AI results are now showing up as financial figures, and 2027 budgets will be set against those numbers. CEOs who want to stay ahead will specify which profit line each project moves and when it stops.
The era of AI announcements is ending. Finance teams are building 2027 plans under regulatory scrutiny, especially from Brussels. The relevant question is not how many pilots are running, but which P&L lines will actually change. Few companies have the discipline to stop projects that are not working. CEOs need to be ready to defend their 2027 plans with specifics that will stand up to a CFO’s questions.
Investors are now comparing AI claims to actual numbers. Janus Henderson reviewed earnings reports from 109 listed companies: 69 percent disclosed margin improvement from productivity gains and AI-linked cost reductions, while 26 percent mentioned revenue growth. Management teams are shifting from reporting adoption metrics to reporting financial results, with some now issuing guidance. This shift allows outside investors to test claims against real outcomes.
The named figures are getting specific. IBM said it realised $4.5 billion in AI-related savings in 2025 and is targeting $5.5 billion in 2026. PayPal has guided investors to $1.5 billion in run-rate savings, most of it driven by AI. Shell reported $400 million in annual savings and a 45 percent cut in unplanned downtime. MercadoLibre said product development expense fell by 120 basis points year on year, and Freeport-McMoRan reported a 10 percent rise in mill throughput with unit costs down by as much as 15 percent. The savings cluster in two places, software development and customer service, which is a useful hint about where to look first. Alphabet, for instance, reports that roughly half of its internal code is now written by AI agents. On the revenue side, Walmart said orders placed through its shopping assistant were 35 percent larger than those placed without it.
A separate survey of 1,000 senior leaders at large European firms, reported in secondary coverage, put the share of AI initiatives that stay at pilot stage indefinitely at 40 to 48 percent. A project with no exit rule rarely ends. It becomes the thing finance keeps paying for and nobody champions. The companies posting savings in their earnings reports are, in effect, the ones that decided which projects to scale and which to drop. The evidence suggests AI is increasing output per employee and slowing hiring, not simply replacing jobs. CEOs should address the people impact in the P&L plan from the outset.
Cancellations are coming. Gartner predicts that over 40 percent of agentic AI projects will be cancelled by the end of 2027, mainly due to rising costs, unclear business value and weak risk controls. Only a small fraction of vendors offer genuinely autonomous products. These are forecasts, not measurements, but the logic holds: projects bought on enthusiasm are the first to be cut when budgets tighten.
AI budgets are likely to grow, but only for projects that prove their value. Surveys point to a recovery in AI budget growth for 2027, with firms focusing on scaling proven, high-value use cases. CFOs are moving from treating AI as a pilot expense to integrating it into core budgets, with increasing pressure to show measurable results. While 75 percent expect larger AI budgets, only 31 percent rate outcomes as strongly positive, rising to 41 percent for projects in production. Funding will go to initiatives that can show their numbers.
The separate AI budget may disappear. Some predict that AI costs will move into the budgets of the departments that use them. This is an opinion, not a finding, but it makes sense. When AI costs sit in a department’s P&L, the business owner is accountable for the result, not just the technology team.
A 2027 plan should be specific. Each initiative should name the P&L line it aims to move, the size of the move and the executive accountable. Exit criteria should be set in advance, both for moving from pilot to production and for stopping underperforming projects. Full costs, including data, integration and people changes, should be in the same budget as the benefit. Boards should see results in financial terms, not just activity metrics. PwC analysis suggests companies applying AI to products, services and customer experience achieved margins nearly four points higher than those that did not. This is correlation, not proof, but it points to gains beyond cost cutting.
Treat early reports with caution. Earnings commentary is self-reported and selective. Savings figures may not be net of AI costs. The companies leading in the Janus Henderson sample are not typical; technology, industrial and financial firms have a head start. Use these numbers as a yardstick for your own questions, not as a promise of what your company will achieve. Test your own business against the same standards before quoting others’ results.
The real test is not whether AI works, but whether a company can prove where it works, especially under regulatory scrutiny. Investors want financial results, finance chiefs are tying budgets to evidence, and few organizations have rules for ending projects that fall short. CEOs who enter the next planning round with named P&L targets, agreed exit criteria, and costs and benefits in one budget will speak the language the room expects.
Which AI project in your 2027 plan would you stop first if it missaskL target, and who decides?
