Three pieces of research landed on the same day, August 17, 2026, and read together they describe a single problem. Enterprise AI adoption is close to universal. Measured cost reduction is at zero. And the cost of running a single agentic workflow is forecast to rise more than fivefold over the next two years. The constraint is not model quality. It is that most organizations cannot say where AI is being used inside their own operation, which means they cannot say what it changed, which means there is nothing to take to the finance function.
Why is AI adoption not producing cost reduction?
The sharpest number comes from TTEC Digital, which published research in partnership with CX Dive titled "The great CX reset: Why outdated operating models are stalling AI's payoff." Across 150 CX, contact center and IT leaders, not one respondent reported achieving cost reductions through AI. Two-thirds said operational costs had gone up instead.
State the limit before the headline gets away from us. TTEC Digital sells CX consulting and AI implementation, and the report is a gated download that argues for the services it sells. The sample is 150 leaders inside CX and IT functions, and the release discloses no fielding window or sampling method. It is good evidence of direction and shape, not of a precise population estimate.
Even discounted, a zero is a striking result. Not a small number. Not a disappointing number. Zero.
What the study points at
TTEC attributes the gap to bolting advanced AI infrastructure onto rigid, pre-AI operating models, and reports that only 1 percent of executives describe their current operating model as highly adaptive and built for continuous change. That framing is correct as far as it goes. Underneath it sits a measurement finding, and that is the one worth reading twice.
What can organizations not actually see?
Only 43 percent of respondents express high confidence that they can clearly account for where AI is being used across the customer journey. Fewer than half. In a function where AI deployment is the stated priority.
Seven platforms and one blind spot
Sixty percent of organizations in the study run seven or more distinct platforms. Leaders describe those systems as technically connected, and the study notes this produces ungoverned automations and redundant tools.
This is the pattern behind almost every workforce data problem we hear described. A CFO at a private-equity-backed IT services holding company told us that reconciling the tools each acquired agency kept running took three days per cash flow forecast. Those organizations do not lack systems. They lack a layer above the systems. AI did not create that problem. It made it expensive.
Governance that exists on paper
Data quality is no longer the binding constraint. Only 15 percent cite unreliable data. Governance is the constraint. Only 29 percent use a formal, cross-functional governance process consistently, while 64 percent apply policies inconsistently across the enterprise.
#### Connected is not the same as observable
A system that exchanges data is connected. A system that produces a durable record of what work was performed, by whom, at what cost, and what changed after a capability arrived, is observable. Most enterprise stacks are the first and not the second. Integration was built to move records between applications, not to answer a question about output. That is why seven connected platforms and no account of where AI is running are compatible statements rather than contradictory ones.
Is the cost of AI about to rise faster than the return?
On the same day, Gartner published a prediction that AI inference costs per agentic workflow will increase more than fivefold through 2028. The mechanism is not price inflation. Tokens, in Gartner's words, are becoming more cost-efficient. The mechanism is that cheaper tokens make more sophisticated workflows affordable, and routing a task to an agentic reasoning model raises provider inference costs by at least five times versus a basic chatbot interaction, often much more as task complexity grows.
Gartner names this the Inference Paradox, defined as better unit economics escalating the overall cost of AI without providing a clear pathway to commensurate and predictable value. Will Sommer, Sr. Director Analyst at Gartner, put it this way: "Product leaders cannot rely on more efficient token economics to rationalize AI costs." He adds that "each successive generation of AI capability will necessitate more, and often more expensive, tokens."
This is a forecast, not a measurement, and no survey sits behind it. Read as a forecast, it still changes the stakes. Costs have already risen with nothing measured on the return side, and the input cost trajectory is up.
Who owns the number?
The third item is the one most likely to be missed, and it is the most useful. In a Q&A published the same day, Gartner VP Analyst Mark Whittle argued that AI work redesign now sits between HR and IT rather than inside either, and predicted that by 2029, 30 percent of organizations will form blended HR-IT teams to accelerate AI enablement.
Among the capabilities Gartner says CHROs and CIOs must intentionally co-own, two are measurement capabilities. The first is "total cost of work analysis," defined as determining the true cost of work that integrates the cost and value created by both humans and AI agents. The second is human-AI performance management. Whittle names the consequence of leaving them unowned: "CHROs may be left managing workforce consequences from AI deployments they did not help shape, while CIOs may become accidental owners of workforce challenges created by AI."
That is an analyst firm describing, without naming it, the gap a Human Capital Operating System exists to close. Total cost of work is not an HR metric or a finance metric. It is both, and it does not resolve inside a system that was built for one of them.
How the three findings compare
| Source | What it actually measures | Method and sample | What it does not establish |
|---|---|---|---|
| TTEC Digital and CX Dive, Aug 17 2026 | Self-reported achievement of AI cost reduction and confidence in AI visibility | 150 CX, contact center and IT leaders. Vendor-sponsored. No fielding window disclosed | That the zero result generalizes beyond CX and IT functions, or beyond this sample |
| Gartner Inference Paradox, Aug 17 2026 | Forecast direction of provider inference cost per agentic workflow | Analyst prediction. No survey or sample size disclosed | Any specific organization's realized AI cost trajectory |
| Gartner CHRO-CIO Q&A, Aug 17 2026 | Where accountability for AI work redesign currently sits | Analyst Q&A. No survey or sample size disclosed | That blended teams improve measured return, which is untested |
| Forrester 2026 Predictions, Oct 28 2025 | Expected deferral of planned AI spend and ability to tie AI to financial growth | Analyst prediction | That deferral reflects poor AI performance rather than poor AI measurement |
What a defensible AI cost reduction number requires
Forrester predicted in October 2025 that enterprises would defer a quarter of planned AI spend into 2027, noting that fewer than one-third of decision-makers can tie AI value to their organization's financial growth. Ten months later, TTEC's zero is what that looks like when the budget review arrives.
Measure the work, not the tool
The evidence that settles an AI cost question does not live in a license ledger or a vendor dashboard. It lives in the systems where output lands: tickets, code repositories, CRM records, support queues. It shows up as change in throughput, cycle time, rework, and the composition of what remains.
A defensible estimate compares teams that adopted a capability against comparable teams that did not, matched on tenure, role mix, work type and tool stack, with a confidence score attached and the limits disclosed. That is controlled cohort analysis with confidence scoring, and it produces a number a CFO can defend precisely because it declines to claim more than it observed. It also does not require watching individuals: aggregated team views require five or more people, and individual signals flow only to a direct manager.
That is what the AI Impact signal family is built to produce, and it is why tool utilization and license waste sit in the same product as the workforce view rather than in a separate procurement report. When 60 percent of organizations run seven or more platforms, the spend question and the work question are the same question. Our measurement approach states the limits in public, including what it does not claim.
Frequently asked questions
Why is AI adoption not producing cost reduction? Because most organizations cannot locate the work AI changed. TTEC Digital's August 2026 study of 150 CX and IT leaders found zero cost reductions, two-thirds seeing costs rise, and only 43 percent confident they could say where AI was being used. A cost cannot be removed from an operating model that cannot see where it moved.
What is the difference between AI time savings and AI cost reduction? Time savings is an input, cost reduction is an outcome, and the second does not follow automatically. Hours returned to an employee become financial return only when the operating model absorbs volume without adding headcount, retires a tool, or shortens a cycle that carries a cost. Otherwise the hours are reabsorbed and the ledger does not move.
Are AI costs expected to rise or fall? Both. Gartner predicts inference costs per agentic workflow will rise more than fivefold through 2028 even as token economics improve, because more capable workflows consume far more tokens. Gartner calls this the Inference Paradox. It is a forecast with no disclosed sample, not a measurement.
Who owns the AI cost of work inside a company? Usually no one. Gartner names "total cost of work analysis" and human-AI performance management as capabilities CHROs and CIOs must co-own, and predicts 30 percent of organizations will form blended HR-IT teams by 2029. Left unowned, Whittle says, CHROs may end up managing consequences of deployments they did not shape while CIOs may become accidental owners of workforce problems.
The next step
The zero in this study is not a verdict on AI. It is a verdict on measurement. A number nobody can trace does not survive a budget review.
Request a Demo to see how Levos measures AI impact across your existing stack, or start with the human capital and financial workforce view if the cost of work is the question in front of you. Levos is currently onboarding design partners at 150 to 500 employees, expanding to 500 to 2,000 employees in the second half of 2026.
