"What's the ROI?" It's the first question boards ask about AI investment, and the framework they use to answer it is almost always wrong. Not because the numbers are wrong, but because the question being measured is wrong. Most AI ROI frameworks measure cost savings. They should be measuring capability gain.
What You Need to Know
- The dominant AI ROI framework ("AI replaced X hours of manual work, saving $Y") captures the least interesting part of AI's value. It frames AI as a cost reduction tool when it's actually a capability creation tool.
- Capability ROI measures what AI enables that was previously impossible: analysing every document instead of sampling, responding in minutes instead of days, serving segments that weren't economically viable before.
- The cost-savings framing actively undermines AI investment. It positions AI as an efficiency play, which means it competes with cheaper alternatives (offshoring, process simplification) rather than being evaluated as a strategic capability.
- Enterprises that measure capability ROI invest more confidently, scale faster, and build compound advantage. Those stuck on cost-savings ROI chronically underinvest.
- The ROI framework you use shapes the investment decisions you make. Choose the wrong framework, and you'll optimise for the wrong outcomes.
68%
of enterprise AI business cases are built primarily around cost reduction and labour savings
Source: Deloitte, State of AI in the Enterprise, 5th Edition, 2024
The Cost-Savings Trap
Here's how most AI business cases work:
"Our team of 12 spends 60% of their time on document review. AI can automate 80% of that work. That's 5.76 FTE-equivalents of savings. At $85K average salary, that's $489K per year. The AI solution costs $200K to build and $50K per year to operate. ROI: positive within 6 months."
The maths is clean. The board approves. And the framing is fundamentally limiting.
Here's what that framework misses:
It caps the value at the current cost. If you're measuring "labour saved," the maximum ROI is the total cost of the labour. AI's actual value often exceeds the cost of the work it replaces, because it does things the manual process couldn't.
It invites comparison with cheaper alternatives. If the pitch is "save $489K on document review," the board will ask: "Could we achieve the same savings by outsourcing?" Or "by hiring three junior analysts instead?" Cost-savings AI competes with every other cost-reduction initiative. Capability AI doesn't.
It measures the wrong baseline. The right question isn't "how much does this cost us today?" It's "what would it be worth if we could do this at scale, in real time, with zero errors?"
Capability ROI: A Better Framework
Capability ROI measures three things that cost-savings ROI doesn't:
1. What Was Previously Impossible
The claims team reviewed a 10% sample of documents because reviewing 100% wasn't economically viable. AI reviews 100%. The value isn't "we saved the cost of reviewing 10%." It's "we now catch issues in the 90% we weren't looking at." What's a missed fraud case worth? What's a compliance violation that would have gone undetected worth?
The question: "What can we now do that we couldn't do before, regardless of what it cost us not to do it?"
2. What Was Previously Too Slow
Customer enquiries took 3-5 business days to receive a substantive response. AI-assisted triage and knowledge retrieval reduces that to hours. The cost-savings framing values the time saved by the response team. The capability framing values the customer retention, satisfaction, and competitive advantage of responding 10x faster.
The question: "What's the value of speed that was previously unavailable to us?"
3. What Was Previously Uneconomical
Serving small-business clients with the same depth of analysis as enterprise clients wasn't economically viable because the margins didn't justify the human effort. AI makes the unit economics work. The real value is a new revenue stream or market segment that was previously inaccessible.
The question: "What markets, segments, or services does this capability open up?"
3.2x
higher AI investment levels at enterprises using capability-based ROI frameworks versus cost-savings frameworks
Source: McKinsey, The State of AI in Early 2025, March 2025
Building the Capability ROI Case
A capability ROI framework has four components:
Component 1: Baseline Capability
Document what you can do today, and what you can't. Be specific about the constraints:
- "We review 10% of transactions for compliance"
- "Average response time is 3.5 business days"
- "We can't economically serve clients below $50K annual revenue"
- "Policy analysis for cross-jurisdictional impact takes 6-8 weeks"
Component 2: Target Capability
Document what AI enables:
- "We review 100% of transactions for compliance, in real time"
- "Average response time is 4 hours, with complex cases escalated within 1 business day"
- "We can serve clients at $10K annual revenue with AI-assisted analysis"
- "Cross-jurisdictional policy analysis is available in 2-3 days"
Component 3: Capability Value
For each capability gain, quantify the value:
- Risk reduction: What's the expected value of issues caught in the 90% we weren't reviewing?
- Revenue enablement: What's the addressable market in the segment we couldn't previously serve?
- Speed premium: What's faster response time worth in retention, satisfaction, or competitive positioning?
- Quality improvement: What's the value of consistent, complete analysis versus sampling?
Component 4: Compound Projection
This is where compound advantage enters the ROI conversation. Each capability built on shared infrastructure makes the next capability cheaper and faster. The ROI of capabilities #1-4 together is dramatically higher than the sum of four individual ROI calculations.
Project the compound value: "Capability #1 costs $200K and delivers $500K in annual capability value. Capability #2 costs $120K (shared infrastructure) and delivers $400K. By capability #4, the platform delivers $2M+ in annual capability value on $600K total investment."
The Board Conversation
The cost-savings pitch sounds like: "AI will save us money."
The capability pitch sounds like: "AI will let us do things we can't do today: serve markets we can't reach, catch risks we're currently missing, respond at speeds our competitors can't match. And each capability we build makes the next one faster and cheaper."
One is a cost line. The other is a strategic investment.
When you frame AI as cost savings, it gets evaluated like every other cost initiative: squeezed, deferred, or substituted. When you frame it as capability gain, it gets evaluated as a strategic investment in competitive advantage.
Dr Tania Wolfgramm
Chief Research Officer
The Practical Shift
You don't need to abandon cost-savings metrics entirely. Labour efficiency is a real and measurable outcome. But it should be the floor of the ROI case, not the ceiling.
Lead with capability: "Here's what we'll be able to do that we can't do today."
Support with efficiency: "And by the way, we'll also save $X on manual processing."
Project the compound: "And each subsequent capability will be faster and cheaper to deliver."
That's the ROI conversation we should be having.
- How do we quantify capability value when we've never had the capability before?
- Use proxy metrics. For "review 100% instead of 10%," estimate the cost of issues in the unreviewed 90% using historical data. For "serve a new market segment," size the addressable market. For "respond 10x faster," measure current customer churn attributable to slow response. The numbers won't be precise, but they'll be directionally correct and more meaningful than cost savings alone.
- Will boards accept capability ROI frameworks, or do they want traditional ROI?
- Boards want confidence that investment will create value. Capability ROI actually provides more confidence because it addresses the strategic question ("what can we do differently?") rather than just the financial one ("how much will we save?"). Present both: capability value as the primary frame, cost savings as supporting evidence.