We scope the role,
build it, then run it.

The four steps

We go from a first call to a running role in four steps.

Nothing gets built until you have agreed the job description in writing. Your side of the work stays small at every step.

First we take one call about the jobs your team repeats, the tools you run and who signs off what. If an AI employee does not suit the work, we say so on the call and stop there.

Then we draw the role from the skill library and prove the role on your own live work. You name the jobs and the sources we may open. You see real output before you agree to run anything.

After that your team hands over a job in Microsoft Teams or Google Workspace, and the work comes back done. Anything you have gated waits for a named person to say yes.

From then on we run the role every day, tune the skills as the work changes and fix the mistakes. Each week a named person at Struan signs the report your team reads.

The first call commits you to nothing.

Before the build

The job description names who owns every part of the work.

You agree the job description in writing before we build anything. The job description says what the AI employee may open, what comes back and who approves.

You name the recurring work worth moving, and the AI employee touches nothing else. The AI employee reads only the inboxes, files and workspaces you approve.

Every job has a named output, so your team knows exactly what comes back. A named person in your team signs off the sensitive work before the work goes out.

Every decision carries a name before the work starts.

Prove the role

We prove the role on your own live work.

You watch the AI employee do the actual job before you agree to run the role. The test runs on your own live work.

You pick one named job and one approved source from the week's normal work. We run the selected skills, and your named person reads the output and checks the approval route.

Your named person and Struan then judge the output against the job description you agreed. You either agree the start, or agree the role is not ready.

Approval gates

You decide which work needs your yes.

A message to a customer waits in the chat until a named person approves the message. Gates come down one at a time, when you decide.

You write into the job description which work needs a yes, and who gives the yes. The AI employee finishes the work and then holds the work back until the yes arrives.

Approved work goes out. Anything unapproved stays put until your named person decides.

You set every gate and only you change a gate.

Fully managed

We run the AI employee and sign off the work every week.

Your team never picks up the running. We answer for the work the AI employee does.

We run the jobs at the times written into the job description. When something goes wrong, we fix the work and tune the skills, and your team never has to work out what went wrong.

You read one report a week, signed by a named person at Struan who answers for the work.

You pay one monthly fee, agreed in writing before we build anything. The fee covers the running, the tuning and the weekly sign-off. Nothing is metered and nothing gets added later.

Nobody on your team ends up managing the AI employee.

Keep control

You can stop the role with one instruction.

One instruction from your team stops the role. We follow a written stop procedure. Every action the AI employee took before the stop stays in the log.