Quick Answer
How Do You Measure AI ROI?
AI ROI (return on investment) compares the value an AI initiative creates with its full cost: AI ROI = (total benefits − total costs) ÷ total costs × 100. To measure it credibly, define the business outcome first, record a baseline before AI is introduced, count every cost (including training and change management), track efficiency, quality, growth and risk benefits, convert time saved into money only when that time is genuinely redeployed, and review results at 30, 90 and 180 days.
Typical payback
2–4 years per use case
First step
A baseline, before you build
Biggest mistake
Treating time saved as money saved
Who owns it
A named business leader
Every leadership team now faces the same question from its board, its CFO or its own conscience: what are we actually getting back from AI? Budgets keep rising. Pilots keep multiplying. And yet, when someone asks for the return, the answer is often a vague mix of “hours saved”, “productivity gains” and “it’s early days”.
This guide is a practical answer for leaders. It explains what AI ROI is, what the latest research says about returns, why AI ROI is so hard to measure, and a six-step framework, with a worked example and a calculator, that you can use to measure the return on your own AI investments.
What Is AI ROI?
AI ROI (artificial intelligence return on investment) measures the business value an AI initiative generates relative to what it costs to build, run and adopt. Like any ROI, it is expressed as a percentage or a multiple: an ROI of 50% means every dollar invested returned $1.50 in value.
What makes AI ROI different from traditional technology ROI is where the value shows up. A new ERP system replaces a process. AI usually changes how people do work and make decisions, so its value is spread across time saved, better quality, faster decisions, new revenue and reduced risk. That is why the same AI tool can show a strong return in one team and none in another.
The AI ROI formula
AI ROI (%) = (Total benefits − Total costs) ÷ Total costs × 100
Payback period (months) = Upfront cost ÷ Net monthly benefit
For multi-year investments, finance teams will also want net present value (NPV), which discounts future benefits to today’s money. Use the same benefit and cost estimates; only the timing treatment changes.
Is AI Delivering a Good ROI? What the Research Says
The honest answer is: for a minority of organisations, yes; for most, not yet. Four major studies published since 2025 point in the same direction, even though each measures something slightly different.
- An IBM study of 2,000 CEOs across 33 countries found that only 25% of AI initiatives had delivered their expected ROI over the previous few years, and only 16% had scaled enterprise-wide.
- Deloitte’s survey of 1,854 senior executives in Europe and the Middle East found that 15% of generative AI users, and just 10% of agentic AI users, reported significant, measurable ROI.
- BCG’s 2025 study of 1,250 executives classed only 5% of companies as “future-built” and achieving AI value at scale, while 60% reported minimal revenue and cost gains.
Four studies, four definitions, one message
Share reporting strong AI returns, by study. Measures differ, so compare the pattern, not the bars.
AI initiatives that delivered expected ROI (IBM, 2025)
25%
GenAI users with significant, measurable ROI (Deloitte, 2025)
15%
Agentic AI users with significant ROI (Deloitte, 2025)
10%
Companies achieving AI value at scale (BCG, 2025)
5%
Sources: IBM CEO Study (May 2025); Deloitte (Oct 2025); BCG (Sep 2025)
Yet investment keeps climbing. Deloitte found that 85% of organisations had increased their AI investment in the previous year and 91% planned to increase it again. Deloitte calls this the paradox of rising investment and elusive returns. Our reading is simpler: most organisations are investing in AI faster than they are learning how to manage it for value.
The AI value gap
Share of companies by AI value achieved, BCG global survey of 1,250 executives, 2025
5% future-built: achieving value at scale
35% scalers: beginning to generate value
60% laggards: minimal revenue and cost gains
What is the average ROI of AI?
There is no reliable single average, and you should be wary of anyone who quotes one without context. The most cited figure comes from an IDC study commissioned by Microsoft, which reported an average return of $3.70 for every $1 invested in generative AI, and around $10 for leading organisations. Those figures are self-reported by respondents and come from a study funded by an AI vendor. They sit alongside IBM’s finding that only a quarter of AI initiatives met their expected ROI.
Both can be true. Averages are pulled up by a small group of organisations getting large returns, while most initiatives deliver little. For a leader, the useful question is not “what is the average?” but “what would it take for our use cases to land in the successful minority?”
How Long Does AI Take to Deliver ROI?
Longer than most business cases assume. Deloitte found that most organisations achieved satisfactory ROI on a typical AI use case in two to four years, compared with seven to twelve months for typical technology investments. Only 6% reported payback in under a year.
AI pays back on a different clock
Typical time to satisfactory ROI, in months
Typical technology investment: 7–12 months
Typical AI use case: 24–48 months
Source: Deloitte, AI ROI: The Paradox of Rising Investment and Elusive Returns (October 2025)
The reason is the shape of the curve. AI value usually follows a J-curve: costs arrive first (licences, integration, data work, training), while benefits arrive only as people change how they work. Leaders who judge a use case at month three often kill it just before it starts paying. Leaders who never set a review point let weak ones run forever. Both are measurement failures.
Why Is AI ROI So Hard to Measure?
In most organisations we work with, the problem is not a lack of AI benefits. It is that nobody set up the conditions to see them. Five issues come up again and again.
- No baseline. If you did not measure how long a process took, or how many errors it produced, before AI arrived, you cannot prove what changed.
- Time saved is not money saved. Ten minutes saved per person per day only creates value if that time goes into something worth more. Otherwise it disappears into the day.
- Benefits are spread thin. General-purpose assistants create small gains across thousands of tasks, which are real but hard to attribute.
- Costs are hidden. Business cases count licences and forget integration, data preparation, security review, training, change management and the time of internal experts.
- Nobody owns the outcome. When AI is “an IT project”, no business leader is accountable for whether the promised value arrives.
The gap is visible in IBM’s research: around 79% of executives in one IBM study saw productivity gains from AI, but only about 29% said they could confidently measure its ROI.
AI ROI is a leadership discipline before it is a spreadsheet. The organisations that measure it well decided what success looked like, and who owned it, before the first licence was bought.
What is the 10-20-70 rule in AI?
The 10-20-70 rule comes from BCG’s research on AI adoption. Companies that get value from AI put roughly 10% of their effort into algorithms, 20% into technology and data, and 70% into people and processes. For ROI, the implication is direct: most of the return, and most of the cost, sits in the 70%. A business case that budgets only for technology will overstate the return and underdeliver it.
Where AI value is won or lost
Share of effort among AI leaders, BCG
Technology & data
People & processes
How to Measure AI ROI: A Six-Step Framework
This is the framework we use with leadership teams. It works for a single use case and scales to a portfolio. The steps are simple; the discipline is in doing them before the AI goes live.
Step 1: Start with the business outcome, not the tool
Write down the decision or workflow the AI will change and the outcome that proves success: “reduce invoice processing time by 40%”, “increase qualified pipeline by 15%”, “cut first-response time in customer service to under two hours”. If a use case cannot name its outcome, it is not ready to be funded. This is also where you name the business owner, the leader accountable for the result.
Step 2: Record a baseline before you build
Measure the current state: cycle time, volume, error rate, cost per transaction, conversion rate, customer satisfaction, whatever the outcome depends on. Two to four weeks of baseline data is usually enough. Without it, every later claim is an opinion.
Step 3: Count the full cost
Include everything the initiative will consume over your measurement horizon, typically one to three years.
| Cost category | What to include |
|---|---|
| Technology | Licences, usage-based model costs, infrastructure, vendor fees |
| Build and integration | Integration with existing systems, data preparation, testing, security and legal review |
| People and change | Training, workflow redesign, change management, internal experts’ time, temporary productivity dip |
| Run and govern | Monitoring, quality checks, model updates, support, governance and compliance effort |
Step 4: Measure four layers of return
AI returns fall into four layers. Most business cases measure only the first; the strongest returns are usually in the second and third.
Layer 1
Efficiency
Time saved, cost per transaction, cycle time, automation rate, cost avoided
Layer 2
Effectiveness
Error and rework rates, decision speed and quality, customer satisfaction, compliance
Layer 3
Growth
Revenue uplift, conversion, retention, new products or services, faster time to market
Layer 4
Resilience & risk
Risks reduced, incidents avoided, capability built, dependence on scarce skills reduced
Track adoption alongside these as a leading indicator: the share of the intended users who use the AI regularly on real work. If adoption is low at 90 days, the financial return will not arrive, whatever the business case said.
Step 5: Convert benefits to money, carefully
This is where most AI ROI numbers go wrong. Hours saved become dollars only to the extent that the time is redeployed to valuable work, avoids hiring, or reduces overtime or contractor spend. We recommend applying an explicit redeployment factor, the share of saved time you can show was put to productive use, rather than assuming 100%.
Value of time saved = Hours saved × Fully loaded hourly cost × Redeployment factor
For quality and growth benefits, use the finance team’s existing values where possible: the cost of an error, the margin on an extra sale, the lifetime value of a retained customer. Agreeing these values with finance upfront is what makes the final number credible in the boardroom.
Step 6: Review at 30, 90 and 180 days, with stop criteria
Set review points before launch. At 30 days, check adoption and quality. At 90 days, check whether the leading indicators are moving. At 180 days, check the financial return against the business case. Agree in advance what result would trigger scaling, redesign or stopping. Stopping a weak use case early is a positive ROI decision: it frees money and attention for stronger ones.
AI ROI Example: A Worked Calculation
The following example is hypothetical, but it uses realistic proportions for a common use case: AI-assisted invoice processing in a finance shared-service team.
| Year 1 | Amount (USD) |
|---|---|
| Licences and usage | 48,000 |
| Integration and data preparation | 70,000 |
| Training and change management | 32,000 |
| Support, monitoring and governance | 25,000 |
| Total cost | 175,000 |
| Time saved: 12,000 hours × $45 loaded cost × 50% redeployed | 270,000 |
| Errors, rework and late-payment penalties avoided | 30,000 |
| Total benefit | 300,000 |
| AI ROI = (300,000 − 175,000) ÷ 175,000 | 71% |
Now see what happens if the same team assumes every saved hour is money saved, the most common mistake in AI business cases.
The same project, two very different ROI numbers
Year-1 ROI in the hypothetical example above
Naive: every hour saved counted as cash
226%
Realistic: 50% of saved time shown to be redeployed
71%
Hypothetical example for illustration. Naive case: (540,000 + 30,000 − 175,000) ÷ 175,000.
A 71% first-year return is still a good investment. But a business case built on 226% sets expectations the project can never meet, and that gap is where boards lose confidence in AI. Note also that this is the return once the system is fully adopted; during the first months of the J-curve, the return will be lower.
AI ROI Metrics: What to Track
Choose three to five metrics per use case: at least one leading indicator you can see within weeks, and one lagging financial measure the CFO will recognise. These are the metrics we see used most effectively.
| Layer | Leading indicators (weeks) | Lagging measures (months) |
|---|---|---|
| Adoption | Weekly active users, tasks completed with AI | Share of the workflow running on the new process |
| Efficiency | Time per task, cycle time | Cost per transaction, overtime or contractor spend, hiring avoided |
| Effectiveness | Error and rework rate, first-time-right | Customer satisfaction, compliance findings, write-offs |
| Growth | Response time to leads, proposals produced | Conversion rate, revenue per customer, retention |
| Risk | Policy breaches caught, outputs reviewed | Incidents, fines or losses avoided |
IBM describes the same split as hard ROI (measurable financial effects such as cost reduction and revenue growth) and soft ROI (benefits such as employee satisfaction, decision quality and customer experience). Both matter, but present hard ROI to the board and use soft ROI to explain it.
How Leaders Improve AI ROI
The research on the organisations that do get strong returns is consistent. They do a few things differently, and almost all of them are leadership choices rather than technology choices.
- They focus. BCG’s future-built companies expect twice the revenue increase and 40% greater cost reductions than laggards in the areas where they apply AI, and earlier BCG research found that AI leaders pursue about half as many opportunities as their peers. Fewer, better-chosen use cases are easier to resource, measure and scale. Our guide to AI use case prioritization explains how.
- They measure differently for different AI. Deloitte found that 85% of its AI ROI leaders use different frameworks or timeframes for generative and agentic AI, rather than one test for everything.
- They invest in people. 40% of Deloitte’s AI ROI leaders mandate AI training. When people do not know how to use AI well, the benefits in a business case never materialise.
- They fix foundations. IBM research suggests that paying down legacy technical debt can improve AI ROI by up to 29%.
- They name an owner. Every use case has a business leader accountable for its outcome, not just a project manager accountable for its delivery
What is the 30% rule for AI?
The “30% rule for AI” is an informal rule of thumb with no single authoritative definition; people use it to describe very different ideas, from the share of work AI can take on to the share of AI projects that fail. The best-documented 30% figure is Gartner’s prediction that at least 30% of generative AI projects would be abandoned after proof of concept by the end of 2025, with unclear business value among the main reasons. That is exactly the problem a measurement discipline is designed to prevent.
What does a 20% ROI mean?
A 20% ROI means the investment returned its full cost plus an additional 20%. For example, an AI project that cost $100,000 and generated $120,000 in measurable value has an ROI of 20%. Whether that is good depends on the time it took, the risk involved and what else the money could have funded, which is why ROI should always be read alongside the payback period.
For leadership teams
Want AI investments your board can see a return on?
Our instructor-led workshops help leadership teams choose the right AI use cases, set baselines and owners, and build an AI roadmap with measurable outcomes.
Frequently Asked Questions About AI ROI
What is AI ROI?
AI ROI is the return on investment from artificial intelligence: the business value an AI initiative creates compared with its total cost to build, run and adopt. It is usually expressed as a percentage, calculated as (total benefits − total costs) ÷ total costs × 100.
How do you calculate ROI on AI investments?
Define the business outcome, measure a baseline, count all costs over your time horizon (technology, integration, training, change management and running costs), measure benefits across efficiency, effectiveness, growth and risk, convert them to money using agreed values, then apply the formula: (benefits − costs) ÷ costs × 100.
What is the average AI ROI?
There is no reliable single average. An IDC study commissioned by Microsoft reported an average self-reported return of $3.70 per $1 invested in generative AI, while IBM found that only 25% of AI initiatives delivered their expected ROI. Returns are concentrated in a minority of organisations with strong strategy, adoption and measurement.
Is AI giving a good ROI?
For a minority of organisations, yes. BCG found that 5% of companies achieve AI value at scale and 60% report minimal gains, and Deloitte found that 15% of generative AI users report significant, measurable ROI. The difference is usually in use-case selection, adoption and leadership ownership rather than the technology.
How long does it take to see ROI from AI?
Deloitte’s 2025 research found that most organisations achieve satisfactory ROI on a typical AI use case in two to four years, compared with seven to twelve months for typical technology investments. Only 6% reported payback in under a year. Narrow, well-defined use cases can pay back faster.
What is an AI ROI framework?
An AI ROI framework is a structured method for planning and measuring the return on AI. A practical framework defines the outcome and owner, records a baseline, counts full costs, measures several layers of benefit, converts benefits to money with explicit assumptions, and reviews results at set intervals with criteria for scaling or stopping.
What is the 10-20-70 rule in AI?
The 10-20-70 rule comes from BCG research: organisations that get value from AI put about 10% of their effort into algorithms, 20% into technology and data, and 70% into people and processes.
What is the 30% rule for AI?
The 30% rule for AI is an informal rule of thumb with no single authoritative definition. The best-documented 30% figure is Gartner’s prediction that at least 30% of generative AI projects would be abandoned after proof of concept by the end of 2025.
What does a 20% ROI mean?
A 20% ROI means an investment returned its full cost plus 20%. An AI project costing $100,000 that generates $120,000 in measurable value has a 20% ROI.
What is the difference between hard and soft AI ROI?
Hard ROI covers measurable financial effects, such as lower costs, avoided hiring or higher revenue. Soft ROI covers benefits that are real but harder to value directly, such as better decisions, employee satisfaction and customer experience. Credible AI business cases lead with hard ROI and use soft ROI to explain and support it.
Do enterprise AI projects deliver a good ROI?
Many do not yet. IBM’s 2025 CEO study found that only 16% of AI initiatives had scaled enterprise-wide. Enterprise projects that deliver strong returns typically focus on a small number of core-business use cases, invest in adoption and change management, and are owned by a business leader rather than IT alone.
What should an AI ROI strategy include?
An AI ROI strategy should include a prioritised portfolio of use cases tied to business outcomes, agreed valuation methods with finance, baselines and owners for every use case, a budget that covers people and process as well as technology, and a regular leadership review that scales, redesigns or stops initiatives based on results.
References
- IBM, IBM Study: CEOs Double Down on AI While Navigating Enterprise Hurdles (May 2025)
- Deloitte, AI ROI: The Paradox of Rising Investment and Elusive Returns (October 2025)
- BCG, The Widening AI Value Gap: Build for the Future 2025 (September 2025)
- Microsoft, IDC’s 2024 AI Opportunity Study: Top Five AI Trends to Watch (November 2024)
- IBM, How to Maximize AI ROI
- BCG, AI Adoption in 2024: 74% of Companies Struggle to Achieve and Scale Value (October 2024)
- Gartner, Gartner Predicts 30% of Generative AI Projects Will Be Abandoned After Proof of Concept by End of 2025 (July 2024)
Ready to build AI capability in your leadership team?Instructor-led, hands-on AI workshops for leaders and managers
Explore AI Training →
About the author
Anupma Singh
Founder & AI Strategy Advisor, 10X AI Leaders · IIT Kanpur
Anupma Singh is the Founder and AI Strategy Advisor of 10X AI Leaders and an IIT Kanpur alumna. She has 25 years of experience across technology, business transformation and AI adoption strategy, including leadership roles at Infosys and Honeywell. She has trained 100+ business leaders and runs instructor-led AI workshops that help leadership teams move from AI-curious to AI-confident.