Twelve AI initiatives with three-person teams is how AI programs stall. The math is unforgiving.
If your business has identified more AI opportunities than it can pursue simultaneously — which it should, if it did the inventory work — then sequencing becomes the actual strategic question. Not “should we do AI.” Not “which vendor.” Which project first, and what happens after.
Here’s the sequencing framework I use with clients. Three factors, applied in order.
First factor: readiness. For each candidate project, honest self-assessment: does the data exist and is it usable? Is the process well-defined enough to specify what “correct” output means? Is there a named owner in the business who will use the output? If any of those are no, this project is not ready, regardless of how attractive the ROI looks. Move it down the list. The projects at the top of your sequencing should be the ones where the friction is lowest, not the ones where the potential is highest.
Second factor: reversibility. Prefer projects where “this didn’t work” is a graceful exit. An automated report you can turn off next Tuesday is more reversible than a customer-facing chatbot integrated with your CRM. Early in AI adoption, choose reversibility over ambition. You’re building organizational muscle for AI, not just deploying one tool. The muscle grows faster when the early attempts are safe to fail.
Third factor: impact. Only after readiness and reversibility have narrowed the field — then rank by expected impact using the Return on Manual Work calculation. Impact matters, but it’s the third factor, not the first. Businesses that lead with impact and treat readiness as secondary end up with impressive-looking projects that stall in production and become the reason nobody trusts AI internally.
The rule that emerges: pick one small, ready, reversible project. Ship it. Measure it. Do the next one. The temptation to run multiple pilots in parallel is real — you want to explore the space, you don’t want to bet everything on one project, you want to prove momentum. Resist. Multiple parallel pilots divide attention, and AI adoption fails on inattention long before it fails on technology.
One project at a time, for the first three or four. After that, your organization has built the pattern-recognition to run parallel initiatives sensibly. Before that, you’re just diluting the wins you could be having.
Sequencing is not glamorous. It’s the discipline that separates AI programs that produce compounding value from ones that produce a portfolio of half-finished experiments. Choose the small win. Ship it. Learn from it. Do the next one. That’s the whole strategy.