AI has reached technology, media and telecoms companies faster than the rules for using it. KPMG in Germany's TMT Monitor 2026, published on 21 May 2026, found that 92% of the companies surveyed use AI, but only about 10% have a fully developed AI strategy, and only a quarter have established AI guidelines. At the same time, 79% expect AI to improve customer service. We think customer service is therefore where much of the ungoverned AI is likely to meet customers first.
What the study found#
The TMT Monitor 2026 is a study by KPMG in Germany of the technology, media and telecommunications sector. KPMG's study page states that it surveyed 180 companies in the sector, supplemented by a case study from each of the three sectors. The press release, dated 21 May 2026 in Berlin, carries the headline finding that AI adoption is growing faster than governance. Neither the release nor the study page, as we read them, describes the survey method or the fieldwork dates.
| Item | What KPMG states |
|---|---|
| Publisher | KPMG in Germany |
| Date | 21 May 2026 |
| Sample | 180 companies in technology, media and telecommunications |
| AI use | 92% use AI in various areas |
| Strategy | About 10% have a fully developed AI strategy; nearly 60% are still planning one |
| Governance | 25% have AI guidelines; 23% have internal oversight bodies; 22% train employees systematically |
| Expected benefits | 79% customer service and user experience; 78% process efficiency; 73% revenue growth |
| Type | Company survey; expectations, not measured results |
Sources: KPMG in Germany press release and TMT Monitor 2026 study page. Method and fieldwork dates are not stated on either page.
KPMG's commentary, from Katja Modder, head of technology, media and telecommunications at KPMG AG Wirtschaftsprüfungsgesellschaft, makes the point plainly: AI has arrived in companies faster than the organisational conditions for using it, and trust has to be built through clear responsibilities, reliable data and effective controls.
Why customer contact is where the gap shows#
Internal AI use with weak governance mostly produces internal problems: an inaccurate summary, a wasted analysis. AI that talks to customers is different. Each conversation is a statement by the company, made to someone who may record it, complain about it or rely on it. When 79% of companies expect customer-service benefits and only about a quarter have guidelines, we would expect a share of that benefit to be pursued without agreed rules for what the AI may say, when it must stop and who answers for it.
Telecoms is a sector where this matters at scale. Operators run large subscriber bases, frequent contact for retention, renewals, pre-sales and collections, and outbound sales that are expensive to staff. The pressure to automate is high, and so is the cost of an automated conversation that misleads a customer about a contract or a price.
What governance means in a live conversation#
KPMG lists the ingredients as responsibilities, data and controls. In a managed telephone process those become operating answers, decided before the first call.
| Question | Our operating answer |
|---|---|
| Who is responsible? | A tandem on our side, a business owner and an operations owner; on the client's side a process owner, a budget owner with final approval and a content approver |
| What may the AI say? | In regulated processes, replies come from a closed, versioned library; no learning in production |
| How is a change controlled? | Changes are approved by the client, and corrections on our side are made straight away; one approved change reaches every live conversation at once |
| Can a call be reconstructed? | Yes: recording, logs, statuses and script version |
| Does the customer know it is AI? | Yes, in every call, in line with Article 50 of the AI Act |
| Who checks what the dashboards miss? | People who listen to selected calls every day, in every project |
Vee operating practice, not drawn from KPMG's study.
None of this slows a deployment down for long. It is what allows a process to scale without each new campaign reopening the same risk questions. In our view, organisations that use AI widely without a strategy tend to discover these questions one incident at a time.
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
The regulatory clock has already started#
For European companies, governance of customer-facing AI is no longer only good practice. The transparency obligations in Article 50 of the AI Act have applied since 2 August 2026, and the European Commission published guidelines on them on 20 July 2026. An AI system that interacts with people must make that clear, with the specific requirements depending on the use case. Financial institutions add their own DORA obligations towards ICT suppliers, and GDPR applies throughout.
We help clients assess their own configuration against these rules; we do not offer certification-like assurances. What we can show is the baseline: Vee Center Sp. z o.o. holds ISO/IEC 27001:2022 certification, banks audit us as a sub-processor, and the AI disclosure is built into every conversation rather than added as a disclaimer.
A strategy for contact starts with processes, not tools#
Nearly 60% of the companies KPMG surveyed are still planning their AI strategy. For customer contact, we would suggest that a strategy begins with a list of processes rather than a list of tools. Which conversations happen at scale, which carry regulatory or financial weight, which are repetitive enough to automate well, and which need a person? The answers differ by sector and market, and they decide the governance each process needs.
In telecoms, the obvious candidates are the high-volume conversations: retention and renewals across the subscriber base, pre-sales for fibre and bundles, and collections. Active outbound sales is a large cost for operators, and it is weaker in consumer markets than in business accounts, where a dedicated manager holds the relationship. A practical division of labour is that the AI takes the first contact at scale and the operator's salespeople receive the customers who are interested, at a pace their team can handle.
This is not new to us. The first outbound campaign we designed, years before generative AI, measured interest in a mobile-internet bundle, and it used a studio voice rather than synthesis because that worked better on cold contact. What has changed since then is not the need for governance but its scope: the more capable the AI, the more carefully its limits have to be set and checked.
What we see in German-speaking markets#
We work in three foreign markets, with a fourth in preparation, and took on our first portfolio in Germany in August 2026, after working in Austria since 2023. Our experience there is consistent with KPMG's framing in one respect: clients ask governance questions early and in detail. Who approves wording, how changes are versioned, how a call can be reconstructed, where data is processed. We regard that as a strength of the market, not an obstacle.
Language and conduct differ too. Words that work in one market do not automatically work in the next, and call windows, retry spacing and the tone of a reminder all have to be set for the market rather than copied. Governance includes those choices, because a conversation that is technically correct but culturally wrong still reflects on the client.
Questions to ask before AI talks to your customers#
- Do we have written guidelines for customer-facing AI, and who owns them?
- What may the AI say without approval, and how is that enforced in the system rather than in a policy?
- Who approves changes, how are they versioned, and how fast do they reach live conversations?
- Can we reconstruct any conversation for a complaint or a regulator?
- How and when does the customer learn they are speaking to AI?
KPMG's TMT Monitor shows adoption running ahead of the rules. In customer contact the rules are part of the product, and they should be in place before the first conversation, not after the first complaint. That is the order in which we build every process we run.
Follow The First Second
Occasional issues on the mechanics and economics of AI-powered customer contact. No noise, no vendor news.
Follow via RSS