Customer service workflows and AI automation

A workflow starts at a trigger and runs from there through the steps you put behind it. You drag them into place, without code, and what you draw is exactly what runs.

You can also describe what the process should do and Cuego puts down the first version for you. You then adjust it as normal: easier to start from something that already exists than from an empty canvas.

The steps each do one thing. Pulling data out of a message, looking up a customer or order, querying your webshop, sending an email, creating a task, or having the AI phrase something.

There are two ways to branch, and the distinction matters. On a hard condition, such as whether an order number was found at all. Or on what the customer turns out to want, where the AI sorts the message into categories you described.

A hard condition is predictable and an AI classification is flexible. So put the irreversible steps behind the hard condition.

A workflow may look up a status and send an update by itself. On a step that changes something in your own system, such as a refund, you can put an approval in between.

You make that choice not on how complicated the step is, but on what it costs when it goes wrong. A wrongly sent status update you correct with a second message. A wrong refund you have to claim back from someone who already received the money.

Every time a workflow runs, the run is kept, and per step what came out of it. So you see which trigger fired, which way a branch went and what each step returned.

Without that history an automation can't be repaired. You know a customer got the wrong message, but not at which step it went wrong, and you start guessing.