Only one company in eight consistently weighs what AI returns against what it costs

The PrompX BriefingN° 0729 October 2026On the table in ZurichOnly one company in eight consistently weighs what AI returns against what it costs Spending on AI has nearly doubled as a share of revenue in a year. The companies pulling ahead are not just the ones spending most. They decided in advance what AI would have to prove. By the PrompX Editors · 4 min read On 30 September, BCG published a number that deserves a place in every board meeting this autumn. Companies now spend about 3.3% of their revenue on AI, up from roughly 1.7% in late 2025. In a single year, AI has nearly doubled its claim on the income statement. The returns are real for some and uneven for most. BCG finds that the 7.5% of companies it calls future-built deliver 2.3 times the shareholder return of laggards. McKinsey’s annual survey finds 37% of respondents attributing at least some EBIT impact to AI, about the same as last year. And on 13 October, the ECB’s chief economist, Philip Lane, said it was too early to detect any material impact on aggregate productivity in Europe. So the question is not whether AI can create value. It is whether your organisation can tell, early enough to act, which of its AI investments are creating it. The figures 3.3% of company revenue is now spent on AI, up from about 1.7% in late 2025. The 7.5% of companies BCG calls future-built earn 2.3 times the shareholder return of laggards.BCG: 1,330 CxOs and senior leaders, 30 September 2026 12% of organisations consistently assess the value of AI against its cost. Among those already reporting returns, the figure is 48%.KPMG, AI Pulse Q3 2026: 2,131 senior leaders in 20 countries, 24 September 2026 8% of CFOs say finance leads enterprise-wide AI value goals and has automated investment triggers tied to the results. Nearly half, 48%, say finance tracks AI value and reallocates capital accordingly.IBM Institute for Business Value: 1,500 CFOs in 33 countries and regions, 30 September 2026 20–27% is the productivity gain technology leaders are capturing across key software development metrics. They expect developer productivity gains to nearly double, from 20% to 39%, within one to two years.Bain & Company, Technology Report 2026: survey of 293 technology leaders, 29 September 2026 What the companies seeing returns do differently They weigh value against cost, every time. The habit of consistently measuring AI’s value against its cost is four times as common among companies already reporting returns as among companies overall. Measurement does not create value. It shows where value is, early enough to move money towards it. Finance holds the pen. Nearly half of CFOs say finance already tracks AI value and reallocates capital accordingly. In the few that lead, the finance function sets enterprise-wide targets and has automated triggers that shift funds when results arrive. Only 8% of CFOs have reached that point. It is the clearest difference between AI as a programme and AI as an investment. They redesign work instead of deploying tools. Bain finds that as many as 90% of enterprises remain focused on deploying tools and narrow use cases. The gap between the gains leaders expect and the 20% to 27% they capture today is the cost of adding AI to old processes instead of rebuilding the process around it. The productivity already exists, at the desk. Fifty-two per cent of euro area workers now use AI at work, twice the share of 2024, according to the ECB, and they report significant time savings, though concentrated in specific sectors and tasks such as coding. That time rarely reaches the income statement on its own. Scaling AI means deciding what the saved time is for. Meanwhile, the IMF warned on 13 October that expectations around AI could prove too optimistic. Boards will ask for evidence sooner, not later. Before an AI pilot scales, it should prove three things: that it works, what it costs to run, and who will stop it if it does not.Three moves for the next ninety days Write the exit criteria before the pilot starts. Agree the measure of success, the ceiling on running costs and the date of the decision. A pilot without a decision date is not a pilot. It is a habit. Give the value case to finance. Let the CFO co-own the targets for AI value and review them quarterly, with the authority to move money from initiatives that miss them to initiatives that beat them. Fund a few workflows, not many tools. Choose two or three end-to-end processes and rebuild them around AI. Broad licences spread value thin; redesigned processes concentrate it where it can be measured. For your next leadership meeting Which of our AI pilots has a date by which it must prove its value, and what happens if it does not? Could our CFO tell us today the return on our three largest AI investments? Which process have we rebuilt around AI, rather than simply added AI to? Now, two minutes. Below this briefing are five statements. Answer them for your organisation as it is today, and see where you stand. Sources BCG, AI is starting to pay off, 30 September 2026. FM Magazine, Companies’ financial value from AI holds firm in 2026 (on McKinsey’s State of AI 2026), 1 September 2026. European Central Bank, Speech by Philip R. Lane, 13 October 2026. KPMG, AI Pulse survey, Q3 2026, 24 September 2026. IBM, As AI scales, CFOs play an expanded role, 30 September 2026. Bain & Company, Technology Report 2026: the AI-native enterprise, 29 September 2026. Bain & Company, Technology Report 2026: the missing architecture for agentic software development, 29 September 2026. International Monetary Fund, World Economic Outlook, Annual Meetings 2026: video news release, 13 October 2026. The PrompX Editors

Where does your organisation stand?

What must enterprise AI prove before it scales?

More briefings

The exit test: could your critical systems move in under a year?

Europe postponed its high-risk AI rules. It did not postpone the rest.

Most companies cannot trace an AI decision back to its data. Can yours?

What each euro of AI must buy in 2027