The Market Just Asked AI for Proof of Effect
This is a reflection on what I see in my own work, not investment advice. I hold no view here on any stock.
On 28 October 2026, two of the largest AI spenders in the world reported their quarter on the same evening. The reactions went in opposite directions, and I think the reason matters far beyond the stock market.
Two reports, one question
Alphabet grew Google Cloud revenue by 103% to $30.7 billion. By most measures that is an extraordinary AI business. Yet free cash flow was negative $9.1 billion, against analyst expectations of roughly break-even to slightly positive, and it was the second negative quarter in a row. The company kept its 2026 capex guidance of $195–205 billion and said spending would “increase significantly in 2027”. The shares fell around 10% after hours and more the next morning (CNBC, Reuters via StreetInsider, CNBC, 29 Oct).
Microsoft grew Azure by 48%. Capex came in at $45.2 billion, below its own guidance of about $50 billion, and free cash flow was $18.6 billion, roughly $10 billion above expectations. The shares rose about 5% (Reuters via StreetInsider).
Both companies are winning in AI. The difference was not growth. It was whether the AI investment could already be seen as cash. The market did not ask “are you doing AI?” It asked “where is the effect?”
The question will not stay with the hyperscalers
I think this is the start of something broader. When investors start pricing proof of effect at the very top of the AI stack, boards and CFOs further down will ask the same question of their own AI budgets. Listed companies first, then everyone who reports to them.
And in many organisations I talk to, or have worked in, the honest answer today is uncomfortable. Here is what I see.
1. Double costs
The most common pattern is that AI is added, not substituted. The team stays the same, the old tools stay the same, the old reports stay the same, and a layer of AI licences and model spend is added on top. Productivity may well go up, but nothing is switched off. On the cost line, AI looks like a new department with no budget it replaced.
Proof of effect starts with a boring question: which cost disappeared when this went live?
2. Many pilots, little operation
The last two years have been a period of wide experimentation, and that was healthy. But a portfolio of twenty pilots with no owner, no operating model and no measurement is not an AI strategy. Recent surveys point the same way: Deloitte finds most companies report faster work but far fewer report transformed processes, and G2 finds oversight and rework eating a large share of the expected agent gains.
The hard part is rarely the model. It is ownership, data quality, monitoring, security and the decision to stop the old way of working.
3. More dashboards, not better decisions
The quietest failure mode is using AI to produce more of what we already had too much of: more dashboards, more reports, more SaaS seats, more summaries. It feels productive. But users rarely asked for another view. They asked to know when something needs their attention and what to do about it.
This is the shift I have spent much of my own recent work on. In a commercial B2B setting at a global manufacturer, we moved away from building more dashboards towards alerting with drill-down: the system tells you when a deviation matters, and lets you go from the signal straight to the underlying data. Fewer screens, faster action, and something that can be measured against the old way of working.
4. The platform should work for you, not the other way around
The second shift is in how the data platform itself is run. I have been building an agent-based setup where AI agents help build and maintain the data platform: pipelines, documentation, tests, quality checks. The point is not novelty. The point is that a capability which used to require a growing team of specialists can be run with a smaller, more senior core, with humans owning the decisions, the architecture and the risk.
That is where proof of effect gets real: not “we added an assistant”, but “this maintenance work no longer needs to be done by hand, and here is how we know the quality held”.
Effect is becoming the scarce skill
If the market keeps asking this question, and I think it will, the scarce resource will not be access to models. Everyone has that. The scarce resource will be people who can:
- pick the right use cases, the ones close to money, risk or customer experience, instead of the most impressive demos;
- remove old costs deliberately, so AI replaces work rather than sitting on top of it;
- get things into operation, with ownership, monitoring, security and GDPR handled from the start rather than after the pilot;
- show results in the language a CFO and a board understand.
None of this is glamorous. It is the same discipline that made earlier technology waves pay off: architecture, governance, change management and a willingness to say no to the tenth pilot.
What I would do on Monday
If I sat with a leadership team this week, I would suggest three simple steps:
- List every AI initiative and next to it write the cost it is supposed to remove or the result it is supposed to move. If neither column can be filled, pause it.
- Choose two that are closest to operation and fund them properly, including the boring parts: data quality, access control, monitoring, legal review.
- Decide upfront what gets switched off when they go live: a report, a tool, a manual process. Then measure it.
The hyperscalers will sort out their own capex. The more interesting question is what happens inside every other company when the same lens is turned on them. I would rather be ready for that question than surprised by it.