AI Strategy
"Which Use Case First?" — The Question That Decides Your AI ROI
Almost every agent in the catalog can be built. The question that has not been answered — and the one that decides whether an AI program compounds or stalls — is which agent to build first.
The default question in every AI program meeting is whether the thing can actually be built. It's the wrong question. The agent catalog is full of well-documented patterns any competent integrator can ship in a few weeks. That question stopped being interesting eighteen months ago.
The unsolved question is which agent to build first. And it matters more than the building, because a first agent that doesn't return a defensible number kills the appetite — and the budget — for the second one before it has a chance to ship.
Most organizations answer this question by polling the executive team. Whichever VP has the loudest pain point gets the first agent. The hit rate of this method is approximately what you would expect.
An honest answer requires measurement. Specifically, measurement of where time actually goes — across the apps people use, the roles they hold, and the gap between the workflow on the slide deck and the workflow on the screen. Surveys don't surface that gap; the desktop does.
Dataken treats first-agent selection as a data problem. OLi watches work in flight, scores candidate use cases against observed time and friction, and produces a ranked shortlist that survives a CFO question. The result is sequencing argued from data, not from whoever spoke last in the meeting.
Key takeaways
- "Can it be built?" is no longer the interesting question — the agent catalog is mature
- The first agent's outcome determines whether the second one ever ships
- Polling the leadership team is the most common selection method and the least reliable one
- Real selection requires measurement of where time actually goes, not where people think it goes
- Sequencing should be argued from observed data, not from the loudest voice in the room