Every AI roadmap we’ve seen starts with the same kind of item at the top: a customer-facing assistant, a personalization engine, an agent that runs campaigns on its own. Those are the use cases that get the budget approved, because they’re the ones that look like the future. They’re also, nearly every time, the ones that take the longest to return anything. The use case that pays for itself first is further down the list, and it usually isn’t on the list at all, because it’s too boring to write down.
Why the Exciting Use Case Is Slow
A customer-facing AI feature has to be right almost all of the time, because the cost of being wrong lands on a customer. That means evaluation, guardrails, review cycles, legal sign-off, and a long tail of edge cases that each need a decision. It also touches systems that were never designed to be touched: the CRM, the support platform, the product itself. None of that is a reason not to build it. It’s a reason to expect the return in quarters, not weeks, and to fund it from somewhere.
We’ve written before about why pilots stall. The short version is that the gap between a demo and a system is mostly integration and trust, and both take time.
What “Boring” Actually Looks Like
The use cases that return first share three properties: the output goes to a colleague, not a customer; a person reviews it before it matters; and the task already happens every day whether AI is involved or not. In practice that means things like:
- Internal lookup. “What did we tell this client about reporting cadence in the kickoff?” answered from the meeting notes and email thread, instead of fifteen minutes of searching.
- Meeting summaries and follow-ups. The notes, the action items, the recap email. Written in a minute, checked in two, sent.
- First-draft reporting. The monthly performance narrative that someone currently writes from scratch, drafted from the numbers and then edited by the person who understands them.
- Reformatting and cleanup. Turning a transcript into a brief, a brief into a slide outline, a spreadsheet export into something readable.
- Search term and query triage. A first pass at sorting a search terms report into obvious junk, obvious keepers, and needs-a-human, before anyone opens it.
None of these will be in a press release. All of them return time in the first week, and time is the only currency an AI initiative can spend before it has earned anything else.
Why This Order Matters Beyond the Return
Starting boring isn’t just about a faster payback. It’s how a team learns what the tools are actually good at, on tasks where being wrong costs a correction rather than a customer. The person who has spent a month editing AI-drafted recaps knows, from experience rather than a vendor deck, where the drafts go wrong and how much review they need. That knowledge is what makes the exciting use case buildable later. Skip the boring phase and you build the ambitious thing on assumptions.
It also settles the workflow question early. Meeting summaries force a decision about where notes live and who reviews them. Internal lookup forces a decision about what’s allowed into the tool and what isn’t. Those are the decisions the ambitious use case will need too, and they’re cheaper to make now.
How to Find Yours
Ask each team one question: what do you do every week that takes an hour, produces a document, and gets read by one person? The answers are the list. Rank them by how often they happen, not by how interesting they sound, and start with the top two. Give it thirty days, measure the hours, and let that number make the case for whatever’s next.
The ambitious use case is still worth building. It just shouldn’t be the thing that has to prove AI works, because it’s the worst-positioned thing to do that quickly. That’s the sequencing we bring to AI strategy work: the roadmap stays ambitious, and the first item on it is the one that pays for the second. If your roadmap starts at the top, let’s talk about reordering it.