An annual report is written for shareholders, financiers and regulators, not for competitors who want to know where the company scores well. Yet it contains more than most readers pick up. Investments in software and data infrastructure, disclosures on efficiency gains, staff development by department, and risk sections on technology together give a picture of where a company is shifting work. The picture is incomplete and delayed, but it is not nothing.
The reason this is relevant does not lie in AI as a theme. It lies in what AI does to the comparison between providers. As long as a dimension such as delivery time, response time or after-sales support depends on how many people a company puts on it, the comparison is a comparison of staffing and organisation. The moment part of that work is done by AI, with or without human oversight, the dimension shifts to something that depends on systems, data and the quality of oversight over them. Whoever sees that before the market does knows which claims from competitors still hold and which have since become outdated.
Several items in annual reports can be traced directly to AI deployment:
None of these items say with certainty that a task has been taken over. They say that money, people or attention have gone towards a shift. What has factually been taken over, what happens under oversight and what remains human work is not stated.
The most common mistake is equating investment with result. A company that reports having invested in AI applications says nothing thereby about whether those applications are already running in production, let alone about the size of the effect on a specific dimension. Investments are announced before they yield anything, and sometimes also when they yield nothing.
The second mistake is interpreting a declining headcount as proof of automation. A decrease in FTE can relate to a reorganisation, a divestment of a business unit or a shrinking market. Where such decreases touch on personnel policy, their own legal requirements apply; the annual report itself does not substantiate a dismissal decision, and that is not what it is written for either.
The third mistake is giving too much weight to the absence of AI terminology. An annual report that does not mention AI does not say that a company does not use AI. Some companies do not report it because it has become operational and is no longer seen as special; other companies do not report it because there simply is nothing.
Annual reports appear annually or semi-annually and are therefore slow relative to how quickly AI applications change in practice. For a company that launches a new application every quarter, last year's annual report is already partly outdated by the time it is read. For companies that invest less frequently, the information remains valid for longer.
The source is therefore mainly suited as confirmation alongside other signals, not as a first or only source. Where price lists show what a shift does to margins and rates, where tenders and public procurements show how a competitor positions itself in a concrete assignment, and where press releases show what a company itself wants to show, the annual report provides the financial and organisational context within which those other signals can be placed. Product documentation also helps here: where the annual report reports an investment, documentation shows whether that investment can already be found in a delivered product.
An indication from an annual report only becomes usable once it is placed alongside your own position. If it turns out that a competitor is structurally investing in a dimension on which you currently still lead, that is a signal to determine what you do with a lag on a dimension before it actually arises. And if you want to claim yourself that you work faster, cheaper or more accurately than the competition, it is worth first reading how you substantiate a claim about your own quality, before a competitor refutes it on your behalf.
The underlying question, which work in your own company can genuinely be taken over by AI, is not answered by an annual report; the work scan from FTE TO AI does that per task, with the distinction between fully transferable, partly with oversight, or human work.
You can now, without opening an annual report, do the free dimension check: you name where you believe you are winning and see which of those claims can be defended with evidence. The full benchmark, with your peer group's score on every dimension, is under construction.