It’s the same question virtually every company trying out generative artificial intelligence is grappling with: where’s the payoff? A recent survey of 600 data leaders conducted by Wakefield Research on behalf of Informatica found that 97% of organizations are working to realize the business value it promised. The squeeze is getting tighter, as companies’ CFOs are under siege and pilots multiply.
The problem is rarely a shortage of good ideas. Benefits are, that is. Not just benefits but the fact that they’re scattered across functions, show up in uneven timelines, and are hard to attribute. But calculating AI ROI isn’t a hopeless endeavor. With the proper framing and operating discipline, leaders can turn from anecdotes to evidence. Here are five expert-backed moves that actually get results.
Put Value In The Language Of The Business
Begin by translating technical success into line-of-business results. Connect each use case back to a value driver the organization is already tracking: revenue lift, cost-to-serve reduction, risk capital savings, cycle-time compression, customer retention, or hours given back to the workforce. If your influencer is the CMO, capture a decrease in churn; for operations, focus on reductions in truck rolls or mean time to resolution.
Accenture leaders frequently explain that no successful CFO “just takes the model’s ROI spreadsheet at face value.” The cure is to condition the measures in your playbook – and that should be the financial playbook you use for any kind of investment – that are the unit economics, baselines, and ranges of sensitivities agreed up front. Don’t get misled by vanity metrics like “queries served,” but build P&L-relevant KLIs and risk-adjusted value.
Instrument On Day 1 With Baselines And A/B Tests
Think about deploying AI as an experiment, not a rollout. Do before-and-after measurements, additional counterfactuals, and holdouts. Quantify adoption and impact with event-level telemetry: time saved per task, deflection rates, error reductions, and downstream conversion.
When it is well designed, the evidence can be compelling. A Stanford and MIT study of AI assistance shows a 14% productivity gain for contact center agents. You can construct similar proof points at your company: control for effects, for instance by comparing teams with and without the tool and tracking changes over weeks, attributing only the delta beyond learning-curve or team-size effects or seasonality.
Get Finance Involved Early In AI ROI Measurement
Engage FP&A to help co-author the benefits logic and ROI model. “The key is to agree on attribution rules, discount rates, and who will pay for shared platform costs before work begins.” Then set up a monthly benefits rhythm—actuals versus forecast, with variance explanations—so that the ROI narrative advances with data, not slideware.
Organizations such as Rabobank demonstrate the impact of this approach with data leaders reporting to CFOs and changing ROI from a one-way justification into a two-way design conversation. At these moments, when finance supports the case, skepticism becomes stewardship and scale.
Invest In Data Foundations And Clear Stop Rules
Generative AI is what you make of it — fueled by whatever data and governance you already have. Poor quality, poor ROI. According to Gartner, bad data has been costing companies $12.9 million a year on average — leakage that is canceling out any benefits from AI. Invest in data lineage, access control, and reference data management so models are based on trustworthy inputs.
At the same time, shield ROI with “stage gates” and kill criteria. Determine up front how success will be measured at 30, 60, and 90 days — target precision, turnaround time, or adoption KPIs. Companies such as Jotun are speeding value by aggregating data to the latest cloud platforms, but they’re also putting in rules to put an end to experiments that don’t pay back. Pairing a low-cost hub with explicit stop signals keeps portfolios focused on the winners.
Tell The Outcomes Story And Connect To Strategy
Executives do not finance models; they fund results. Frame outcomes in the story of what you want your CEO to prioritize — growing markets, productivity, customer trust, or resilience. AWS data leaders sometimes recommend translating metrics to the language of stakeholders: for sales, incremental pipeline; for service, higher first-contact resolution; for compliance, less manual review with equivalent effectiveness.
For consumers, this saga is important as well. A Deloitte survey finds consumers would pay a premium for “responsible” AI tools, leading to the conclusion that risk-aware design and transparency can be value creators in their own regard. McKinsey’s study of AI adoption reaches a similar conclusion: the companies that are most successful at generating impact through AI combine technical execution with change management and skills — key characteristics of a strategy-led narrative, not one led by available tooling.
Now package all of the above into a simple, repeatable playbook:
- Define value in business terms.
- Instrument impact from day one.
- Co-own ROI with finance.
- Build on strong data foundations, including stop rules.
- Communicate outcomes alongside corporate strategy.
Do that regularly and the 97% issue becomes a competitive advantage — since you will be one of the few who can show the receipts.
