Start With One Workflow: How to Pick Your First Automation

The most common way to fail at AI is to try to do all of it at once. "We're adopting AI" becomes a sprawling initiative across every department, nothing ships cleanly, and a year later there's a lot of activity and not much to show. The teams that succeed do the opposite. They pick one workflow, nail it, and build from there.

Choosing that first workflow well matters more than almost anything else, because it's where you build the muscle, contain the risk, and prove the value that funds everything after. Here's how to pick.

Why just one

Starting with a single workflow does three things at once. It lets your team learn how to build, deploy, and operate an agent on something small enough to actually finish. It contains the blast radius if you get it wrong. And it produces a real, measured result you can point to instead of a projection. Breadth comes later. The first one is about learning and proof.

The scorecard

A good first workflow scores well on most of these. Run your candidates through it.

High volume and repetitive. The work happens often and looks similar each time. Volume is what creates the leverage, and repetition is what an agent handles well. A rare, one-off task isn't worth automating first, however annoying it is.

A clear definition of good. You can say concretely what a correct result looks like. If you can't, you can't tell whether the agent is working, and you're not ready to automate it yet.

Low cost of being wrong. Pick something reversible and low-stakes for your first attempt. You want room to learn without a mistake being expensive or public. Save the high-stakes work for after you've earned trust.

Data is available. The information the workflow needs already exists and is reachable. If the data is missing or a mess, your first project becomes a data cleanup project, which is a fine thing to do but not the same thing.

You can measure it. There's a number, time, volume, error rate, that tells you whether it worked. Measurability is what turns your first project into the proof that funds the next one.

What not to pick first

Avoid the high-stakes, judgment-heavy, relationship-driven work for your opener, and avoid anything sitting on messy or contradictory data. Those are real opportunities, but they're advanced ones. Leading with them is how a first project becomes a cautionary tale.

This is Stage 3 of the Agentic OS Maturity Model done deliberately: your first real agent, on one well-chosen workflow. And the data criterion is the one people underrate, which is why cleaning your data before you automate so often turns out to be step zero. Pick the workflow where the 70/30 split is obvious and the stakes are forgiving, and you set yourself up to climb.

If you want help picking the first workflow that builds momentum instead of a cautionary tale, that's exactly where I start with clients. Let's talk.