Can You Measure the Return on Investment of Your AI?
Capgemini, IBM, Bain, Accenture, PwC, Deloitte, KPMG and McKinsey — each ran their own global study, separately, on their own respondents. And yet the number that keeps repeating across all eight is the same number: the gap.
IBM found the average enterprise-wide return on AI investment sits at 5.9% — below the typical 10% cost of capital — while only around 25% of AI initiatives deliver the ROI expected of them.
McKinsey found only 5.5% of organisations are seeing real financial returns from their AI spend, even though adoption is now close to universal.
Accenture found only 13% of organisations are creating significant enterprise-level value from generative AI.
KPMG found that while 74% say AI is delivering business value somewhere, only 24% can point to ROI across multiple use cases — a figure that actually fell from the year before.
Bain found only 15% of AI decision-makers can point to an EBITDA lift in the past twelve months, and fewer than a third can tie AI's value to a P&L line at all.
PwC found the top 20% of companies are capturing 74% of AI's entire economic value — the rest are splitting what's left.
Capgemini found agentic AI success rates have climbed from 5% to 14%.
Deloitte found a more optimistic 74% saying advanced initiatives meet or exceed expectations — though even they note scaling to strong ROI typically takes six to twelve months, longer than most boards are budgeting for.
Where do companies get stuck?
Most organisations can point to AI activity. Fewer can point to AI outcome, priced in dollars, tied to a use case a CFO would sign off twice. The firms breaking that pattern aren't running more pilots. They're running fewer, with a named owner, real governance, and a number attached before the budget clears.
Speaking to senior industry leaders, decision makers and Technology Experts, everyone is having exactly this conversation. I wanted to know whether the frustration I was hearing in the room matched what the data was actually saying. It does. So let's unpick the key challenges.
1. What is the predicted cost?
Most cost conversations start in the wrong place — a per-seat licence number, presented as if that's the whole bill. It never is. The real cost curve includes data readiness, token usage, integration, governance, organisational change management, and the ongoing cost of running the thing once it's live — the part IBM's research suggests most organisations are still underestimating, given how far the average return sits below the cost of capital.
Cost has to be forecasted at the use-case level, before you commit budget. Current cost, but also future predicted cost.
2. What are the use cases?
This is where the debate actually splits the room. PwC's research found the highest-performing companies weren't pointing AI at cost-cutting at all — they were pointing it at growth and business reinvention, and earning roughly 7.2 times the financial gain of their peers as a result. Accenture found the opposite instinct is far more common: efficiency-first thinking that caps the upside before it starts.
I think you need both, but you need to know which is which before you build the shortlist. A scored, prioritised list — value against feasibility against risk — beats an enthusiastic list of everything AI could theoretically touch.
3. What processes have been identified for optimisation?
Bain's research put this plainly: the model is the easy part. It's the data, the process, and the change management that actually determine whether value shows up. Automation bolted onto a broken process just makes the broken process faster.
Process discovery has to come before automation — decomposed, measured, and honestly assessed for whether it should be simplified or eliminated before anyone talks about which technology automates it. Decide which parts of the process can be fully AI automated, human assisted or need to remain human only.
4. How do I measure success, and where is the ROI coming from that funds the investment?
This is the question boards actually lose sleep over, and it's the one the eight reports answer most consistently. KPMG's research found the strongest outcomes weren't correlated with how much AI an organisation had deployed — they were correlated with governance discipline and clear accountability. Deloitte's research suggests boards should expect six to twelve months before strong ROI shows up, which is longer than most internal budget cycles allow for.
My recommendation is fund it in sequence. Let a small number of quick, well-governed wins generate the case — and the cash — that funds the transformational initiatives next. Measure outcome, not activity, and price every use case in the currency your CFO already uses. The good old Crawl / Walk / Run approach still applies.
5. What technology should I use — or is Copilot going to fix all my issues?
No single tool fixes an organisation that hasn't done the first four questions properly. PwC's research found the gap between leaders and laggards had little to do with how much AI technology was deployed — it came down to what the leaders pointed it at and how far they redesigned their workflows around it. KPMG found the same pattern: the strongest performers weren't the ones with the most AI, they were the ones with the clearest governance and cost visibility.
This whole conversation should be technology-agnostic. The platform matters far less than most vendors would like boards to believe. Buying the right tool without doing the strategy, process, architecture and governance work first is how organisations end up as one of the majority in these eight reports, not one of the minority.
Where this leaves the debate
AI is just another technology and to implement it, organisations need to follow the best practice technology transformation steps. Think about your overall business objective and role of AI, your key usescases, what process steps you want to automate, organisational change management, governance and control – next to selecting the right tools of course.
Next in this series, I want to take the other side of the ledger — not what AI costs to do properly, but what it costs to do nothing at all, including the uncomfortable thought experiment of what would actually happen to your business if AI adoption simply stopped tomorrow.
If you want this kind of thinking before it lands in the next board pack, subscribe to my CEO Update.
The best is yet to come.
— Anita Parer, CEO & Founder, Kyte Consulting
Sources
- Capgemini, Unlocking Business Value with Agentic AI — https://lnkd.in/eiUfpes6
- IBM, How to Maximize AI ROI in 2026 — https://lnkd.in/e7VQ4cDs
- Bain & Company, What to Expect from AI in 2026 — https://lnkd.in/e9XvyJXi
- Accenture, Making Reinvention Real with Gen AI — https://lnkd.in/eNqaxyZd
- PwC, 2026 AI Performance Study — https://lnkd.in/ecvcKF7M
- Deloitte, State of AI in the Enterprise 2026 — https://lnkd.in/e3ycBbsK
- KPMG, Global Tech Report 2026 — https://lnkd.in/eWjsiuGR
- McKinsey, The State of AI in 2025 — https://lnkd.in/eN9T2Z4w
Subscribe to my CEO Update
Short notes from me on what's actually changing in AI and Oracle delivery, written for leaders who don't have time for noise. No marketing. No promotion. Just the signal.