Who’s Grading AI’s Homework?

AI spending is not slowing down. Gartner expects worldwide AI spending to reach $2.7 trillion in 2026, up 49.5% on last year, and McKinsey found that 93% of organisations in its survey had gone over their AI budgets. Boards approved much of this on a promise. Now they are rightfully demanding proof of returns.

The question is who supplies it. Every AI vendor reports on its own tool, using its own definitions, and some now sell ROI tools that measure their own products. That is a student marking their own homework. No finance director would accept a number produced that way anywhere else in the business.

Adoption is on the up. Value isn’t.

Nearly nine in ten organisations now use AI regularly, yet only 37% can attribute any EBIT impact to it, a figure unchanged in a year. Deloitte sees the same gap on the ground. Access to sanctioned AI tools grew by half in twelve months, but fewer than 60% of the people with access use it in their daily work.

The firms getting returns are the ones measuring them. Gartner’s high performers track AI returns continuously, manage AI as a portfolio and stop what isn’t working, and they see positive returns on 81% of their initiatives. Low performers don’t know the return on 29% of theirs. Measuring the right thing matters too. McKinsey warns against “tokenmaxxing”, chasing consumption instead of outcomes. A count of users and tokens says nothing about whether your sales team is winning more business.

The gap is widest for mid-sized businesses. In McKinsey’s data, 54% of large organisations are scaling AI across the enterprise, against about a third of smaller ones. Mid-sized firms face the same board questions as the largest enterprises, with far leaner teams to answer them.

Three tests for an AI measure

The same rules for every vendor. A Copilot, Claude or Gemini user and AI usage should mean the same thing.
A published method. Finance and audit should be able to check every number.
Freedom to recommend spending less. If a tool can never tell you to cut a licence, it is a sales tool.

Keep your evidence your own

There is a dependence risk too. Gartner predicts that by 2028, 70% of enterprises will abandon agentic AI built through vendor forward-deployed engineering, as costs rise and internal teams struggle to change it themselves. Measurement carries the same risk. If the vendor builds the tool, holds the data and defines success, your evidence leaves when they do.

Why Europe should care most

In EMEA, 32% of companies rely on foreign suppliers for most of their AI stack, against 11% in the Americas, and 77% of organisations now weigh an AI supplier’s country of origin when buying.

Europe also asks more of the evidence itself. GDPR expects employee data to be kept to a minimum. Works councils in Germany, the Netherlands and elsewhere have a say over any technology that can monitor staff. The EU AI Act requires organisations to make sure their people are AI literate. European organisations need evidence they own, gathered without watching their employees.

How Temporall does it

Temporall has spent ten years doing the data engineering behind work and AI intelligence. We don’t resell any third-party technology, so we can tell you when the evidence says to cut a licence.

Our Work Graph applies one set of definitions across Microsoft 365 Copilot, Claude Enterprise, ChatGPT Enterprise and Gemini. It filters out activity that isn’t real use, and scores every person daily on AI use and proficiency. Our method is published, and every number links back to the vendor data it came from. Through MCP it connects to business systems such as ServiceNow, so you can link proficiency to results.

We never deploy as a browser extension or desktop application, and we don’t read documents, messages or calendars. Temporall is hosted in Europe and the US and certified to ISO/IEC 27001 and ISO/IEC 42001. It needs no data warehouse and no data engineers.

On the Enterprise plan, AI Governance is included from day one. It gives IT, risk, legal and finance one shared record: an inventory of every AI tool and agent, policy exception reports and an AI Literacy Register for EU AI Act Article 4.

Get an independent grade

Boards will keep funding AI. What they will stop accepting is evidence marked by the people selling it.

Our AI Baseline is a fixed-price, two-week engagement that gives you an independent view of adoption, proficiency and licence use, with every number traceable to its source. Temporall targets deployment in a week and a 90-day payback, depending on scope and realised savings.