Gartner's prediction that agentic AI will resolve 80% of common customer service issues without human intervention by 2029 is among the most quoted numbers in our industry. It is a forecast, published in March 2025, about a subset of issues, and it is often repeated as if it described today's market or every call. Read precisely, it is a useful statement about direction. Read loosely, it becomes a promise no regulated business should sign.

What Gartner predicted#

The source is a Gartner press release dated 5 March 2025, titled in substance "Agentic AI will autonomously resolve 80% of common customer service issues without human intervention by 2029". The release also forecasts that the shift will lead to a 30% reduction in operational costs. Unlike most of the studies we read in this series, it is not a survey result. It is Gartner's prediction about the state of customer service four years after publication.

The prediction at a glance
ItemWhat the release states
PublisherGartner, press release
Date5 March 2025
TypeForecast (prediction for 2029), not a survey finding
HeadlineAgentic AI to resolve 80% of common customer service issues without human intervention by 2029
Economic claim30% reduction in operational costs, forecast alongside the 80%
ScopeCommon issues, not all customer contacts

Source: Gartner press release, 5 March 2025, as summarised in our research file. The cost figure is part of the forecast, not a measured saving.

Reading the small print#

Four elements of the prediction deserve more attention than the headline number.

  • Common. The 80% applies to common service issues, not to all conversations. Routine questions and simple transactions are the natural candidates. Disputes, hardship, complaints, bereavement and anything requiring judgement are, by definition, less common and harder.
  • Autonomously. Gartner describes resolution without human intervention. In a regulated process, "no human in the conversation" is not the same as "no human in the process": someone still defines the rules, approves the content and owns the exceptions.
  • By 2029. This is a date in the future. It says nothing about what share of issues is resolved this way today, in any sector or country.
  • 30% lower costs. The cost reduction is predicted together with the 80%, not observed, and it should never be borrowed as a saving for any specific supplier or process, including ours.

What the prediction gets right#

The direction is plausible, and it matches what we have argued for years. In May 2020 our CEO told ITwiz that automated voice would carry out the whole process, not a fragment of it. In our view, that holds for selected areas, not for everything. The point of agentic AI, in Gartner's sense, is the move from answering a question to completing a task: confirming an appointment and updating the calendar, recording a promise to pay and passing the status to the bank, checking eligibility and booking a slot.

That move is real, and it depends far more on integration than on the model. A conversation that understands the customer perfectly but cannot write the outcome back into the client's systems has resolved nothing. In our managed service, integration covers REST APIs, SFTP flat files, HTTPS exchange, VPNs and the telephony itself, and it is often the longest part of a launch.

Resolution also has to survive contact with real customers over time. Speech changes by region and age group, customers find phrasings no designer anticipated, and the client's own rules change. In every project our people listen to selected calls every day, especially unusual ones, and one approved change applies to every live conversation at once. An agent that resolves common issues in 2029 will need the same kind of operating discipline, whoever runs it. The forecast describes what the technology could do; it does not describe who will keep it doing that, week after week.

Using a forecast in a business case#

Forecasts like this one end up in board papers and procurement documents, often stripped of their qualifiers. The safer practice is to treat the 80% as context and build the business case from the bottom up. Start with your own contact reasons and volumes. Separate the issues that are genuinely routine from those that carry regulatory, financial or reputational weight. Then estimate, process by process, what share can be completed in one automated conversation and what must be routed elsewhere.

The same applies to the cost claim. A forecast 30% reduction across an industry tells you little about your own cost structure, your contracts, your telephony or the compliance work an automated process adds. Our rule is simple: compare against a comparator on the same buckets and portfolios, measure the incremental effect, and never present an analyst's expected saving as a result. In collections, for example, we say that Vee does not deliver 70%; it delivers the improvement over what would have happened anyway.

Why regulated contact will not look like the average#

In banking, insurance, healthcare and utilities, the uncommon issues are often the ones that matter most. A debtor who mentions financial hardship, a customer who withdraws consent to recording, a relative reporting a death, a dispute with legal implications, a request for erasure of data: each of these needs a defined route to a person, a stop or a referral. They are not failures of automation. They are part of a responsible design.

That is why we never describe our service as running without people. Our conversations are executed by Vee's AI, and Vee employs no call agents, but every process includes agreed transfers to the client's own staff. In regulated processes the replies come from a closed, versioned library, there is no learning in production, and every call can be reconstructed from its recording, logs, statuses and script version. Our technology posture is summarised in one line: deterministic where it matters, generative where it helps.

Building a voicebot is not the same as serving a customer. Between the two lies a very long road. – Dawid Wójcicki, Founder & CEO of Vee

Gartner itself points in a similar direction elsewhere. In a separate survey of 3,566 customers published in August 2026, 87% said companies using generative AI for service must provide access to a human agent. And in January 2026 Gartner predicted that generative AI cost per resolution would exceed $3 by 2030, above the cost of many B2C offshore human agents. Both are worth holding next to the 80%: high automation of common issues is compatible with a designed human route, and resolution by AI is not automatically cheap.

What we see in production#

We have run automated telephone contact since 2012 and in production at scale since 2017. To 31 October 2026, our AI has handled 48,626,959 cases (cumulative, Vee operating data; a case is a record assigned for handling, not a person or a call). Much of that is outbound, which is different from the inbound service issues Gartner's prediction mainly describes: reminders in early collections, appointment confirmations, renewals, surveys, lead qualification.

Where a process is narrow and well defined, a large share of cases can be completed in a single conversation. In soft collections for an insurer we measured a 61.9% first contact closure rate, and in inbound FAQ handling for a mobile-payments service a 73.5% first call resolution rate (both historical, one process each). Those are not 80%, they are not common issues across an industry, and they should not be compared with Gartner's forecast. They show what a specific, well-run process achieves, which is the only level at which such a number means anything.

61.9%First contact closure rateShare of cases closed in the first contact.Soft collections, insurance, historical. One process; not comparable with Gartner's 2029 forecast.

Questions to ask a supplier who quotes the 80%#

  1. Which issues in our process count as common, and what share of our volume do they represent today?
  2. What happens to the cases that are not common: who receives them, through which route, with what context?
  3. Which outcome is written back into our systems, and how is it verified?
  4. Is the cost figure you cite a forecast, a survey expectation or a measured result in a comparable process?
  5. Can every automated decision be reconstructed for an audit or a complaint?

The 80% forecast is a reasonable signal that automation will take on more of the routine work. The decision about your own process should rest on your own data and the cases that are not common. That is where we start every diagnostic.

Follow The First Second

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

Follow via RSS