Senthil Info

automating business processes, honestly

What Gartner Actually Said

Gartner is the most-cited source in this field and the most loosely quoted. Three 2026 findings do most of the work in current discussion, and each is narrower than its retelling.

Reviewed August 9, 2026. Gartner sells research to buyers rather than tools; it also has commercial relationships across the vendor market. Treat as a serious source with a commercial position. For a product-side reference related to workforce operations, Monitask also covers cognitive offloading.

Finding one: the delivery rate

April 2026. Only 28% of AI projects in IT infrastructure and operations deliver the promised return. Roughly one in five fails outright. For broader background and an independent point of comparison, see ACM.

Three things about the scope.

It is IT infrastructure and operations specifically, not business process automation generally. Applying it to an invoice-processing project is an extrapolation.

The two figures are compatible. 28% delivering and 20% failing outright leaves about half partially delivering — which is where most projects actually land, and the half nobody quotes.

And "promised return" is the bar. Not project completion. A project that finished, works, and did not hit its business case counts against the 28%.

Finding two: what separates them

The most useful and the least quoted, because it is not a number.

Successful projects share two properties: integration into existing workflows rather than running as a parallel process, and full executive support before and during the rollout rather than only at launch.

Failed projects share one dominant cause: misaligned expectations — leadership assuming the technology would immediately automate complex tasks or deliver cost reductions on an unrealistic timeline.

Gartner's own framing is direct about the implication: neither success factor requires a different vendor.

Finding three: headcount

May 2026. A survey of 350 executives at companies above $1 billion in revenue, on organisations piloting or deploying autonomous AI. Around 80% reported workforce reduction — and the reduction rate was nearly identical between companies reporting high ROI and companies reporting flat or negative ROI.

The observation is that staff numbers fell. The finding is that they fell regardless of whether the automation worked.

That breaks the inference, not the observation, and it is the single most useful thing anyone has published for a board conversation about automation returns.

How these get misquoted

"Gartner says 80% of AI projects fail." Conflating the RAND figure, the Gartner infrastructure figure and the headcount survey into one number that none of them states.

"Only 28% of automation projects succeed." Wrong subject — infrastructure and operations, not automation generally — and wrong bar, since it measures promised return rather than success.

"Gartner says AI causes layoffs." The inverse of what the May survey found.

And any of the three without a date. This field moves quarterly, and a 2024 Gartner figure is a different claim from a 2026 one.

Using them properly

For a board conversation: the 28% figure with the subject stated, or the headcount finding, which is more surprising and harder to argue with.

For a project decision: the success and failure factors, because they are actionable and the percentages are not.

And always with the year and the population. "Gartner, April 2026, on AI in IT infrastructure and operations" is a citation. "Gartner says" is not.

The limit worth stating

All three describe large enterprises. There is no equivalent data for smaller organisations, and the figures are quoted at them constantly.

The causes transfer. The rates do not.

The short version