The finding that should worry every CFO
Harness published data on July 30 showing that roughly one in four dollars spent on AI goes to waste, drawn from a survey of 700 engineering leaders and practitioners. The waste is not mysterious. More than half of respondents said their organization lacks a dedicated owner for AI costs, leaving responsibility diffused across engineering, platform, and FinOps teams. Only one in five can determine the source of an unexpected AI cost spike within hours. Harish Doddala, VP of cloud and AI cost management at Harness, put it directly: "The visibility problem, the ownership problem, the forecasting problem show up whether you are spending $300K a month or $3M."
The mechanics of the leak are worth naming. AI spend spans infrastructure, foundation models, SaaS subscriptions, and managed services, and most organizations run three or more providers with different pricing structures. AI copilots and coding assistants add cost that looks like an ordinary software license, which makes it easy to lose track of. More than half of organizations forecast AI spend through guesswork rather than data, and over 40% still manage it in spreadsheets. That is a recipe for the exact waste the survey measured, and it is the norm, not the exception.
Retail is spending straight into the gap
Retail is not a bystander to this problem. Global AI spending in retail is projected to reach $19.9 billion in 2026, up from $6.4 billion in 2021, and the appetite is still climbing. Deloitte found that 82% of retail executives plan to increase AI investment over the next twelve months, and NVIDIA's 2026 read put that figure at 97% of retailers. Every story we have covered this season, from Kohl's shopping assistant to grocery shelf-scanning vision, adds another line to that budget. The spend is real and accelerating, and the governance the Harness report says is missing is missing in retail too.
That combination is what makes retail exposed. A sector increasing AI investment at double-digit rates, running multiple model and platform providers, and lacking a single cost owner is precisely the profile Harness describes as leaking a quarter of its spend. The retailers writing the biggest AI checks are often the ones with the most fragmented ownership, because the spend originated in pilots scattered across merchandising, marketing, supply chain, and store operations. Nobody built the cost-control layer, because in a pilot the numbers were small enough to ignore. At scale they are not.
The returns do not yet justify the run rate
The uncomfortable counterpart to the spending curve is the return curve. McKinsey's State of AI found only 6% of organizations generating genuine enterprise-level value from AI. In retail specifically, Deloitte puts enterprise-wide AI deployment at just 7 to 10 percent, which means most of the sector is still in pilots and point solutions. Bain's 2026 work found that while 37% of companies targeted cost reductions of 11 to 20 percent from AI, nearly 40% landed below 10 percent instead. The money is going out faster than the value is coming back, and that gap is what waste looks like at the portfolio level.
The barriers are consistent across the research. Bain found 41% cite data access and integration as the biggest obstacle, and BCG estimates 70% of AI value comes from people, processes, and operating model, the areas most retailers leave largely untouched. Stanislas Vignon, Head of Insights at LVMH, said it cleanly: "AI is not a magic wand. If you don't have the right data, it doesn't work." The pattern is that retailers are funding models and tools while underfunding the data and process work that actually converts spend into results.
Why the costs stay invisible
The visibility problem is structural. Cloud infrastructure has a decade of FinOps tooling behind it, so a spike in compute usually gets caught. AI spend does not have that maturity. Foundation model tokens, per-seat copilot licenses, vector database queries, and managed agent services each bill differently, and they rarely roll up into one dashboard. When only one in five organizations can trace a cost spike within hours, the other four are finding out about overruns weeks later on an invoice, long after the runaway job or misconfigured agent could have been stopped.
Diffused ownership makes it worse. When AI cost sits across engineering, platform, and FinOps with no single accountable owner, every team assumes another is watching the meter. That is how a proof-of-concept that cost a few thousand dollars a month becomes a production workload burning six figures before anyone notices. The Harness finding that spend behaves the same at $300K and $3M a month is the tell. The dysfunction tracks one variable, whether someone owns the number, and budget size barely moves it. Most retailers have not yet named that person.
FinOps for AI is the discipline retail keeps skipping
The answer is not novel, which is part of why it gets skipped. Retail already knows how to do FinOps for cloud, and the same playbook applies to AI: unified visibility across every provider, a named owner accountable for the total, forecasting grounded in usage data rather than guesswork, and alerts that fire in hours instead of at month-end. The difficulty is organizational, not technical. It requires pulling AI cost out of the scattered pilot budgets where it was born and giving it a single home with real authority to question spend.
For a retail technology leader, this is a governance move that pays for itself quickly. If a quarter of AI spend is genuinely waste, then cost visibility is one of the highest-return projects available, because it recovers money already committed. We would rather see a retailer spend the next quarter instrumenting its AI costs than approving another round of models it cannot yet measure. The discipline also improves the ROI story upward, because a leader who can show where every AI dollar goes can defend the budget that funds the projects that actually work.
What to fix before the next budget cycle
The near-term steps are unglamorous and effective. Name a single owner for AI cost, with a mandate that spans engineering, platform, and finance. Consolidate every AI provider bill into one view so the total is visible in one place. Replace spreadsheet forecasting with usage-based projection, and set spike alerts that reach a human within hours. None of this requires new AI capability. It requires treating AI spend with the same rigor retail already applies to cloud, labor, and inventory, which is the standard the Harness data says most organizations have not met.
The strategic payoff is credibility. Retail is heading into budget season with AI spend rising and boards starting to ask harder questions about return. The leaders who can answer where the money went, what it produced, and where the waste was cut will keep their funding. The ones who cannot will face the skepticism that 83% of retail executives in one survey already voiced, the belief that the AI investment boom will end before 2030 because it failed to generate real value. Cost governance is how a retailer stays on the right side of that skepticism.
