AI ROI: why investment fails to reach the bottom line

AI ROI has yet to show up at most companies that invest in the technology. In PwC's global survey of 4,454 chief executives in 95 countries and territories, conducted from September 30 to November 10, 2025, 56% said AI had brought neither higher revenue nor lower costs over the previous 12 months, and only 12% reported both gains. The most frequent cause lies in the choice of problems AI is asked to solve.

In an article published in Harvard Business Review in September 2026, David Rogers, a faculty member at Columbia Business School, and Krishnan Sankaranarayanan, who leads AI transformation at the chemical company Eastman, attribute the gap between investment and return to a failure of strategy. Companies point AI at low-value problems, and no one ties the project to a line of the income statement before it starts. The authors' remedy fits into four steps, detailed in the sections below.

Where does AI investment get lost?

It gets lost in projects that solve visible, inexpensive problems. Rogers and Sankaranarayanan list the most common: internal friction that annoys leaders and costs the company little, a new feature customers do not value, a use case copied because another company already did it, a dashboard that shows data in a new way without changing any decision, and a tool that saves employees minutes without the savings showing up in any budget. In each case the immediate gain is small, and the competitive advantage is nil, because competitors have access to the same models and the same use cases.

PwC reaches a similar diagnosis from a different angle. According to the survey, 26% of chief executives saw costs fall because of AI, and 22% saw them rise. The firm concludes that isolated, tactical projects seldom deliver measurable value, and that returns come from enterprise-scale deployment aligned with business strategy, built on foundations of technology, a road map, risk management, and culture. The 12% that achieved revenue gains and savings at the same time are the furthest ahead on those foundations.

Technology, adoption, and alignment: one factor at zero sets the whole to zero

AI value fits into an equation: value equals technology times adoption times strategic alignment. Technology covers the model, the infrastructure, compute, and data. Adoption is actual use by teams, the goal of change management. Alignment measures how much the problem being solved weighs in the company's priorities, balance sheet, and income statement. Because the three multiply, any one of them near zero brings value to zero: an excellent model without the right data, a well-designed solution no one uses, a powerful tool aimed at an irrelevant problem.

Companies have learned to look after the first two factors and neglect the third. This happens when AI sits with IT or an innovation unit that has no profit-and-loss responsibility, when leaders expect a model connected to all the data to spare them hard choices about which problems to solve, or when the program starts from the technology and from the fear of falling behind competitors. The clash between IT and AI teams explains part of that mismatch.

An example isolates the third factor. A solution that saves an employee one minute on a routine task is worth little at a company where that minute disappears into the slack of the day and the task sits off the critical path. At another, the same minute is worth a great deal: the task is the bottleneck that limits throughput, such as fitting one more package onto a truck, or it is repeated tens of thousands of times a day, or it cuts an employee's exposure to a hazardous environment. Technology and adoption are identical in both cases, and value changes with where the task sits in the economics of the business.

Value drivers by industry and position in the value chain

A value driver is a lever that, when moved, changes the profitability of the business in a way that matters, and it varies by industry. In heavy industry, the drivers are procurement, which accounts for a large share of revenue, and asset availability, because an idle plant raises the cost of every unit produced. In B2B services, the driver is sales productivity; at a subscription business, customer retention and acquisition cost; in retail, pricing and inventory; in consumer packaged goods, product ideation and the marketing mix.

Within the same industry, the driver changes with the company's position in the value chain. Two chemical makers show the difference. Westlake produces chlorine and polyethylene, traded on world markets, in facilities that cost billions of dollars, and earns its profit through operational efficiency, which calls for AI in plant operations. Givaudan produces high-margin fragrances, flavors, and cosmetic ingredients, spends far more on R&D as a share of revenue, and earns its profit through the speed and cost of launching products, which calls for AI in the lab and in product development.

Nestlé illustrates a driver of the second kind. According to the company, a proprietary generative AI tool cut product ideation from six months to six weeks in early tests reported in June 2024. In a market where the first product on the shelf takes the sale, development time is the line that counts.

What to do: list three to five value drivers for the business with the CFO and the heads of each unit, and remove from the AI portfolio any project that moves none of them.

The income statement, line by line

With the drivers defined, the next step is to read the income statement of the company or business unit line by line and ask what AI could move, and by how much. Revenue, cost of goods sold, and operating expenses should each generate a hypothesis about value. At an oil and gas company with assets concentrated upstream, reserve life and production cost are the critical lines, and they point to AI that reduces production downtime, to protect the top line, and AI that optimizes extraction, to extend the reserves.

The same reading points to new revenue from capital already spent. For the oil company in the example, it points to asset leasing and trading of the company's own crude on spot and futures markets, each supported by AI.

In a government agency, the budget plays the role of the income statement. The largest expense lines, such as recurring service contracts and public works, are where an efficiency gain shows up in budget execution. The article on agentic AI in procurement follows the same logic: procurement attracts investment in agents because its results are measured in money, contract by contract.

What to do: bring last year's income statement to the meeting that sets AI priorities, write next to each relevant line a hypothesis with an estimated value and an owner, and rank the project portfolio by that value.

Set the metric before the project

The last step is to decide how to measure AI's effect on the chosen driver, before building the solution. The rule is blunt: if a company cannot say in advance how it will measure a project and who will sign off on the value, it is not ready to begin. The metric comes from the problem: the false-reject rate on a production line, retention and viewing time at a subscription service, demand-forecast error and the holding cost of excess inventory at a retailer, the duration of each development stage at a consumer goods company.

Aramco publishes a metric of its own. Alongside its annual results, the company reports Technology Realized Value, the value it attributes to AI, digital, and other technology solutions. According to its March 2026 results release, the figure reached $5.3 billion in 2025 and $11.3 billion since 2023. It is the company's own estimate, without a separate audit, and it appears in the same document in which Aramco reports its earnings.

Measuring in advance also changes the standing of AI inside the organization. When every initiative is tied to a value driver and an income statement line, the people who own profit and investment decisions gain confidence in the program, and that confidence spreads through the rest of the company.

What to do: for every approved project, record before it starts the metric, the baseline, the target, the measurement date, and the name of the person who signs off on the result, and review any project that reaches that date without the number.

Conclusion

AI ROI appears when the program starts from the economics of the business. The four steps do exactly that: choose the value drivers, adjust them to the company's position in the value chain, read the income statement for hypotheses with an estimated value, and set the metric and the owner before the first model. Technology and adoption remain necessary, and strategic alignment decides where they point. Because competitors can use the same models, lasting advantage comes from a company's knowledge of its own economics.

The link between governance and business indicators is what Takter's ITIL® Maturity Journey examines. Using the ITIL maturity model, Takter assesses the five components of the service value system, from governance to continual improvement, and connects governance to indicators such as return on assets (ROA) and return on equity (ROE). The assessment shows the organization's current stage and the areas to improve in order to deliver predictable results.

Sources

• PwC, 29th Global CEO Survey, January 2026.
• David L. Rogers and Krishnan Sankaranarayanan, Closing the Gap Between AI Investment and Financial Return, Harvard Business Review, September 2026.
• Nestlé USA, Unlocking New Opportunities with Gen AI, June 2024.
• Aramco, Fourth quarter and full-year 2025 results, March 2026.