Workflow automation on your own systems

A workflow is a route that work travels again and again. Something comes in, a few steps happen to it, and something goes out. A work order arrives, somebody reads it, looks up the job it belongs to, copies the hours across and books them. Four steps, every day, always in the same order.

As long as people walk that route by hand, every step costs time and now and then something slips through. Workflow automation means we put the route into code, exactly the way you walk it, and the system walks it as long as nothing unusual happens.

Every workflow begins with a trigger. In practice it is usually one of these: an email with an attachment in a shared inbox, a new file in a folder or on a drive, a row that changes in a system, a form filled in on your site, a message in a team chat, or simply the clock, every working day at seven.

We agree that trigger before we build, because it also decides what happens when things go wrong. A workflow watching an inbox has to know what it does with an email that looks almost, but not quite, the way it expected.

We build on the systems that are already there: the mailbox, the document folder, the accounting package, the ERP, the planning board, the CRM, and the spreadsheets that in practice still sit next to all of it. Through an API where there is one, and otherwise through the route that does exist, an export, a database, or a file dropped every night.

Moving to different software is not a condition for starting. That is rather the point: the route you walk today is the route we automate.

At the end there is something concrete. A booked line. A draft email waiting in the drafts folder. A report sitting in the team folder on Monday morning. A status updated in the system everyone already looks at.

Alongside that, a log. Per run it records what came in, which steps the system took and what went out. Without that log an automation is a black box, and nobody trusts a black box for longer than a fortnight.

The judgment. When the system is unsure about what it sees, it puts the case forward instead of guessing. Anything that goes out in your name or moves money is approved by a person before it leaves.

If something breaks, the workflow falls back on the manual route. An outage should not stop the operation; it should be a message saying today it has to go by hand.

Work that runs differently every time, and work where the contact itself is the value. Also work that comes back once a quarter: the maintenance on it costs more than it returns.

So we start with one route that runs often enough for the difference to show. Once that one works, the second is much quicker, because the connections are already in place.