Choosing your first AI use case: boring beats impressive

Asked which process to automate with AI first, most leadership teams pick the most visible one: the customer-facing chatbot, the AI feature for the sales deck. It is the single most reliable way to join the failure statistics. The right first use case is almost always boring, and its boringness is precisely what makes it succeed.

The reason is mechanical, not aesthetic. Your first AI project carries a second job on top of its actual job: it sets the organization’s belief about whether AI works here. A boring project that ships and saves 300 hours a quarter buys you the mandate for the ambitious ones. An impressive project that stalls poisons the well for two years. Optimize the first move for probability of success, not for applause.

The four criteria that predict success

After enough projects, we score candidates on four questions.

High volume. The process happens hundreds or thousands of times a month. Volume is where the ROI lives, and it also means abundant real examples to build and test against from week one.

Clear ground truth. For any given input, competent people agree on what the correct output is. Invoice data extraction has ground truth. "Write a good marketing post" does not. Without ground truth you cannot measure quality, and without measurement you cannot prove the system works or catch it degrading.

Tolerance for review. The workflow can absorb a human checking the AI’s output, at least initially, without destroying the economics. A system drafting responses for your team to approve can ship at 90 percent accuracy. A system acting directly on customers needs a bar you should not attempt on project one.

Measurable baseline. You know, in numbers, what the process costs today: hours per week, error rate, backlog age. This is what makes the after comparable to the before, and it is the foundation of any honest ROI claim, as we argued in measuring AI ROI without lying to yourself.

Score each candidate 0, 1, or 2 on each criterion. A candidate at 6 or above is worth piloting. Below 5, keep looking. It takes an hour with your ops leads, and the ranking it produces is usually surprising: the winner is rarely the process anyone was excited about.

Why the impressive use case fails first

Run the flashy candidates through the same grid and watch them collapse. The public chatbot: stakes are maximal because errors reach customers, ground truth is fuzzy, review tolerance is zero because responses are live. The strategy copilot for executives: low volume, no ground truth, no baseline. These fail the grid on every line, yet they dominate first-project selection because selection happens by demo appeal.

There is a second-order cost too. High-stakes projects attract oversight, and oversight without early wins turns into slow death by review committee. The boring project ships while the impressive one is still in legal.

Where to look for the first process to automate

Look where documents meet decisions in your operations. Invoice and order processing, claims intake, KYC checks, contract review against standard clauses, support ticket triage and drafting, RFP first drafts, reconciliation and exception handling in finance. These score high on all four criteria, which is why document processing is the least sexy, highest ROI use case in most portfolios.

Two reassurances worth giving your team. First, imperfect data is rarely disqualifying: modern systems handle messy documents well, and readiness is a per-workflow question, not a company-wide prerequisite, as we covered in is your data ready for AI. Second, boring does not mean small. A mid-size insurer processing 40,000 claims documents a year is looking at thousands of hours; the process is dull and the number is not.

Boring is the strategy, not the compromise

The sequence that works: pick the highest-scoring boring candidate, pilot it on real data with the graduation criteria decided in advance (the discipline in scoping a POC that can actually ship), ship it, publish the measured result internally, then let the win fund the next project. Ambition comes third, after evidence and after trust, and it comes faster this way than by aiming high first.

If you want a second opinion on your shortlist, ranking candidate use cases is how our custom AI engagements usually begin, and the scoring conversation takes an afternoon, not a strategy phase.