Most finance leaders see AI as a series of funding decisions: which pilot to approve, which vendor to renew, which business case looks least speculative this quarter. That framing makes sense. It’s also the reason so many AI investments never pay back. When every initiative arrives as a new line item, each one has to justify itself from scratch. Each one also rebuilds the same data plumbing underneath, and each one argues about whose number is right. That’s why data modernization has become a finance issue as much as a technology one.
The results speak for themselves. In IBM’s 2026 CEO Study, CEOs who had redesigned five core areas of the business, including finance, were 4x more likely to have realized the full benefits outlined in their AI business cases. And CEOs most committed to reinvesting productivity gains and scaling AI systematically had scaled 23% more AI initiatives enterprise-wide than those least committed. Meanwhile, 83% of CEOs say AI success depends more on people’s adoption than on the technology itself.¹
That isn’t a technology problem. It’s a capital allocation problem, which makes it a CFO problem.
The latest tech news, backed by expert insights
Stay up to date on the most important—and intriguing—industry trends on AI, automation, data and beyond with the Think Newsletter, delivered twice weekly.
Three trends are converging on finance leaders right now, and any data modernization strategy has to account for all three.
- AI spend isn’t showing up in the P&L. Boards are funding AI because they can’t afford not to. But a model that answers questions nobody trusts, or a pilot that never leaves the lab, is pure cost. An AI initiative that lands as a new budget line fights for survival every cycle. One that visibly lowers cost-to-serve or speeds up a revenue decision earns its place in the budget that already exists.
- You’re paying for the same data many times over. Most enterprises copy the same customer, product, and asset data into warehouses, marts, BI extracts, and now AI pipelines. Every copy carries storage, compute, and reconciliation cost. Flexera’s 2026 State of the Cloud Report found that wasted cloud spend rose to 29% after five years of decline, which the report attributes to growing cost complexity from AI and new cloud services.²
- Nobody agrees on the number. Ask three teams for net revenue retention and you’ll often get three answers, each defensible and none reconciled. Today that costs time in the close and credibility in the boardroom. Once AI agents start answering those questions at scale, it becomes an audit exposure.
Each of these is manageable on its own. Together they explain why AI ROI is so elusive. Finance is being asked to fund innovation on a flat budget, cut the cost of data it already owns, and stand behind numbers that machines now generate, all at once.
The good news is that nobody has to build the stack of tomorrow from scratch. Most enterprises already run a modern data platform, whether that’s Snowflake, Databricks, Google BigQuery, or a mix. The gap was never storage or compute. It’s that the data on those platforms isn’t packaged, defined, or governed in a way that finance can count as an asset. Closing that gap is what data modernization means for a CFO.
So the question changes. Not “which AI should we fund next?” but “which data is reusable enough to fund once and spend many times?”
In finance terms, the data modernization path to answering that question runs through four stages:
- Migrate: cut the run rate. Retire legacy warehouses and duplicate copies. Open table formats such as Apache Iceberg let teams keep a single physical copy and share it across engines instead of paying to move it.
- Modernize: make spend visible. Apply FinOps discipline to data and AI workloads. Allocate compute and AI usage costs to the business units that consume them, so spend tracks value rather than habit. Flexera reports that 49% of organizations now use unit economics to tie cloud spend to outcomes, up from 40% a year earlier.²
- Contextualize: define every metric once. Put business definitions into a governed semantic layer so every report, dashboard, and AI agent uses the same logic.
- Monetize: turn data into decisions and revenue. Publish trusted, reusable data products that speed internal decisions and, where it makes sense, create new revenue through partner or customer data sharing.
Most data modernization programs spend their energy on stages one and two. The real payback sits in stage three.
Consider a single metric, net revenue retention. Finance owns the definition: expansion minus churn, on active contracts, measured monthly at the account level. In most enterprises that definition lives in a spreadsheet, a slide footnote, and a dozen slightly different SQL queries.
A governed semantic layer changes that. The definition is written once as code, owned by a named person in finance, tested, and versioned. Then it’s enforced natively on whichever platform the data lives on: semantic views in Snowflake, governed metrics in Databricks Unity Catalog, or the Looker semantic layer on Google Cloud. A dashboard, a planning model, and an AI agent asking the same question all get the same answer.
The same logic applies to access policy. Rules like “only HR business partners see salary detail” or “EU customer data stays in region” can be written in business language once and compiled into each platform’s native controls. Audit, lineage, and quality evidence then flows back as a single scorecard.
For a CFO, that’s the difference between hoping AI tells the truth and being able to prove it does. One definition. One enforcement model. One evidence trail for the auditors.
Two credible narratives are circulating in finance leadership about AI spend, and a pragmatic strategy recognizes the tension between them.
The first camp says freeze it. Until AI proves its ROI, every dollar should be treated as speculative, and the burden of proof sits with the business.
The second camp says the risk is falling behind. Competitors are moving, and waiting for certainty means paying more later to catch up.
In a sense, both are right. The answer isn’t choosing between caution and speed. It’s changing what gets funded. Build from the core out: foundation and definitions first, then governed data products, and AI agents last. An agent is only as trustworthy as the data beneath it, and funding the foundation first means each later AI use case costs less and carries less risk than the one before it.
A data product is a governed, reusable package of data with a named owner, a clear definition, quality commitments, and access rules attached. It’s built once and consumed many ways: through SQL, dashboards, applications, APIs, or AI agents.
That shift matters on the balance sheet. A project is a sunk cost. A data product is closer to an asset: funded once, reused across teams, and measurable through usage. Sharing capabilities such as Snowflake Secure Data Sharing, Databricks Delta Sharing, and BigQuery sharing let teams grant access to live data without making new copies. Usage metering then makes internal cross-charging practical.
The pattern holds across industries. A utility that builds one governed view of each customer premise can reuse it for outage communications, billing analytics, and rate-case support. A manufacturer that packages line performance data once can serve the plant floor, the supply chain team, and the finance close from the same source. The use cases differ. The investment logic is identical.
A practical data modernization strategy doesn’t start with a transformation program. It starts with one decision: name two or three pilot domains and the business owners accountable for them.
From there, a CFO can bring four questions to the next budget review:
- Which metrics do we argue about most? Those are the first candidates for a governed semantic layer.
- How many copies of our core customer, product, or asset data are we paying for? That’s the run-rate opportunity.
- Can we allocate data and AI spend to the business units that consume it? If not, FinOps comes before the next AI pilot.
- Which AI use cases would reuse the same data products? Fund those first, because each one lowers the cost of the next.
The organizations that come out ahead won’t be the ones with the most AI pilots. They’ll be the ones that treat data modernization as an investment discipline: governing data as an asset, defining what their numbers mean, and funding foundations that keep paying back.
Abhi Majumdar
Associate Partner, Data & AI Leader, Industrial and Communications
IBM Consulting
