If you're rolling out AI and not seeing a return on investment, this is probably the reason.
To get ROI, you need to roll it out organisationally, not functionally.
The gap shows up clearly in the research. In Gallup's workplace AI research, 65% of employees in organisations using AI say it has had a positive effect on their personal productivity. Yet only around one in eight strongly agree that AI has transformed how work gets done in their organisation. Those are two differently worded measures, not opposite answers to one question, but together they illustrate the point: individuals are getting faster while organisations barely change.
One of the reasons AI pilots rarely pay off is that you're only as fast as your slowest point. AI demos almost never show the handoffs between departments, because that is exactly where AI productivity goes to die.
Let me show you what I mean.
3x one team, and the work has to go somewhere
Say you use AI to 3x your marketing output. Nice. But that extra output doesn't just evaporate. It flows straight into the teams downstream: more leads for sales, more contracts for legal, more work for delivery, ops, HR and IT.
You made one team faster. You made every other team busier.
You're only as fast as your tightest bottleneck
If those teams haven't been upgraded too, the value gets stuck. Marketing is running at 3x, but if legal is still at 1x, everything past legal moves at 1x.
The extra work marketing created doesn't become value. It becomes backlog. Your system speed is set by your slowest critical function, not your fastest one.
You only get the ROI when every function can handle it
You don't get the return from 3x-ing marketing until sales can handle the leads, delivery can fulfil the work, and legal can process the contracts. Only when every critical function in the chain is running at 3x do you actually see 3x ROI.
Until then, you haven't created value. You've just moved the bottleneck around.
This is why AI has to be rolled out organisationally, not functionally
A functional rollout (one team, one tool, one workflow) gives you local wins and a system that's still stuck. An organisational rollout increases the speed of the whole system, and that's where the real, compounding ROI lives.
One team hits a wall. The whole organisation compounds.
A loop is a unit of organisational redesign
The unit of change should not be the individual AI user or even the department. It should be a repeated piece of work that crosses the necessary handoffs and ends in a useful business state.
That is why AI loops matter: they let the organisation redesign the work from trigger to outcome instead of making one participant faster.
A loop's "when done" forces the workflow to optimise for a cross-functional business state. A proposal loop doesn't finish when the document looks good; it finishes when the customer has a proposal it can accept and the business has a commitment it can deliver profitably. That definition drags sales, delivery, legal and finance into the same piece of work.
If you want to start pulling on this thread, begin with how to choose your first AI loop, see why small AI builds get adopted faster, and then look at how master loops and sub-loops scale AI across a business.
And it's why costs go up before they come down
Rolling out organisationally means paying for more than tokens. It's tokens across teams, plus enablement and training, workflow redesign, tech integrations, and change management and governance, usually all before the productivity gains land.
That's not a warning sign. It's the shape of doing it properly.
Bottom line: AI ROI isn't incremental. Stop optimising one team and start removing the bottlenecks across the organisation, so the whole system can run faster.
