  1. Who on your side gives final approval of changes, and who approves content?
  2. How quickly does the supplier correct something it missed, and how is that defined in the agreement?
  3. How are client-side changes estimated and approved before they go live?
  4. Is customer-facing content generated or taken from a closed, versioned library?
  5. Who approves each version, from which date does it apply, and can you see the history?
  6. How fast does an approved change reach live conversations, and how is it rolled back?
  7. What are the SLA times for reaction, repair and communication when something fails?

A supplier that answers these plainly is one you can disagree with productively. That is what a managed service should be: clear rules about who decides, how a change is approved and what is written down, so that the conversation can stay on results.

Follow The First Second

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

Follow via RSS