How it works
You define the job. We manage the agent.
Managed AI workers for defined business tasks. We configure them around your knowledge, connect them to your systems, and run them against an agreed scope. You buy the output.
- 01
Choose the work.
Tell us the task, expected volume, and required output.
We start from a job, not a technology: what slows you down, how often, and what a finished piece of that work looks like.
If the task cannot be described, it cannot be priced. We will say so rather than sell you a pilot.
- 02
Agree the scope and price.
Define successful completion, connect the relevant knowledge and systems, and set approval rules.
We write down what counts as a completed output and how quality is judged. That definition is the thing you are buying.
We connect the approved knowledge and systems, and agree which actions need a person.
Any onboarding or integration fee is quoted here, before you commit.
- 03
Start and scale.
Review delivered outputs, track spending, and expand the work as results justify it.
You see what was delivered, accepted, escalated and charged. Rework caused by the agent is ours.
Expansion follows evidence — another workflow or an adjacent team, once the first job works.
Start with one workflow. Expand when it works for your business.
We would rather prove one job than launch a programme.
What onboarding actually involves
Honest about the first few weeks.
Hiring starts with a scoped enquiry, not a sign-up form. There is no self-service checkout, and we are not going to pretend otherwise.
Knowledge approval is usually the gate
Assembling and approving that material usually sets the timeline, not the technology.
System access takes as long as it takes
It depends on your environment and IT process. We scope it rather than guess optimistically.
Approval rules are a business decision
We bring a recommendation; you set the line.
Oversight
Your knowledge. Your rules. Your oversight.
Defined per engagement, in writing, before anything runs.
Approved sources
What the agent may read, and nothing beyond it.
Allowlisted actions
What it may do, defined action by action.
Review points
What waits for a person before it takes effect.
Escalation
When and how a human takes the request over.
Output reporting
What was delivered, accepted, escalated and charged.
Records
What the agent did and why, available for review.