The AI Compass — practical AI for business leaders.
Ask a CFO in 2026 whether AI is central to their strategy, and the answer is overwhelmingly yes — 85% say so, and every single one surveyed plans new AI investment in the next six to twelve months, according to Coupa’s 2026 Strategic CFO Report, based on a survey of 600 senior finance leaders across North America, Europe, Japan, and Australia.
Ask the same CFOs whether they’re confident they can actually pull it off, and the number flips: 92% say they’re not sure they can execute on the strategy they just told you was central to the business.
That gap — full conviction, shaky execution — is what we are calling the CFO’s confidence trap. It isn’t unique to finance. But finance is where it shows up most starkly, because finance is one of the few functions where getting the sequencing wrong doesn’t just waste a budget line — it produces a bad number that ends up in a board deck, an audit, or a regulatory filing.
This article looks at why the gap exists, what’s actually driving the cancellations happening quietly behind the optimistic headlines, and what the CFOs who’ve closed the gap are doing differently.
The conviction is real, and so are the early wins.
Start with what’s genuinely working, because it’s nothing. Organisations further along in AI-driven financial planning report up to a 40% increase in forecast accuracy and speed, and monthly closing cycles are increasingly being replaced by continuous finance models that run scenarios in near real time.
Genpact’s CFO has described agentic accounts payable work — more accurate autonomous data capture, greater touchless processing, better cash visibility, and stronger supplier relationships, all while reducing cost. HPE’s CFO has said intelligent agents are already automating quarterly close, forecasting, and analysis, delivering real-time insight rather than a lagging monthly snapshot.
CFOs’ stated priorities for where they’re putting this to work are consistent and sensible: financial planning and forecasting, supplier risk assessment, spend analysis, and eliminating manual processes. None of this is hype. It’s the ordinary, compounding value of automation applied to a function that has always run on data.
Where the trap actually bites.
Here’s the number that should give every finance leader pause: 41% of CFOs believe agentic AI — systems that can plan and act with real autonomy, not just chatbots that answer questions — will deliver their greatest long-term financial returns. Only 13% have deployed it successfully. That’s not a small implementation lag. That’s two-thirds of the CFOs who believe most strongly in agentic AI’s potential currently unable to make it work.
They’re not alone, and the pattern isn’t unique to finance. Gartner made a stark, well-sourced prediction in mid-2025: more than 40% of agentic AI projects will be cancelled by the end of 2027. Crucially, Gartner’s analysts were explicit that the cause isn’t the technology falling short — it’s escalating costs, unclear business value, and inadequate risk controls.
A separate part of their analysis is worth sitting with: of the thousands of vendors now marketing “agentic AI,” Gartner estimates only around 130 actually deliver real autonomous capability. The rest are largely existing chatbots, assistants, and robotic process automation tools relabelled for the moment — what the industry has started calling “agent washing.”
The broader numbers back up how uneven the payoff has been. A Deloitte survey of roughly 1,800 executives across Europe and the Middle East found only 6% had achieved payback on their AI investment in under a year; most expected two to four years. And the pressure isn’t easing — a Kyndryl study found 61% of CEOs now feel more pressure to prove AI’s return on investment than they did twelve months ago. Boards want the payoff on the timeline the strategy deck implied, not the timeline reality is delivering.
In finance specifically, the gap between stated usage and real usage tells the same story. CFO Connect’s 2026 research found 56% of finance professionals now report using AI in their work, up sharply from 17% in 2023 — genuine, rapid adoption. But only 17% are actively using it inside core finance workflows like reconciliation, variance analysis, or forecasting. Most usage still sits in administrative tasks: drafting, summarising, basic research. Adoption is real. Transformation, so far, mostly isn’t.
Why finance gets caught in this particular trap.
Three things make finance more prone to this gap than most functions, not less.
1. The stakes of being wrong are asymmetric. A marketing team that ships a mediocre AI-generated campaign wastes a budget and a news cycle. A finance team that lets an ungoverned agent miscategorise spend, misstate a forecast, or process a payment incorrectly has produced an error with audit, compliance, and sometimes legal consequences.
That asymmetry rightly makes CFOs more cautious about scaling past the pilot stage — but it also means the gap between “we invested” and “we deployed successfully” shows up more visibly here than almost anywhere else in the business.
2. Most finance data isn’t ready for what’s being asked of it. A recurring theme across this year’s CFO commentary is sequencing. The widely echoed framing from Bain’s Finance Transformation research is blunt: AI-first finance transformation requires deliberate sequencing — data foundation first, automation second, intelligence third.
CFOs who invest in AI before fixing data quality get poor results; those who sequence correctly get compounding returns. Most organisations, understandably eager to show board-level AI progress, have been tempted to skip straight to the intelligence layer.
3. Legacy infrastructure resists exactly the kind of integration agentic AI needs. Deloitte’s CFO technology guidance for 2026 points to the same obstacle from a different angle: legacy systems, unclear data architecture, and immature governance are the core infrastructure blockers keeping agentic AI’s potential — Gartner projects it could handle 15% of everyday work decisions and be embedded in a third of enterprise software by 2028 — from actually reaching finance teams.
What the CFOs closing the gap are doing differently.
The pattern among finance leaders who are ahead isn’t that they moved faster. It’s that they moved with more discipline, on a narrower front.
1. They sequence deliberately. Data quality and architecture come before automation; automation comes before anything described as “intelligent.” Skipping straight to an agentic pilot on messy data is the single most common way CFOs end up in the 87% who haven’t deployed successfully despite believing in the technology.
2. They treat every deployed agent like an accountable hire, not a tool. Workday’s CFO, Zane Rowe, has framed this well: the leaders getting real value are moving beyond pilots and treating agentic systems as genuine team members that take on defined work and drive outcomes — which means each one needs an owner, a measurable scope, and, critically, an override switch. Gartner’s own analysis of why projects get cancelled makes the same point from the failure side: the agents most likely to survive past 2027 won’t be the most capable models — they’ll be the ones with a name attached to their accountability and a documented way to pull them back.
3. They pursue narrow, well-scoped use cases with clear ROI, not broad transformation bets. Gartner’s explicit recommendation for this stage of the market is to pursue agentic AI only where it delivers clear, demonstrable value — rethinking a workflow from the ground up where it’s warranted, rather than bolting an agent onto a legacy process and hoping. Vague enthusiasm for “agentic finance” as a category is exactly the pattern behind most of the cancellations Gartner is forecasting.
4. They budget and message a two-to-four-year payback window, not a two-to-four-quarter one. Setting that expectation with your board up front is cheaper than resetting it after a pilot underdelivers against an unrealistic timeline nobody actually believed in.
What this means for you.
You don’t need to be a CFO for this pattern to be relevant — the same conviction-execution gap shows up in marketing, HR, and operations, just with lower-stakes consequences when it goes wrong. A few things worth doing regardless of your function:
Separate your confidence in the strategy from your confidence in your readiness to execute it. They are different questions, and Coupa’s data suggests most leaders are currently answering only the first one honestly.
Before funding the next AI pilot, ask what it will be built on. If the honest answer involves inconsistent, siloed, or poorly governed data, that’s the actual project — not a prerequisite you can skip.
When a vendor uses the word “agentic,” ask for a demonstration of autonomous, multi-step reasoning on your own data, not a product tour. Gartner’s math on real versus rebranded vendors suggests a healthy scepticism here will save you a cancelled project later.
Give every AI deployment, agentic or not, a named owner and a defined way to shut it off. It’s the simplest governance step in every account of what separates the deployments that survive from the ones that get quietly cancelled.
The confidence isn’t the problem. Confidence that’s outrun the groundwork required to justify it is. The CFOs — and the leaders in any function — who close that gap in the next two years won’t be the ones who talked about AI most boldly. They’ll be the ones who did the unglamorous data and governance work first, and let their confidence catch up to their capability rather than the other way around.
Sources:
- Coupa, The Strategic CFO Report 2026 (survey of 600 senior finance leaders, North America, Europe, Japan, Australia)
- The CFO, “The AI-First CFO: Building High-Performance Finance Teams in 2026”
- Deloitte, CFO Guide to Tech Trends 2026 (citing Gartner and Deloitte’s Tech Trends 2026, December 2025)
- Houseblend, “AI Agents in Finance 2026: A CFO Guide to Reality vs Hype”
- CFO Connect, State of AI in Finance 2026: Adoption Trends, Tools, and CFO Roadmap
- ChatFin, “CFO Digital Transformation: AI-First Finance Roadmap for 2026” (citing Bain Finance Transformation 2026)
- Fortune / Yahoo Finance, “CFOs Predict How AI Will Continue to Shape Finance in 2026” (comments from CFOs at HPE, Workday, Genpact)
- Gartner, “Gartner Predicts Over 40% of Agentic AI Projects Will Be Cancelled by End of 2027” (press release, June 25, 2025)
- Forbes, “Why 40% of Agentic AI Projects May Be Cancelled By 2027” (Robert Szczerba, July 7, 2026), citing Deloitte EMEA executive survey and Kyndryl CEO research
As always at The AI Compass: every statistic above is sourced from a named, dated report — verify anything you plan to build a decision around, and treat single-source numbers with appropriate caution.
#AICompass #CFO #FinanceTransformation #CorporateFinance #AgenticAI
