Air IT Group’s AI Decision Gap research, covered on November 5, 2026, surveyed 504 decision-makers at UK organisations and found that 83% report at least some AI spending with no measurable return, or that they do not measure returns at all. The same study found that 41% purchased or introduced an AI tool before agreeing what business problem it should solve. On Basil C. Puglisi’s reading, those two numbers belong together. A tool bought before the problem has nowhere honest to post a return, and a board that accepts a two-year payoff story without a pre-written measure is reviewing the same gap at a larger scale.
RGP’s September 2026 CFO Perspectives survey of 400 CFOs and equivalent finance leaders at organisations with $250 million or more in revenue, released November 3, 2026, shows how that gap looks from the finance seat. While 52% of CFOs report positive AI returns today, just 10% describe them as significant. Within 24 months, 54% expect significant positive returns. Once a 24-month expectation reaches a board, it can harden into a delivery date, and the survey already names the capacity and integration barriers that sit between the promise and the ledger.
The Decision Gap starts before the purchase
Among Air IT respondents with at least one active AI initiative, 74% said not all initiatives had both an agreed business objective and a measure of success. Of those whose organisations formally track AI pilots, 62% abandoned at least one in the previous 12 months. Almost half, 47%, said AI made a task or process take longer once checking, correction, setup and management were included, and only 13% said their organisations consistently convert time saved through AI into measurable higher output or productivity.
Lee Johnson, Air IT Group’s chief technology officer, put the sequencing failure plainly: “If the problem and the outcome aren’t clear from the outset, it becomes much harder to know whether AI is actually delivering value.” The failure sits in decision design and shows up later as an ROI story no one can defend.
Puglisi’s Factics method was built for the same sequencing rule. Facts establish what is known, tactics establish what to do, and goals, outcomes and KPIs establish what happened because the action was taken. The test is timing. “If I cannot define the measurement before I take the action, the action is not real,” he writes on the Factics page, because a KPI written after results arrive is a narrative and a KPI written first is a commitment that can fail. Buying the tool before naming the problem inverts that order, which is why Air IT’s 83% figure should not surprise anyone who has watched an AI purchase outrun its brief.
The 24-month promise meets a capacity ceiling
RGP titles one of its findings “The 24-Month Promise: AI Expectations Exceed Current Returns.” The leading barriers to realizing AI value are integrating AI into existing systems (23%), scaling successful pilots (14%), and data quality or readiness (13%). Fewer than 1% cite funding as the primary barrier. Nearly half of CFOs, 49%, say their transformation agenda exceeds their organisation’s ability to execute, and among that group only 24% report positive AI returns, compared with 79% where capacity matches or exceeds transformation demands.
Authority and control split in the same direction. RGP found that 64% of CFOs report high accountability for AI adoption and just 18% report high control. Only 13% say major initiatives run through enterprise-wide governance with a single point of accountability, and 58% of that group say at least three-quarters of initiatives delivered on time, against 14% at siloed organisations. A 24-month ROI promise made without a named owner, a pre-written measure and enough capacity to change the surrounding work needs those gaps closed before finance should treat it as a capital plan.
How Puglisi would govern the buy and the promise
The client work starts before procurement. The owner writes the business problem in one sentence, the baseline that shows the problem today, the expected effect and the KPI that will prove or fail the claim. That packet is the Factics brief for the purchase, and no AI tool enters the stack until the brief exists.
Checkpoint-Based Governance, Puglisi’s framework for placing a named human decision at defined points in an AI workflow, then sits on two rulings. The first is the buy itself. A named owner accepts, modifies or rejects the purchase against the written problem and KPI. A decision to modify can shrink scope or delay the spend until the measure is clear. The second is the 24-month promise before it reaches the board. Finance reviews the projected return against the same baseline and records a decision to accept, modify or reject. A rejected projection stays out of board materials until capacity and ownership are named.
During the pilot, the same owner reviews the KPI on a fixed cadence. The owner assesses time savings against the approved business outcome and records a decision to modify or reject unsupported claims rather than celebrating the clock improvement as productivity. Abandoned pilots enter the same record, so the next purchase brief has to answer what failed last time.
What a client tracks
A short set of measures keeps the Decision Gap visible. The first is the share of AI purchases that carry a written problem, baseline and KPI before the order is placed. The second is the share of active initiatives that have both an agreed objective and a success measure, the gap Air IT found when 74% said not every initiative had both. The third is the share of board or investor AI ROI statements that finance has accepted under Checkpoint-Based Governance before publication. The fourth is the share of initiatives with claimed time savings that meet their pre-agreed business outcome, because Air IT found only 13% of respondents consistently convert saved time into higher output or productivity.
The promise needs a signature before the calendar
Air IT shows UK SME decision-makers still purchasing AI before the problem is named, and RGP shows CFOs still projecting significant returns two years out while capacity and control lag. Puglisi’s answer is a written problem and KPI before the buy, and an accept, modify or reject ruling on the 24-month promise before the board hears it. When those two checkpoints are missing, the Decision Gap and the 24-month promise are the same failure at two stages: before the purchase and before the board.
Sources
- Air IT Group. (2026). The AI Decision Gap [Research report page]. https://www.airit.co.uk/the-ai-decision-gap/
- IT Europa. (2026, November 3). UK firms face AI reality check as returns fail to match ambition. https://www.iteuropa.com/news/uk-firms-face-ai-reality-check-returns-fail-match-ambition
- Puglisi, B. C. (n.d.). Factics: The method behind valuable content and trusted AI. basilpuglisi.com. https://basilpuglisi.com/factics/
- RGP. (2026, November 3). RGP study: 89% of CFOs report positive financial outlook, yet nearly half say transformation demands exceed capacity [Press release]. https://rgp.com/press/rgp-study-89-of-cfos-report-positive-financial-outlook-yet-nearly-half-say-transformation-demands-exceed-capacity/
- RGP. (2026, October 22). Where AI transforms business economics. https://rgp.com/insights/where-ai-transforms-business-economics/
- Trim, N. (2026, November 5). Four in five UK businesses struggle to prove AI investment is paying off, Air IT Group research finds. Business Info Magazine. https://businessinfomag.uk/B2B-Technology/four-in-five-uk-businesses-struggle-to-prove-ai-investment-is-paying-off-air-it-group-research-finds/
Questions readers ask
What is the AI Decision Gap in Air IT Group’s research?
Air IT Group uses the term for the disconnect between investing in AI and having the evidence and clarity to determine whether it is delivering value. Its research of 504 UK organisation decision-makers found that 83% report at least some AI spending with no measurable return, or that they do not measure returns at all, and that 41% bought or introduced AI before agreeing the problem it should solve.
What did RGP’s CFO Perspectives survey find about AI ROI?
In RGP’s September 2026 survey of 400 CFOs at organisations with $250 million or more in revenue, 52% reported positive AI returns today but only 10% described them as significant, while 54% expect significant positive returns within 24 months. Integration, pilot scaling and data readiness ranked as the leading barriers, and funding almost never did.
Why does buying AI before naming the problem damage ROI?
Without an agreed problem, baseline and success measure, there is no honest place to post a return. Air IT found that 74% of respondents with an active initiative said not every initiative had both an agreed business objective and a measure of success, and 62% of those who track pilots abandoned at least one in twelve months.
How does Factics apply to an AI purchase?
Basil C. Puglisi’s Factics method requires a fact, a tactic and a KPI written before the action. Applied to AI buying, the owner writes the business problem, the baseline and the measurable outcome before the tool is purchased, so the result can fail the claim rather than rewrite it afterward.
What is Checkpoint-Based Governance for a 24-month AI promise?
Checkpoint-Based Governance is Basil C. Puglisi’s framework for placing a named human decision at defined points in an AI workflow, where the person accepts, modifies or rejects what is proposed. For a two-year ROI claim, finance rules on the projection before it reaches the board, using the same pre-written baseline and capacity constraints the pilot will face.
#AIg
Leave a Reply