Gartner Says CX Tech Spend Will Double by 2028. The Resolution Gap Explains Why.

In a 60-day window, Gartner published two predictions that any CX leader sizing AI spend over the next 24 months should read side by side.

On January 26, the firm forecast that cost per resolution for generative AI in customer service will exceed $3 by 2030, higher than many B2C offshore human agents. On March 31, the firm forecast that over half of customer service organizations will double their technology spend by 2028 without an equivalent reduction in headcount.

Read together, these forecasts describe a CX economy where the spend curve goes up, the headcount curve stays roughly flat, and the cost per AI resolution drifts toward the offshore agent line it was supposed to undercut. The story that AI makes customer service cheaper, the assumption underwriting most CX software budgets since 2023, is no longer holding cleanly.

Gartner names two variables behind the trend: talent and tooling complexity. Both are real, but there's a third variable underneath them, and it's the one most teams aren't tracking. That variable is resolution: specifically, the gap between what AI handles and what AI actually solves. It decides whether each new dollar of CX tech spend produces an outcome or just produces a transaction.

The resolution gap, in dollars

Most CX dashboards report "automation rate" or "deflection rate" as the headline number. A ticket comes in, the AI responds, the conversation ends without an agent, and the platform marks the ticket handled. The dashboard ticks up.

What the dashboard doesn't always show is whether the customer's problem actually got solved. That's the resolution gap: the space between interactions the AI touches and issues the AI fixes.

Volume doesn't disappear when the gap is wide. It re-enters the system in three ways:

• As re-contacts, when the customer comes back through the same channel because their problem persists
• As cross-channel escalation, when the customer tries chat, gives up, calls in, and reaches a human who starts from scratch
• As churn, when the customer concludes the company can't help and stops trying

Each of those outcomes carries a cost the deflection dashboard doesn't capture. The McKinsey AI in Customer Service 2026 sample puts the unit economics in sharp relief: an AI-handled resolution costs about $0.62, against $7.40 for a human-handled one (cited via Digital Applied's 2026 benchmark compilation). But that math only works at the resolution number, not the deflection number. Every interaction logged as deflected and then bounced back as a human-handled escalation means two bills: one for the AI license, and again for the agent picking up the dropped thread. When the gap is wide, that math compounds, and the spend curve Gartner is forecasting through 2028 is what it looks like at scale.

This is why cost per resolution is the benchmark that matters when comparing AI economics to the human baseline, not cost per interaction, not deflection rate. Gartner's $3-per-resolution forecast for 2030 is a measurement of what AI costs when it actually solves the problem, not what it costs when it handles a message.

What's actually happening inside the spend curve

Gartner's October 2025 survey of 321 customer service leaders found that only 20% of organizations had reduced agent headcount because of AI. The other 80% are paying for the AI and paying for the agents.

That's not a transitional state on the way to lower costs. It's the structural result of how AI is being deployed in most CX functions. As CXOVoice put it, customer service costs are rising because AI is adding layers rather than removing them, the stack now includes conversational AI, routing, knowledge, analytics, orchestration, and oversight, each with its own licensing, implementation, and integration costs.

None of those layers retires an existing layer. What Gartner is forecasting through 2028 isn't a one-time migration cost, it's a structural shift from payroll-heavy CX budgets to software-heavy CX budgets, with the total cost base climbing rather than falling.

Inside that shift, the resolution gap decides which side of the ledger AI spend lands on. When AI resolves the issue end-to-end, the software layer earns its place: it does work that would otherwise require a human, and the unit economics work. When AI deflects without resolving, the software layer adds cost without retiring any of the existing cost, and the dollar that was supposed to come out of the agent budget ends up funding the AI on top of it.

The compounding math

Take a typical setup: 10,000 inbound tickets a month, a vendor dashboard reporting 60% automation rate, and 6,000 tickets marked "handled" without a human.

The audit work that distinguishes deflection from confirmed resolution typically lands somewhere different. Industry research on RAG-based deployments consistently finds 15% to 25% of "deflected" tickets contain wrong or incomplete answers. If 25% of those 6,000 deflected tickets re-enter the system within 72 hours (as re-contacts, escalations, or cross-channel attempts) then 1,500 interactions that looked automated weren't.

The true autonomous resolution rate on this volume isn't 60%. It's 45%. And the gap shows up in three line items most CX dashboards aren't wired to surface:

• The AI vendor's bill is unchanged. The license is priced on interactions, and the AI handles the interactions.
• The agent team's workload includes those 1,500 re-entered tickets, often without context, because the handoff is broken.
• The customer experience absorbs the cost of asking twice. The vendor charges for 6,000 interactions. You paid for 4,500 resolutions.

The resolution gap is what decides whether the AI line item is a substitute for human handling or an addition to it. When the gap is small, AI spend offsets agent spend. When the gap is wide, AI spend stacks on top of agent spend, and that's the curve Gartner is forecasting through 2028.

The architecture variable

The resolution gap isn't a model problem. It's an architecture problem.

Most AI agents in production today can read but can't act. They retrieve information from a knowledge base, summarize policy, and suggest next steps. What they generally can't do, without a deeper integration, is process the refund, update the account, modify the order, or change the appointment. The architectural dividing line is read-only versus read-write access to the systems where the customer's actual problem lives.

A read-only AI is structurally capped at the share of tickets that can be resolved by surfacing information. For most enterprise CX programs that's a tier-1 band of password resets, order status lookups, and policy questions, real volume, but a ceiling. The complaints, exceptions, billing disputes, and account changes that drive support costs sit on the other side of the read-write boundary.

This is why "automation rate" numbers above 70% on broad query sets are worth probing. They usually reflect either an aggressive triage layer that excludes the hard tail before the AI sees it, or a definition of resolution that counts conversation completions without confirming the outcome. Production deployments with read-write integration to backend systems tend to land in the 70% to 85% confirmed resolution band on broad query sets that include policy-dependent tickets.

The architectural decision is what determines whether each new dollar of CX tech spend closes the resolution gap or widens it. Read-only versus read-write, overlay versus rip-and-replace, AI as a layer on top versus AI inside the helpdesk, these aren't implementation details. They're the spend multiplier. Maven's evaluation framework walks through the dimensions in more depth.

What resolution-first looks like in production

The companies closing the resolution gap share one architectural trait: read-write integration to the systems where customer problems actually live. The production data that closes the resolution gap looks specific:

• Mastermind reached 93% autonomous resolution within six weeks of deployment.
• Papaya Pay reached 90% autonomous resolution in three weeks.
• Clio moved from a legacy chatbot to autonomous resolution at 80%, resolved 60% more tickets, and shortened live support response by 4x.
• Rho held 95% CSAT while handling 12% more monthly contacts without adding headcount.

That last data point is worth sitting with. Gartner's March 31 release predicts that the doubling of CX tech spend through 2028 won't be matched by a reduction in headcount, and warns that companies cutting agents to fund AI will risk operational disruption, degraded customer experience, and expensive rollbacks. The forecast assumes AI doesn't resolve enough on its own to handle growing volume.

Rho's number is the clearest illustration of resolution-first spend math in production. CSAT held at 95%, contact volume grew 12%, and the team didn't grow. That's what volume capacity looks like when the resolution rate is real: more customers handled, same team, CSAT intact.

That's the version of the spend math that makes the AI line item earn its place. Tech spend goes up, agent spend stays flat, customer volume goes up, and the total cost base lands lower per resolved issue. It only works when the resolution number is high enough. Which is to say, when the gap between deflection and resolution is small enough that AI volume converts cleanly into AI outcomes.

Three questions before you double the budget

Gartner's 2028 spend forecast doesn't have to land on every CX P&L. The variable that decides whether it does is the architecture and economics of the AI layer being added. Three questions cut through the headline numbers any vendor will lead with.

1. Can the AI take actions in the systems where the customer's problem lives, or only respond from documents? The read-only versus read-write line is the architectural variable that caps how much of the support cost base the AI can address. An AI that surfaces order status is useful. An AI that processes the return, updates the account, and confirms the change is what closes the resolution gap.
2. What's the 72-hour re-contact rate on resolved tickets? This is the audit metric. If 25% of "resolved" tickets generate a follow-up within three days, the effective resolution rate is materially lower than the headline. A vendor confident in the resolution quality of their AI publishes this number. A vendor that hasn't measured it usually has a reason.
3. What's the confirmed resolution rate, not the deflection rate? A confirmed resolution counts only tickets where the customer's problem was solved and the customer didn't come back. Deflection counts any ticket the AI touches. Vendors that have measured both will share both. Vendors that haven't, or won't, are typically optimizing for the number that flatters them.

The three answers together determine whether every new dollar of CX tech spend buys resolution or buys activity.

The variable Gartner didn't name

The two 2026 forecasts pull in the same direction. Spend goes up, cost per AI resolution climbs toward the human floor, and headcount stays roughly flat. The combined effect is a structural increase in CX cost base for the half of organizations doubling their tech spend through 2028.

What the forecasts don't say, because they're measuring tech spend, not architecture, is that the resolution gap is the variable that decides which half a given organization lands in. The companies that treat resolution as the metric of record, and architect their AI layer to act in the systems where customer problems actually live, won't see the spend curve double without outcomes. The companies that treat resolution as a synonym for deflection will spend more in 2026, 2027, and 2028, and resolve about as much as they did in 2024.

The category language is shifting to reflect this. Autonomous resolution is becoming the metric of record. Resolution Rate is on track to replace older CX benchmarks as the primary measure of whether AI in support is working. Vendor and analyst commentary is consolidating around the same point: in 2026, the dashboard your AI vendor leads with is the one you should probe hardest.

The 2028 spend curve Gartner is forecasting isn't inevitable. It's the consequence of architecture decisions most CX teams are making right now.