What to automate first.

When we walk into an operation for the first time, the list of things that could be automated is never short. Quotes retyped from email into the ERP. Tickets triaged by hand every morning. A report assembled from four spreadsheets on a Friday afternoon. The temptation is to start with the biggest leak. We almost never do.

Start where a mistake is cheap

The first workflow sets the tone for everything after it. If it goes wrong in a visible way, the programme loses trust it will take months to earn back. So we look for the process where an error is annoying rather than expensive: a wrong internal reminder, a mislabelled ticket, a draft that a person still reads before it goes out.

That rules out invoicing, pricing and anything customer-facing on day one, however attractive the savings look on paper. Those come second, once the team has watched an agent behave well for a few weeks.

Prefer the boring, frequent task

A task that happens two hundred times a day is a better first candidate than one that happens twice a month, even if the monthly one is more painful. Frequency gives the agent enough real cases to be measured on, and it gives your team a daily reason to notice that something has changed.

• It has a clear input. An email, a form, a file landing in a folder.
• It has a clear output. A record created, a reply sent, a field updated.
• The rules can be written down. If two experienced people disagree about the right answer, it is not a first workflow.
• Someone owns it today. That person becomes the approver, and later the champion.

Keep a human in the loop on purpose

The first agent we ship almost always proposes rather than acts. It drafts the reply, fills in the record, flags the exception, and a person clicks approve. This feels slower than it needs to be, and that is the point. Approval traffic tells us exactly where the agent is unsure, and it tells your team the system is not going to surprise them. Before any of that, the agent runs in shadow mode and touches nothing live.

The goal of the first workflow is not to save the most time. It is to make the second workflow easy to say yes to.

What we measure

Before anything is built we time the process as it runs today: how long a case takes end to end, how many hands it passes through, how often it is retyped. Those numbers come from your operation, not from an industry benchmark, and they are the only ones we report against later.

If you are choosing between candidates right now, pick the one that would make one specific person's Monday quieter. Then tell us about it.