Customer service leaders are spending much more on AI from budgets that are barely growing. Gartner's survey of 199 leaders, published on 26 August 2026, found that AI spending had risen 38% while overall service and support function budgets had risen just 2%. When one line grows that much faster than the total, something else is being squeezed, and the question for every buyer is whether the new spend pays back the line it replaced.

What the survey found#

The source is a Gartner press release dated 26 August 2026, reporting a survey of 199 customer service and support leaders fielded in April and May 2026. The headline is the contrast between two growth rates: 38% for AI spending by those leaders and 2% for their function budgets as a whole. The release also ranks channels: leaders rate live chat and the phone as the most valuable today, and expect generative AI chatbots to lead within two years, followed by live chat and generative AI voicebots. Voice stays near the top even in a forecast led by chat.

The survey at a glance
ItemWhat the release states
PublisherGartner, press release
Date26 August 2026
Sample199 customer service and support leaders
FieldworkApril–May 2026
AI spending+38%
Function budgets+2%
ChannelsLive chat and phone most valuable today; in two years GenAI chatbots first, then live chat and GenAI voicebots
TypeLeader survey, not a market forecast

Source: Gartner press release, 26 August 2026. We report the figures as published; we have not seen the underlying dataset.

What the numbers do and do not support#

This is a survey of leaders describing their own spending. It is not a forecast of the market for voice AI, and it is not a measure of what that spending achieved. A 38% rise says nothing about the base it started from. If AI was a small line, a large percentage increase is a modest sum; if it was already substantial, the shift is significant. The release, as summarised in our research file, does not give that base.

The arithmetic still matters. If total budgets grow by 2% and one component grows by 38%, the difference must be funded by holding other components flat or cutting them, unless AI is a very small share of the total. In our view, the typical candidates are agent headcount, outsourcing contracts, legacy tooling and training. Gartner says leaders are redirecting spending away from labour and overhead and towards technology, but the release does not break that down by line, so we will not guess further. What we can say is that AI is no longer an innovation budget sitting on top of operations. In many service functions it is now competing with operations for the same money.

Spending is not the same as return#

Gartner published a second survey in November 2026, of 275 customer service leaders, which found that only 4% had achieved the return they expected from their AI initiatives. The two studies used different samples and should not be combined into one calculation. Placed side by side, though, they describe a familiar pattern: money moves faster than results.

In our experience the gap usually has little to do with the model. It opens in the work around it: process design, integration with the client's systems, telephony, testing, the rules for when a conversation should hand over to a person, and the daily listening that catches what reports miss. Our CEO puts it in terms of a lesson companies keep relearning.

Companies have learned this lesson many times: they buy good software and it stays in a drawer, because nobody can design the process on their own. – Dawid Wójcicki, Founder & CEO of Vee

Where the money should be measured#

If AI spend now competes with operating budgets, it should be judged in the same currency: results per process, not activity. For automated telephone contact, the unit that most often hides the truth is the minute. A minute rewards long calls, while the result often comes from shorter ones. That is why new agreements at Vee are no longer billed by the minute, and why, from 15 September 2026, we stopped publishing regular minute and case-volume figures in our market reports in favour of economic value and measurable results.

Our own volumes show why the minute is a weak proxy. In 2025 we handled 11.61 million cases and 7.18 million AI conversation minutes; since 2026 minutes have fallen faster than cases, because calls are shorter, the campaign mix changed and foreign processes grew (Vee operating data). A buyer paying per minute would have seen less to pay for while the work handled kept growing. In 2022 we made the same point in public: a solution with a low unit price and a low success rate can cost more than twice as much per success as one with a higher unit price and a much higher success rate.

How we structure what a client pays for#

  • A one-off setup for preparing and launching the process.
  • Base operation, charged per case or per account and independent of outcome.
  • An optional performance layer, linked to a result that can be measured in the client's own systems, with the definitions and measurement rules set out in the agreement.

We do not publish prices, and a structure is not a promise of savings. It does make the spending conversation simpler: the client can see what it pays to run the process and what it pays for the result, which is the comparison a squeezed budget demands.

Why the voicebot finding matters to regulated firms#

The finding that voice remains among the channels leaders value most is consistent with what we see. In banking, consumer finance, insurance, healthcare, telecoms and energy, the phone remains where many decisive conversations happen: a debtor in early delinquency, a policy renewal, a missed medical appointment, a technician visit that needs confirming. These are also the conversations where a mistake costs money or puts a licence at risk.

Spending on voice in these sectors is therefore not only a technology choice. It brings obligations: AI disclosure under Article 50 of the AI Act, which has applied since 2 August 2026; data-processing agreements and impact assessments; reconstruction of every call through recordings, logs, statuses and script version; and, for financial institutions, their own DORA duties towards suppliers. A budget line that ignores those costs will look cheaper than it is.

Scale is the other consideration. In March 2020, demand from large clients rose roughly fivefold and we raised channel capacity for one mobile operator from 250 to 900 (historical, Vee operating data). That was possible because scale came from infrastructure rather than hiring. When budgets barely grow, the ability to absorb volume without adding headcount is a large part of what AI spending is supposed to buy, and it should be tested, not assumed.

If the money comes from headcount, read the other surveys too#

The obvious assumption is that AI spending is funded by fewer agents. Gartner's own data is more cautious. In a survey of 275 leaders published in November 2026, only 21% said they had reduced human-agent headcount because of AI, and 7% had already rehired. An earlier Gartner survey of 321 leaders, fielded in September and October 2025, found that 85% were expanding the responsibilities of human agents. These are separate samples, so we do not net them against the 38%. Together they suggest that, for most organisations, AI spending is not yet replacing people on the scale the budget shift might imply.

That has a practical consequence. If AI spend rises while the human operation stays roughly the same size, the organisation pays twice for a while: once for the new tools and once for the team that still carries the work. The way out is not a bigger tool budget. It is a process that actually runs end to end, with a clear line between what the AI executes and what is deliberately routed to the client's own staff. Vee employs no call agents; the conversations are executed by Vee's AI, and our people design, listen, analyse and improve. That division is what makes the cost of the service visible in one place.

Questions to ask before the next AI line item#

  1. Which existing cost does this spend replace, and how will we know if it does?
  2. What is the unit of payment: minutes, cases, accounts or results? Does it reward the outcome we want?
  3. Which result will be measured, against which comparator, on which buckets and portfolios?
  4. Who designs and runs the process after launch, and who pays when something they missed goes wrong?
  5. What compliance costs (disclosure, reconstruction, audits) sit outside the quoted price?

Gartner's survey shows that service leaders have already moved money towards AI. The harder decision now is how they measure what it returns. We built our model around the result per case rather than the minute because that is the comparison a flat budget eventually forces on everyone.

Follow The First Second

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

Follow via RSS