Deloitte Digital’s 2026 Global Contact Center Survey says that the contact centres furthest behind on AI are held back by integration, legacy systems and security and compliance, and that the most mature report far higher profitability. Both findings are useful, provided we read the second as a correlation and remember that one survey, reported twice, is still one survey.
Below we set out what Deloitte published, the limits of each figure, and how the picture compares with running telephone customer contact for regulated enterprises.
Two publications, one dataset#
Deloitte Digital released the 2026 Global Contact Center Survey on 4 June 2026, with a press release following on 9 June 2026. The respondents were contact-centre leaders across industries and geographies, surveyed in late 2025 and early 2026; Deloitte describes them as hundreds of leaders and does not publish an exact sample size. We worked from the public summary and the press release; access to the full report was not confirmed, so we report only figures stated in those two documents.
The press release is not a second, independent source. It presents the same measurement. Articles that cite “two Deloitte studies” on contact-centre AI are counting one dataset twice.
| Finding | Figure | Published in |
|---|---|---|
| Weaker organisations citing integration problems | 72% | Survey summary, 4 Jun 2026 |
| Weaker organisations citing legacy systems | 58% | Survey summary, 4 Jun 2026 |
| Weaker organisations citing security and compliance | 53% | Survey summary, 4 Jun 2026 |
| Higher profitability reported by AI-mature organisations | 85% | Survey summary, 4 Jun 2026 |
| Contact centres using agentic AI | 35% | Press release, 9 Jun 2026 |
Deloitte Digital, 2026 Global Contact Center Survey and press release. Survey of contact-centre leaders, late 2025 to early 2026. Subgroup comparisons; correlation, not causation.
The release adds one more observation that sits awkwardly next to the adoption figure: more than half of surveyed consumers said service quality stayed the same or got worse in 2025. More agentic AI in more contact centres has not, on this evidence, translated into a better experience for the people calling. The release does not tell us which contact centres those consumers dealt with, or whether they met AI at all, so the two findings should not be joined into a claim that agentic AI made service worse. What they do support is a narrower point: adoption on its own is not evidence of a better experience, and it should not be reported as one.
Maturity correlates with profit. That is all it shows.#
The 85% profitability figure will be quoted widely, and it should be quoted carefully. It compares groups of organisations sorted by their AI maturity. It does not show that becoming mature caused the profit. Firms that are already well run, well funded and well integrated are more likely both to deploy AI successfully and to be profitable. The direction could run either way, or both could share a third cause.
In our own reporting we hold ourselves to the same discipline. We compare against a comparator on the same buckets and portfolios, and we talk about incremental effect. As our team puts it internally, Vee does not deliver 70%; Vee delivers the additional points above what would have happened anyway.
Integration, legacy, compliance: the real definition of immaturity#
The barriers reported by weaker organisations are more instructive than the maturity score itself. Nearly three quarters cite integration, more than half legacy systems, more than half security and compliance. None of these is a model problem. They are problems of plumbing, history and governance.
Integration#
A telephone conversation creates value only if its result lands somewhere useful. In our processes data moves through REST APIs, SFTP flat files or HTTPS, and every case returns a status on four axes: technical, dialogue, processing and business. The business status is the client’s version of the truth, in its own language, and the one its analysts act on. Without that loop an AI conversation is a recording, not a process.
Legacy#
Most regulated enterprises will not replace their core systems to suit a contact-centre project, and they should not have to. Our job is to understand the client’s reality rather than change it, and to connect telephony, dialler, schedules and data exchange to what already exists. That includes the telephony layer itself, numbers, SIP trunks, routing and transfers, which we run rather than leave to a separate supplier.
Security and compliance#
Banks, insurers and healthcare providers audit their suppliers, and they are right to. Vee Center Sp. z o.o. holds ISO/IEC 27001:2022. Every content change in a regulated set-up is versioned and approved; every call can be reconstructed from recording, logs, statuses and script version; and 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 a certificate of compliance, and nobody should.
35% use agentic AI, and quality did not improve#
The press release puts agentic AI use at 35% of contact centres. Read beside the consumer finding, it suggests that adoption and experience are moving at different speeds. TELUS Digital’s vendor-sponsored 2026 survey of 815 CX leaders describes the same pattern as deployment outpacing optimisation, and it fits our experience: launching a process is the start of the work, not the end of it.
What closes that gap is the unglamorous part. In every project our quality team listens to selected calls every day, particularly unusual ones. Analysts look for correlations and run A/B tests. Changes go live once and apply to every conversation at the same time, because we do not have to retrain a floor of people. That is the operating rhythm that turns adoption into results.
Every day, in every project, our people listen to selected calls, especially unusual ones, to hear with their own ears what technology still cannot. – Dawid Wójcicki, Founder & CEO of Vee
Maturity is a rhythm, not a score#
Surveys have to sort organisations into groups, and a maturity label is a convenient way to do it. In operation, maturity looks less like a level reached and more like a routine kept. Our managed production starts intensively after launch, then settles into a weekly and later a monthly rhythm with the client, and the rhythm does not stop after the first year. Long-standing processes still have weekly meetings, daily listening and joint decisions on the next base, bucket, process or country.
That rhythm needs a shared vocabulary for results. We measure on four rungs: reach, because without reaching the customer there is nothing else; continuation, the inverse of hang-ups after the company name is heard; obtaining the key information, such as a promise to pay or a confirmed visit; and the real business effect, such as the payment itself. A contact centre that reports only the first rung, or only talk time, will find it hard to know whether it is maturing at all.
There is also the question of what happens when something goes wrong. In our model, anything on our side that needs correcting is corrected straight away, and changes the client requests are approved on its side; how changes are scoped and settled is set out in the agreement. That sounds administrative, but it is part of what keeps a process improving: nobody has a reason to leave a known problem in production.
How it compares with what we see#
Deloitte’s barriers match the order of difficulty we meet in projects. Integration and data exchange take more of a launch than dialogue design. Security reviews by banks are thorough and necessary. The model behind the conversation is rarely the deciding factor; we choose the speech engine per language and keep replies in regulated processes inside a closed library precisely so that the hard problems can be solved where they live.
Where maturity shows, it shows in specific numbers with a specific scope. In soft collections for an insurer, historically, 61.9% of cases closed at first contact. In one inbound FAQ process in mobile payments, historically, first-contact resolution reached 73.5%. Neither number is a promise for another process; both come from processes that were integrated, measured and run for long enough to improve.
What to ask before calling yourself mature#
- Does every call return a usable status? Case-level, daily, in your own business language.
- Which legacy systems must stay? Design around them rather than waiting for a replacement programme.
- What can your auditors reconstruct? Recording, logs, statuses and the exact script version for any call.
- Is experience measured, not assumed? Track the customer’s side alongside cost, so adoption does not hide a decline in quality.
- Is the effect incremental? Ask for a comparator on the same portfolios, not a before-and-after chart.
Deloitte’s survey is a fair map of where contact centres get stuck. It does not prove that AI maturity creates profit, and it does not need to. Its practical lesson is that the barriers are integration, legacy and governance, and that each of them is solved in operation, one process at a time.
Follow The First Second
Occasional issues on the mechanics and economics of AI-powered customer contact. No noise, no vendor news.
Follow via RSS