Two-Thirds Report AI Wins, Almost None Wake the CEO
Azeem Azhar polled roughly 160 IT vice presidents this summer. Two-thirds pointed to measurable AI results. Only about eight said the results were significant enough to interrupt the CEO's holiday.
Two-Thirds Report AI Wins, Almost None Wake the CEO
Azeem Azhar polled roughly 160 IT vice presidents this summer. Two-thirds pointed to measurable AI results. Only about eight said the results were significant enough to interrupt the CEO’s holiday. That gap — 152 quiet wins to 8 loud ones — is the real story of enterprise AI in 2026.
The number matters because capex does not care about anecdotes. Hyperscalers have guided to record data-centre spending in 2025 (European Commission AI approach). The Stanford AI Index 2025 puts private AI investment at $252 billion for 2024, up sharply year-on-year. Meanwhile the curve of measurable value inside enterprises is looking suspiciously flat.
The overstatement incentive
BCG’s 2025 survey, cited in the same report, found that 70% of chief executives feel personally judged on AI outcomes. Give a leader a scorecard and they will report the score. The Azhar dataset says the score most IT VPs would defend under oath — the “call the CEO on holiday” threshold — sits closer to 5% than 67%.
That is not fraud. It is calibration drift. A retrieval-augmented chatbot cutting eight minutes off a service ticket is a real result. It is not board-material until you multiply it by volume, sustain it for two quarters, and net out the model bill.
Boards need a validation gate, not another dashboard.
ai_roi_gate:
threshold_material: "revenue_or_cost_shift >= 0.5% EBITDA"
min_observation_window_days: 180
auditor: finance_controller # not the AI product owner
net_of:
- inference_cost
- migration_and_retraining
- regulatory_risk_reserve
reject_if_source: self_reported_by_project_teamWhat “significant” actually looks like
The NIST AI Risk Management Framework is explicit: performance claims must be tied to measurable, contestable operational metrics. The EU AI Act, phasing in through 2027, requires providers of high-risk systems to log post-market performance — the same evidence a serious CEO would want before cancelling a holiday.
A useful mental model for boards evaluating AI wins:
Our consulting practice applies these three gates to every executive AI review we run for mid-market boards, and we mirror the same signals inside our agentic architecture work so the ROI conversation stays grounded in production telemetry, not slide-ware.
The CAI position
Eight out of 160 is not a failure of AI. It is a failure of measurement discipline. The MIT Sloan Management Review’s work on organisational AI learning is consistent on one point: firms that separate the reporter from the auditor sustain the wins that survive board scrutiny. That is the fix, and it costs nothing to implement.
Book a 30-minute governance review via our consulting desk and we will tell you which of your current AI wins would survive an independent audit.