Only 22% of organisations have scaled AI across multiple business units or adopted an AI-first approach, according to a Gartner survey released on 1 September 2026. The survey covered 1,303 respondents at organisations with at least $50 million in annual revenue, and was fielded between January and April 2026.
The finding sits awkwardly beside the spending numbers. Eighty-five percent of functional leaders said they plan to increase AI spending in 2026, and they devoted an average of 12% of their budgets to AI in 2025, DIGIT reported. About 11% of organisations did not know what their own function had spent on AI.
/ 01What the survey found
The sharpest contrast is between high and low performers. High performers reported positive returns on 81% of their AI initiatives. Low performers did not know the rate of return on 29% of theirs, TechInformed reported. The difference was less about which tools they bought than about how they managed them: high performers track returns continuously, run AI initiatives as a portfolio, and regularly move resources away from, or stop, projects that underperform.
The survey also compared what organisations pursue with what pays off. The most pursued use cases were cybersecurity threat detection and response (54%), IT service desk automation (54%) and automated code generation and refactoring (44%). The use cases most often reporting positive returns were different: intelligent IT asset and cost optimisation (40%), synthetic data generation (28%) and code generation (23%). Code generation was the only use case in both top threes. Productivity dominated intentions, with 75% of functional leaders naming it as a target and about 30% of functional AI spend going towards it.
Tina Nunno, a distinguished vice president and Gartner Fellow, said that “CEOs and CIOs must pinpoint which AI use cases truly drive positive financial returns,” and that the lack of financial visibility “heightens risk as spending accelerates.” TechInformed notes that Gartner’s release does not show that the high performers’ practices caused their returns; the survey reports correlation.
/ 02Why it matters
The headline number is not surprising to anyone who has watched enterprise AI programmes over the past three years. What is useful is the detail. The problem the survey describes is not that AI fails to work. It is that most organisations cannot tell where it works, because they cannot connect a project’s cost to a measurable outcome.
That points to a structural issue. Pilots are typically built by one team, on one dataset, with one model, and measured by enthusiasm. Scaling across business units requires the same capability to work on other teams’ data, under other teams’ definitions, with results that a finance function can check. Each step exposes the fact that “customer”, “order” or “incident” means slightly different things in different systems, and that nobody wrote down which meaning the pilot assumed.
/ 03Measurement needs a model of the business
Gartner’s high performers measure returns continuously. That is only possible if the organisation can trace an AI output to a business outcome: which decision it informed, which process it changed, which cost it moved. In most enterprises those links live in people’s heads and slide decks.
An ontology-driven knowledge graph makes those links explicit. When initiatives, systems, data sources, processes and outcome measures are modelled as entities with defined relationships, portfolio questions become answerable: which AI initiatives touch the procurement process, which data they depend on, and what changed after they went live. When the same graph also records where each AI answer came from, a failed result can be diagnosed rather than simply abandoned.
There is a cost angle too. A model that has to reconstruct business context from raw text on every request uses more compute than one that is handed structured context. Moving meaning into a governed semantic layer, rather than asking larger models to infer it repeatedly, is one way to make the returns side of Gartner’s equation easier to show.
/ 04What to do
- Count the spend. If your function is in the 11% that cannot say what it spent on AI, start there.
- Tie every initiative to an outcome measure before it leaves pilot, and record the link somewhere other than a slide.
- Stop things. Gartner’s leaders reallocate from underperforming projects. That requires knowing which ones they are.
- Fix shared definitions early. The second business unit is where hidden differences in meaning surface. A shared ontology is cheaper before the rollout than after.
Gartner’s low performers do not have a model problem so much as a bookkeeping problem: spend, systems and outcomes are recorded in different places, and nothing joins them.
Put the AI portfolio in the same graph as the business it serves. Each initiative links to the processes it changes, the datasets it reads, the cost centre that funds it and the metric it was meant to move, all typed against one agreed definition of order, customer or incident. Questions a finance lead actually asks then come back with evidence: which initiatives depend on the procurement dataset, what cycle time did after the service-desk agent went live, which pilots share a data source and could be merged. When a project underperforms, the trace shows whether the model, the data or a definition mismatch was at fault, so stopping it is a decision rather than a hunch. Structured context also lets smaller models carry the load, which is where our figure of up to 50% lower GPU spend comes from.
How we’d model it
AI initiative—changes→Business processAI initiative—fundedBy→Cost centreAI initiative—measuredBy→Outcome metricAI initiative—dependsOn→Dataset
- Less Than A Quarter of Orgs Have Scaled AI Across Their Business — DIGIT digit.fyi ↗
- Gartner finds only 22% have successfully scaled AI — TechInformed techinformed.com ↗
See how Aktiver turns raw data into knowledge graphs your AI can reason over — deterministic, traceable, behind your firewall.