Where to start automating commerce operations.

A practical way to choose the first workflows to automate, and the three tests that keep a good idea from becoming an expensive one.

Every commerce team has a list of things it would like to automate, and the list is almost always too long. The useful question is not what AI can do. It is which workflow to start with, so that the first result is real, visible, and leaves the team more capable than it found them.

Start with the work, not the tool

Walk through a week in each team: customer service, merchandising, finance operations, order exceptions, content. Look for work that is repetitive, shaped by rules, and high in volume. A good sign is a person following a checklist they could recite from memory. A bad sign is a task that depends on judgment nobody has written down.

Three tests for a first workflow

1. Is it frequent enough to matter?A workflow that happens twice a month will never repay the effort of automating it. Volume is what turns a small saving into a real one.
2. Can someone describe “done correctly”?If two experienced people disagree about the right answer, automation will only disagree faster. Clarity comes before speed.
3. Is the downside contained?Ask what happens when it is wrong. Start where a mistake is cheap and visible: drafts a person reviews before anything is sent, suggestions before actions.

Common starting points in consumer brands

These are illustrations, not a menu. The right first workflow depends on where your own team loses the most time.

• Customer-service triage and draft replies, reviewed by a person before they go out.
• Product-content enrichment: descriptions and attributes drafted at scale, checked before publishing.
• Returns and exchange handling, where the rules are clear and the volume is high.
• Order exceptions and inventory alerts that today depend on someone noticing.
• Reconciliation and reporting preparation that eats the first days of every month.

Design for the handover from day one

Decide who will own the workflow after it goes live, and what number it is supposed to move. Hours returned, cost per ticket, and cycle time are all good measures, but pick one before you start, not after. A workflow with a named owner, a measure, and a way to tell when it is drifting survives. Without those, it quietly decays once the people who built it move on. This is the difference between automating a task and building a capability.

What to leave for later

Be cautious with workflows that depend on systems whose data disagrees, and with customer-facing decisions that have no review path. They may be the right ideas at the wrong time. Put them on a separate list, write down what would make them ready, and come back.

What to do on Monday

List five workflows your teams would most like to be rid of. Run each through the three tests. Pick the one that passes, give it an owner and a number, and start there. If none passes, that is useful to know too, and it tells you where the groundwork is.