Here’s a number that should make every CFO uncomfortable. A new 2026 survey — 200 senior executives across telecom, retail, and e-commerce — found that 90% of organizations increased AI investment over the past two years. Nine out of ten companies, significant money, all moving in one direction.
And then this: only 12% of those same organizations can rigorously measure what that investment actually returns.
That gap is not a marketing problem. It’s not a data science problem. The reason most companies can’t answer the board’s question — “what is our AI actually returning?” — is that the measurement infrastructure either lives in finance, or it doesn’t exist at all. Right now, most finance teams aren’t building it.
The Board Is Already Asking
The same survey found that 86% of marketing leaders have been asked by their board, CFO, or investors to present hard evidence of AI’s business impact on a monthly or quarterly basis. So 86% are being asked. And 12% can actually answer.
I’ve been seeing this play out firsthand in C-suite conversations. The questions are coming. And there’s a growing realization that a lot of current AI investment is being defended on vibes — not necessarily wrong, but very hard to hold up when a board starts pushing.
Only 16% of leaders feel very confident they can defend their AI budget with hard evidence. The remaining 84% are caught between mounting pressure and a lack of proof.
Why the Existing Tools Don’t Work
The standard frameworks finance teams reach for — ROI, ROMI, LTV/CAC, attribution models, marketing mix modeling — were all built on three assumptions. Investments are discrete (you can ring-fence them). Channels are separable (you can isolate contribution). Costs are visible (they show up in one place). AI breaks all three at once.
Think about it like electricity in a manufacturing facility. It powers everything simultaneously. No single department owns the full bill. And if you tried to calculate the ROI of electricity on one specific product line, you’d struggle — not because electricity isn’t valuable, but because the cost accounting to isolate its contribution was never built that way. That’s exactly where most companies are with AI right now.
ROI undercounts because AI infrastructure, data prep, and talent are scattered across department budgets. Attribution models over-credit AI by counting its influence multiple times across a customer journey. Marketing mix modeling runs on quarterly cycles — too slow and too aggregated to isolate AI-specific effects. Each framework has the same core problem: it was designed for a world where AI didn’t exist.
Four Structural Breakdowns — and What Finance Has to Do About Each
The survey identified four specific gaps blocking measurement. Each one has a direct finance implication.
Cost fragmentation (62% struggle here). For every $1 spent on AI model development, organizations need approximately $3 for surrounding infrastructure and change management. Most companies count the software license and miss the iceberg. Talent and integration costs — spanning HR, IT, and department budgets — leave the total cost base incomplete by an estimated 30–50%. Finance needs to own a complete cost map, not just the vendor contract.
Revenue attribution complexity (58%). When AI augments decisions across dozens of touchpoints simultaneously, isolating its contribution is genuinely hard. The organizations that have solved it — the 12% who can actually measure — consistently use one approach: controlled testing. An APAC telco ran a six-month randomized holdout test and proved their AI-optimized campaigns outperformed manual ones by 31% on primary conversion. That number held up because they built the control group to prove it. Holdout group design should be a standard part of any AI initiative approval process — the same way you require a business case before approving capital spend.
CX-to-revenue disconnect (55%). AI improves NPS. Satisfaction scores go up. But without attribution to retention, lifetime value, and bottom-line revenue, the initiative looks like overhead when budgets tighten — even when it’s quietly driving numbers that other teams take credit for. 57% of organizations cannot connect AI-driven satisfaction changes to revenue impact at all.
Governance and integration gaps (50%). Without a pre-AI performance baseline, you cannot measure incremental improvement. Without unified data pipelines connecting cost data to revenue data, you can’t calculate efficiency. The survey’s top performers committed 6–18 months to building connective tissue across finance, procurement, CRM, and cloud billing. This is infrastructure work — closer to a BI data layer project than a marketing dashboard.
Two Things Finance Has to Own
Every top-performing organization in this survey showed the same division of responsibility. Finance owns comprehensive cost capture — infrastructure, talent, integration, cloud billing, everything. Marketing or the business unit owns revenue attribution through controlled testing. That split is explicit, agreed upon, and reviewed quarterly.
If your organization is spending on AI and nobody in finance has formally taken ownership of building a complete cost map, that’s a gap with your name on it. Because when the board asks the CFO “what is our total AI spend and what are we getting for it,” the answer should be a number somebody in finance built.
The second thing is measurement methodology. Most companies track activity proxies — campaigns launched, models deployed, engagement rates. Only 21% can link AI spend to revenue at the campaign level. The difference is almost always controlled test design. Push for holdout groups. Push for propensity-matched controls. Push for pre-AI baselines documented before systems go live. These aren’t optional rigor. They’re the minimum bar for a number that will hold up in a board presentation.
Key Takeaways
- 90% invested, 12% can measure it. The gap is structural — traditional ROI frameworks were never built for always-on, multi-touchpoint AI systems.
- AI’s total cost is undercounted by 30–50%. Software licenses are visible. Talent, integration, and cloud infrastructure — spread across multiple budgets — are where the real hidden costs live.
- Controlled testing is the only defensible method. A/B holdout groups and propensity-matched controls are how top performers prove their AI works. Without a control group, you can’t isolate AI’s contribution.
- No pre-AI baseline means no measurement. Document performance benchmarks before systems go live — once AI is running, you’ve lost the counterfactual.
- The window is narrowing. Within two years, rigorous AI measurement will shift from competitive advantage to baseline expectation. Finance teams that build this infrastructure now will be the ones who can actually answer the question.
Want the CPE credit? Take the full lesson on EverydayCPE and earn 0.2 CPE credits: [lesson link]
Leave a Reply