EY’s study of AI in customer service finds that fewer than a quarter of organisations have turned AI into realised value, and that nearly all of those leaders run voice AI at scale. The weaker groups are held back less by technology than by governance, compliance and change management, which is exactly where regulated customer contact is won or lost.
The study is useful and also small, self-reported and led by one market. Here is what it says, how far each figure can be stretched, and how it compares with what we see running telephone contact for regulated enterprises.
What EY published#
The report is “Demystifying AI in customer service: Navigating AI-led transformation to value realization”, announced by EY on 4 November 2026 through its India newsroom, with the full PDF available via an external download. It draws on more than 100 survey responses and executive interviews with customer-service leaders, with fieldwork between the fourth quarter of 2025 and the first quarter of 2026. The release says respondents span industries and geographies but does not break them down by country.
EY divides respondents into three groups by how far they have progressed from AI adoption to value.
| Finding | Figure |
|---|---|
| Achievers | 23% of respondents |
| Challengers | 33% of respondents |
| Aspirants | 44% of respondents |
| Weaker groups citing governance, compliance or change management as barriers | More than 40% |
| Achievers running voice AI deployments at scale | 91% |
| KPI improvement reported by top performers | Over 50% on at least three to four core KPIs, as reported by Achievers |
EY, Demystifying AI in customer service, 4 Nov 2026. 100+ survey responses and interviews combined, Q4 2025–Q1 2026; country mix not disclosed. Self-reported results, not an audit of causation.
How far the figures can be stretched#
Three limits apply. The sample is modest: a little over 100 respondents means that each maturity group is a few dozen organisations, and the Achievers perhaps two dozen. The results are self-reported, so “over 50% improvement” is what top performers say about their own KPIs, not what an auditor measured. And the release does not say where respondents are based, so the distribution of Achievers and Aspirants may look different in European banking or insurance.
The 50% figure needs particular care. EY says Achievers reported more than 50% improvement on at least three to four core metrics, citing examples such as customer experience, revenue uplift, average handle time and agent productivity. The public material does not tell us from what baseline, over what period or against what comparator. It is a reasonable signal that leaders feel they have made large gains. It is not a benchmark anyone should put in a contract.
The largest group is still at the start#
The distribution deserves as much attention as the leaders. Aspirants, at 44%, are the largest group, and together with Challengers they make up more than three quarters of respondents. Even allowing for the small sample, the picture is of a market where most organisations have adopted AI in some form but have not yet turned it into value they can point to.
That is not a new pattern. In late 2020 we argued that growth in automated customer contact would depend on the quality of the technology and on deployment experience, not on the availability of tools; a year later we observed that many firms had rushed into cheap, ineffective bots and been disappointed. EY’s groups describe a similar divide in 2026, with a broader range of AI and a larger budget behind it.
For a regulated enterprise sitting among the Aspirants, the useful question is not how to buy more AI, but which single process could be run well enough, and measured cleanly enough, to produce the first piece of realised value. One proven process is a better foundation for the second than a portfolio of pilots.
Voice at scale is what the leaders have in common#
The most striking number is the 91%. Among the organisations EY calls Achievers, nine in ten run voice AI deployments at scale. Correlation again, not proof, but it lines up with a view we have held for a long time: the telephone remains where much of the value of customer service sits, and voice is harder than text, which is why getting it right separates operations.
Voice at scale is not a model choice. It is telephony, numbers, routing, transfers, speech recognition tuned per language, dialogue design and a voice that customers will keep listening to. We run that whole stack as a service. Up to 31 October 2026, Vee had handled 38,325,799 connected AI conversations and 28,569,847 minutes of AI conversation time, both cumulative Vee operating data, with human agents not included in the minutes.
Scale also changes what “improvement” means. A single change to a dialogue, once approved, applies to every live conversation at the same time. That is one reason why voice operations that are run well can move a KPI quickly, and why one that is run badly can move it the wrong way just as quickly.
Governance, compliance and change: the barriers that matter#
More than 40% of Challengers and Aspirants, taken together, name governance, compliance or change management as what holds them back. For banks, insurers and healthcare providers this will sound familiar. The question is rarely whether an AI conversation can be built. It is whether it can be approved, audited, explained and changed safely.
Governance#
In regulated processes, our replies come from a closed, versioned library; language models, where used, classify what the customer means but do not generate the answer; nothing learns in production. Each content change is versioned and approved by the client. Every call can be reconstructed from recording, logs, statuses and script version.
Compliance#
Vee Center Sp. z o.o. holds ISO/IEC 27001:2022, and banks audit us as a sub-processor. Our conversations always disclose that the customer is speaking with AI, in line with Article 50 of the AI Act, whose transparency obligations have applied since 2 August 2026. We help clients assess their own configuration; we do not offer certification-like assurances, and a supplier that does should be questioned.
Change#
Change management is often read as training staff to accept new tools. In our model the client’s team does not have to learn to operate the AI; we run it. The change that matters is in decision-making: who approves content, who owns the budget, who receives transferred calls and how results are judged. We work as a tandem on our side, a business owner and an operations owner, opposite a process owner and a budget owner on the client’s side.
We are not a software house. We don’t build a process, switch it on and say: now it will fly by itself. – Dawid Wójcicki, Founder & CEO of Vee
From pilot to realised value, in practice#
EY’s framing, from transformation to value realisation, is the right one. In our experience the route has recognisable steps. A diagnostic of goals, data and current results, including listening to the client’s own recordings. Contract and data-protection work. A specification covering dialogue, integration and telecoms. Tests end to end. A controlled pilot on a sample of the base, never the whole of it. Then managed production, intensive at first, then weekly and monthly, and expansion to new buckets, portfolios, processes or countries only as results are proven.
Realised value is then measured on the client’s own metric, against a comparator. That is how we can state, with scope, that in early-stage collections on-time payment frequency has historically risen by 7–13% versus a human-agent comparator, or that in soft collections for an insurer, historically, 61.9% of cases were closed at first contact. Those are smaller and more specific numbers than “over 50%”. They are also numbers with a definition attached.
What to ask before calling yourself an Achiever#
- Which KPIs improved, and against what? Name the metric, the baseline, the period and the comparator.
- Is voice in scope? If most of your service value sits on the phone, a text-only programme will not show it.
- Who approves content and changes? Write down the governance before launch, not after the first incident.
- Can every conversation be reconstructed? Recording, logs, statuses and script version, for any case your auditors choose.
- Who runs the process after go-live? Realised value depends on daily listening and weekly decisions, not a launch date.
EY’s study gives a useful vocabulary and a plausible pattern: the leaders run voice at scale and have solved governance, the rest are stuck on it. Read with its limits in mind, it supports a practical conclusion for regulated enterprises. Value comes from operating the process well, measuring it honestly and changing it under control, which is the work we take on for our clients.
Follow The First Second
Occasional issues on the mechanics and economics of AI-powered customer contact. No noise, no vendor news.
Follow via RSS