Gartner’s latest survey says knowledge foundations and human-AI team design separate the winners. Here’s what that means, and what it implies about the resolution rates the industry keeps promising.

Customer service has led every other function in AI adoption. It hasn’t led in AI returns.
Last week, at its Customer Service & Support Conference, Gartner released a survey of 275 service and support leaders conducted this summer. Only 4% said their AI initiatives had delivered the ROI they expected.
That number isn’t an outlier. Earlier this year, Gartner reported that service teams’ AI spending had grown 38% while their overall budgets grew just 2%. Across the AI use cases Gartner analyzed, roughly a quarter produced positive returns, a quarter produced negative returns, and the largest group had no clear ROI at all. Meanwhile, more than half of service leaders expect their own incentives to be tied to AI outcomes.
More money, more pressure, unclear results. Something in how support teams are approaching AI isn’t working.
Gartner calls the organizations getting it right “AI-forward” and describes three habits:
All three are worth reading closely. Three implications stand out for midmarket support teams.
The most striking number in Gartner’s findings isn’t 4%. It’s 1.8.
AI-forward organizations invest nearly 1.8 times more in data and knowledge management foundations than in the AI use cases themselves. Read that again: the teams getting returns spend more on the foundation than on the AI.
That matches what we’ve argued for a while: the model isn’t what makes AI work in CX; your knowledge base is. An AI agent can only answer as well as the knowledge it’s grounded in. When the knowledge base is thin, outdated, or scattered, the AI either escalates or guesses. Neither produces ROI.
The practical implication: a business case that treats knowledge cleanup as a one-time setup task is set up to disappoint. Knowledge needs an owner, a process, and tooling that finds gaps continuously. That’s why we built Knowledge Builder to capture gaps from tickets the AI couldn’t answer and draft articles from the answers your team already gives, and why Knowledge Manager is a first-class role in Flexivity.
Gartner’s third habit is the one most likely to be misread. Sladdin put it plainly: “Humans and AI are not fit for the same job.”
Gartner doesn’t use the word co-pilot. But think about what an AI-accelerated team looks like day to day. AI handles what it’s good at: reading every ticket, classifying it, summarizing long threads, finding the right article, drafting a reply, resolving routine requests outright. People handle what they’re good at: judgment, empathy, exceptions, and the conversations where a customer relationship is on the line.
The connective tissue between those two is agent assist. It’s where AI’s work reaches human judgment. When a ticket needs a person, does that person start from a twenty-five-message thread, or from a summary, the customer’s sentiment, matching knowledge, and a draft reply? That difference is where much of the value in a hybrid team lives.
Teams that treat AI only as a deflection tool skip this layer, and they leave the largest share of their tickets, the ones still handled by people, untouched by AI.
Gartner doesn’t say this directly, but its findings point to it: much of the ROI gap is an expectations gap.
Vendor marketing routinely cites autonomous resolution rates of 75% or higher. Those numbers are real, for specific intents. Password resets, order status, simple policy questions: AI can and should resolve a large share of these end to end. That’s good for customers, who get instant answers, and good for support teams.
But a 75% rate on your best-suited ticket types isn’t a 75% rate across your whole queue. Support volume includes billing disputes, bug reports, complex account changes, frustrated customers, and edge cases no knowledge base anticipates. Set an organization-wide target based on the best-case number and you’ve built a business case that was never going to close. Tie leaders’ incentives to it, as a majority now expect, and the 4% figure stops being surprising.
We’ve written before about what your resolution rate isn’t telling you. Gartner’s own guidance, to stop asking how to remove humans and ask where they add the most value, is a rejection of resolution rate as the headline measure.
None of this argues against ambitious AI. It argues for honest AI.
That’s the hybrid model we built Flexivity around: Progressive Autonomy to set how far the AI goes per class of tickets, an agent co-pilot inside Zendesk and osTicket for everything people handle, and Knowledge Builder to keep the foundation growing. Pricing is per conversation, so the business case doesn’t depend on hitting an inflated resolution number.
The 4% didn’t get there by automating the most. They got there by building the right foundation and putting people and AI where each does best. That’s a standard any support team can meet.