KPMG's The Big Shift argues that agentic AI will change customer experience in two ways at once: by orchestrating journeys behind the scenes and by taking part in them directly. Both roles are plausible, and both raise the same question for a regulated business, which is who governs each decision the AI makes or carries out. The report is an advisory publication from KPMG Australia rather than a statistical study, and it is most useful read as a framework to test against your own operations.
What the publication is#
The Big Shift: preparing for the rise of agentic customer experience is KPMG Australia's 2025–26 Customer Experience Excellence report, released in late 2025. It sits alongside KPMG's global Customer Experience Excellence study for the same year, which reports 80,594 consumer interviews about 2,684 brands in 16 countries and regions, and it applies that programme's thinking to the Australian market. KPMG withdrew that global report in June 2026 after most of its citations were found to be untraceable; The Big Shift, which shares parts of its Total Experience text, is no longer at its original KPMG address, so we link an archived copy. Its themes are agentic customer experience, trust, the Total Experience model and the governance of customer touchpoints.
| Item | What we know |
|---|---|
| Publisher | KPMG Australia |
| Title | The Big Shift: preparing for the rise of agentic customer experience |
| Edition | Customer Experience Excellence FY25–26, released late 2025 |
| Market | Australia, within KPMG's global CEE programme |
| Themes | Agentic CX, trust, Total Experience, governance of touchpoints |
| Type | Advisory analysis, not an independent study of telephone contact |
Source: KPMG Australia, The Big Shift (PDF, archived copy; no longer at its original KPMG address). We quote no figures from it.
Two roles for agentic AI#
The report's most useful idea is the distinction between two roles. As an orchestrator, agentic AI coordinates journeys: deciding which step comes next, which channel to use and when to involve a person. As a participant, it deals with customers directly, resolving their needs in the conversation itself. KPMG places both inside its Total Experience model, which joins customer, employee, partner and digital interactions into one system rather than a series of separate touchpoints.
The distinction matters because the two roles carry different risks. A participant that says the wrong thing harms one conversation. An orchestrator that makes the wrong decision, for example contacting the wrong segment, at the wrong time, with the wrong message, can harm thousands. Governance has to be designed for each role separately.
How we divide those roles in regulated contact#
Our practice, built over years of running telephone contact for banks, insurers, healthcare providers, telecoms and energy companies, divides the two roles quite sharply. It is not the only possible design, but it shows what governing them looks like in operation.
| Decision | Who owns it | Why |
|---|---|---|
| Which accounts, portfolios and buckets to contact | The client's own analysts | Allocation is a business and risk decision that stays with the client |
| How to reach them: number, window, attempts, SMS, voice | Vee, within agreed operating rules | Reachability is our craft; the rules are agreed before launch |
| What the conversation may say | Approved content in a closed, versioned library | Every change in a regulated process is versioned and approved |
| When to hand over, stop or refer | Exception paths agreed with the client | Hardship, disputes, consent withdrawn, requests for a person |
| What the outcome means for the business | The client, in its own systems | The business status is the client's version of the truth |
Vee operating practice. Not drawn from KPMG's report.
In KPMG's terms, our AI is a participant in every conversation and a partial orchestrator of contact: it decides retries, timing and channel within rules the client has approved. The strategic orchestration, which customers to contact and why, stays with the client. We think that is the right split for regulated firms today, because the consequences of allocation decisions sit in the client's risk, capital and conduct obligations, not ours.
Our job is to understand the client's reality, not to change it. What we can change is how effectively the customer contact is run. – Dawid Wójcicki, Founder & CEO of Vee
What partial orchestration looks like day to day#
Orchestration within approved rules is less glamorous than the word suggests. It means deciding how, when and through which number and channel each customer is contacted, and stopping when the agreed limits are reached. Each of these decisions affects whether the customer is reached at all, and reach is the first rung of everything else: without reaching the customer, there is nothing else.
These decisions are also where an agentic system can most easily overstep. A model optimising for contact rate alone might call more often, earlier or later than a bank's conduct rules allow. That is why the limits come first and the optimisation second. The windows, attempt caps and stop conditions are agreed with the client before launch, and the system works inside them. We describe the running elements, such as number management, SMS and retry timing, as they are today, and keep further automation of orchestration as a direction of development rather than a claim.
Trust is designed, not declared#
The Big Shift treats trust as a precondition for agentic experience. In a regulated phone conversation trust is concrete. It means the customer knows they are speaking to AI, which Vee discloses in every call in line with Article 50 of the AI Act, applicable since 2 August 2026. It means identity is verified before any debt detail is disclosed, as a rule of the platform rather than a line in a script. And it means the conversation cannot improvise about a customer's money, health or contract.
That last point is where generative AI needs most care. Our technology posture runs from fully deterministic dialogue through small models that only classify intent, to large language models used for classification without generating replies, to generative variants for low-sensitivity processes. In regulated processes, replies come from a closed, versioned library and there is no learning in production. Deterministic where it matters, generative where it helps. An agentic system that cannot be held to approved content has not earned the trust KPMG describes.
Governing the touchpoint#
The report's emphasis on touchpoint governance is, in our experience, the part most organisations underestimate. Governance is not a policy document. It is a set of operating answers: who approves a change, how quickly a fix reaches live conversations, and how any single call can be reconstructed after a complaint.
- Change control. Client-side changes are approved by the client; if something on our side needs correcting, we correct it straight away. One approved change applies to every live conversation at once.
- Reconstruction. Every call can be rebuilt from its recording, logs, statuses and the script version in force at the time.
- Listening. In every project our people listen to selected calls daily, especially unusual ones, because dashboards do not hear what customers actually say.
- Security baseline. Vee Center Sp. z o.o. holds ISO/IEC 27001:2022 certification; banks audit us as a sub-processor, and their own GDPR and DORA obligations shape the contract.
None of this is specific to agentic AI. It is what running an automated process at scale requires. Since 2012 we have handled 48,626,959 cases with our AI (cumulative to 31 October 2026, Vee operating data; a case is a record assigned for handling, not a person or a call), and the governance questions have not changed with each new generation of models. They have only become more important as the models became more capable.
What not to take from the report#
The Big Shift is a forward-looking framework from one KPMG member firm for one market. It is not a survey of European banks or insurers, and it says nothing about Polish, Austrian or German telephone contact. It should not be cited as evidence that agentic AI already improves customer experience in regulated sectors, or that any particular architecture is safe. Its value is that it names the roles and the governance questions clearly.
Questions to ask before giving AI an agentic role#
- Which decisions will the AI take as orchestrator, and which as participant?
- Which of those decisions must stay with our own analysts and risk owners?
- What may the AI say without approval, and how is that limited in regulated processes?
- Who approves a change, how fast does it reach live conversations, and how is it rolled back?
- Can we reconstruct any single interaction, months later, for a regulator or a complaint?
KPMG is right that agentic AI will reshape customer experience. In regulated contact, the reshaping will be decided less by what the AI can do than by how clearly its roles are governed. That is the work we do with every client before the first call is made.
Follow The First Second
Occasional issues on the mechanics and economics of AI-powered customer contact. No noise, no vendor news.
Follow via RSS