How to measure returns on AI
The article argues that token use and other input metrics show adoption, not value; firms should measure outcomes such as productivity and customer satisfaction while guarding quality. AI often produces a J-curve—initial disruption before gains—so good measurement also tracks implementation health, employee response and in-house capability. For financial returns, avoided hiring is easier to attribute than incremental revenue, which needs baselines and A/B testing.
Why it matters
Dealroom should measure AI by faster and better workflows plus avoided hiring—not prompts or seats—without killing experimentation before the J-curve turns positive.
Executive takeaways
- AI adoption is entering its grown-up phase: usage is not ROI; quality, the J-curve and organisational learning matter.
- Europe’s dependence on American software and defence is concentrated in critical systems; sovereignty requires real alternatives and portability, not slogans.
- AI is shifting value from web distribution to scarce human data: Reddit can license the corpus, but AI answers are already eroding the traffic loop that created it.
- Much of the anti-data-centre backlash is political theatre around speculative projects, while chips, workers, power and credible developers remain the binding constraints.
- Europe’s strategic risk is increasingly operational: Germany’s exclusion strategy is strengthening the AfD, while interceptor shortages are turning Russian strikes into direct economic warfare against Ukraine.
What The Economist may be missing
- The AI-ROI framework underweights option value and second-order product or revenue gains; premature scorecards can optimise away the experimentation that creates them.
- The sovereignty and Palantir arguments overweight incumbent performance and underweight bargaining asymmetry, open standards, portability and procurement’s role in creating credible European suppliers.
- The data-centre argument is very American: in Europe, grid connections, permitting and social consent can be hard constraints rather than electoral theatre.
Read the full article: The Economist
