0147 skills
The building blocks
The smallest units of work. Retrieving information, classifying it, reading a document, spotting a deviation, drafting a reply, updating a system, asking a human to approve.
04the process
From scan to live in seven steps.
A hundred and eight roles are written out and waiting, across ten business domains. Implementing is not inventing, it is choosing and configuring: your knowledge, your systems, your rules, your approvals. That saves the weeks others spend on a blank page.
THE IMPLEMENTATION PIPELINE, from scan to a monitored employee
01AI Scan
02Workflow mapping
03Role selection
04Build & integration
05Approval & control
06Live implementation
07Monitor & optimize
→what you start from
The role exists before you call us.
Most AI projects start with a blank page. Ours start with a library: ten business domains, forty role families, and 108 roles written out as complete recipes. So implementation is not invention. It is selection, configuration and proof, and that is a far shorter road.
0247 handlings
Single handlings
Each does one narrow handling well: gather competitor ads, lift the fields off an invoice, report how certain the answer is. An employee chains dozens of them together.
03108 roles
The role library
Not features, but work you would recognise on a payroll: a customer care employee, a proposal preparer, a stock planner. Each one carries its own recipe before a customer ever asks for it.
04yours
Your employee
One of those roles, plus everything that makes it yours: your industry knowledge, your policy, your tone of voice, your systems, your approval rules. The recipe is shared. The configuration never is.
05combined
A department
Employees hand work to each other with evidence, a confidence level and a recommended next step attached. That is the point where it stops behaving like a tool.
06always on
Who may do what
Who may do what, what waits for a human, what every action cost, and a full trail of what happened. You can stop any employee at any moment.
THE TEN DOMAINS
→inside one employee
It checks before it acts.
A digital employee is not one clever instruction. Every piece of work runs through the same nine steps, in the same order, every time. At step seven it holds its own work against your rules. That is the reason step eight can be left alone.
The ninth step feeds the first: what it learns this week decides where it looks next week.
step 01 · watch
It watches where the work arrives
An inbox, a shared folder, a system reporting an event, a public source, or a number crossing a threshold. It does not wait for someone to forward it something.
For example: A mail lands on info@. It sees it straight away, you forward nothing.
Which systems it can watch →
step 02 · read
It lifts the fields that matter out of what arrived
From an attachment, a form or a conversation it pulls the data you need, and it remembers where each field came from.
For example: From a PDF invoice: supplier, amount, invoice number and date.
Where the data comes from →
step 03 · sort
What is this, how urgent, and whose is it
It determines type, priority and risk, and assigns it to the right queue or the right person.
For example: A complaint about a late delivery. High. Belongs to customer service.
Which roles do this work →
step 04 · retrieve
It gathers the context before it concludes anything
Which customer, which order, which agreement, which policy applies, and has this happened before. Without that context every answer is a guess.
For example: Order 8841, shipped on Tuesday, delayed at the carrier.
Where that context comes from →
step 05 · weigh
Compare, score, and reach a conclusion
It weighs the case against earlier ones and against the standard, and reaches a conclusion it can explain. Not a score, but a score with a reason.
For example: Third delay for this customer, so this is a risk case.
How a role reaches a judgement →
step 06 · draft
It prepares the draft
The reply, the record, the plan, the task. Still a draft: up to this point nothing has been sent and nothing has changed.
For example: A mail with the new delivery date, written in your tone of voice.
How it learns your tone →
step 07 · check
Before anything moves, it holds its own work against your rules
Does the evidence hold, does your policy allow this, does it fit your brand, and does it even have permission. Fail any one of those four and it goes to a human instead of out the door.
For example: It may not promise a discount above ten percent, so this one reaches you.
How the control layer works →
step 08 · act
Only now does something move
Send, file, update, schedule, or put it in front of you. What it may do alone is your call, and you can move it back at any moment.
For example: Mail sent, ticket updated, and you see it back in the overview.
What it may and may not do alone →
step 09 · learn
It records what it did and remembers the correction
Every action lands in the trail, gets measured against the agreement, and a correction from you becomes a lesson. That is why the questions decrease instead of continuing.
For example: You edited its reply. Next time it writes it that way straight off.
What it delivered →Click a step to hold it still and read on.
01day 1
AI Scan
We start with a structured scan of your operation: which workflows repeat, where hours disappear, which tools the work lives in. Output: the one workflow with the highest return to start with.
02week 1
Workflow mapping
We map how the workflow actually runs, not the idealized version. Inputs, decisions, exceptions, tools, and the moments where human judgment genuinely matters.
03week 1
Role selection & configuration
We don't invent a role. We take the one from the library that matches the workflow and configure it for you: exactly which tasks it takes over here, which systems it touches, what it may never do, and how success will be measured. You approve that before we build.
04week 2-3
Build & integration
The role already carries its own recipe. What we build is your version of it: the connections to your systems, your knowledge, your policy, your tone. Then we test it against real historical cases until its output meets the bar. Your team keeps its own tools. You get one dashboard that shows what happened, and what is waiting on you.
05week 3
Human approval & control setup
We configure the control layer: who approves what, role-based access, data boundaries, logging and fallbacks. Nothing critical executes without a human decision, unless you decide otherwise.
06week 3-4
Live implementation
The employee goes live in supervised mode: every action reviewed at first. As trust builds on real results, you decide where routine actions can run with lighter review.
07ongoing
Monitoring & optimization
We track performance against the agreed metrics, handle edge cases, and improve the workflow month over month. You see exactly what the employee did, and what it delivered.
→why this works
It goes live only once it could handle last month.
01
Never a blank page
We start with one scoped workflow, using a role that already exists, proven on your own data from last month. If it works, then we talk about the second. If it does not, you lost one workflow and not a programme.
02
Real cases, not demos
Employees are tested against your historical data before go-live. If it can't handle last month's reality, it doesn't ship.
03
Trust is earned in stages
Autonomy expands only where results justify it, and you decide where. Supervised first, always.
→common questions
About the process.
How much of our time does an implementation take?+
Plan on two or three short working sessions during mapping and design, quick reviews during the build, and a handover at go-live. We do the heavy lifting; you make the decisions.
What happens if mapping shows it's not a good fit?+
Then we tell you, before you commit to the build. A wrong pilot costs us more than a lost deal, so honest go/no-go advice is part of step two.
Can we pause or stop after go-live?+
Yes. Every employee has an immediate stop mechanism you can hit at any moment. The licence itself runs for six months, the trial period, after which you extend, scale down to a lighter licence or stop at no cost. Nothing about the setup locks you in technically either: your data and tools remain yours.