Consulting
Pick the one workflow worth automating, and decide its controls first
You can probably name five jobs an AI agent might do and not know which to start with, or what the agent should never touch. The sprint settles both in about two weeks. You finish with one chosen workflow, a plain list of controls and a scope someone can build from.
What the sprint does
One workflow chosen, its controls written down
We sit with you and the people who do the work and look at the candidate jobs. We pick the one where an agent saves the most time for the least risk. Then we write down what the agent may read, what it may change, which steps a person must still approve, and who owns it once it is running.
You can read the result without us in the room. Each choice has the reason beside it, so you can take the scope to us, to another builder or to your own staff.
Fit
Who the sprint suits
It suits the owner or operator who decides what the business spends time and money on.
The buyer
- An owner or general manager who can approve a project.
- A team lead who knows how the work runs day to day and will look after the agent afterwards.
- A bookkeeper, office manager or compliance contact, when client data is involved.
The moment it helps
- You have several ideas for AI and no way to rank them.
- Someone on staff already pastes client details into a chatbot and you want rules.
- A tool demo impressed you and you want to know if it fits your own work.
- You plan to set up an agent team and want to know which skills to build first.
What is an AI governance sprint?
A short, fixed-scope piece of consulting that decides whether one workflow should go to an AI agent, and under what controls. It measures how the work runs today, checks what data the agent would touch, and sets the limits and the review steps. It ends in one of three decisions: build it, fix a prerequisite first, or leave it alone with the reasons written down.
What we need from you
What you bring
The sprint stays short because these are agreed before it starts.
Show the detail
Five things to line up before day one.
- A decision-maker
- Someone who can approve a build, present at the start and at the final readout.
- A person who does the work
- Someone who knows how the job runs, for about a day of their time across the sprint.
- Two or three colleagues
- People who do or depend on the workflow, for one conversation each.
- Numbers from the business
- How many requests, how long each takes, how much rework. Rough figures from your own records are fine; guesses are not.
- Access, or samples
- Read access to the files and systems involved, or a set of sample documents with names and client details removed.
Method
How a workflow is scored
Seven dimensions, each scored one to five with the evidence written beside the score. If a workflow scores below three on Value, Ownership or Risk, we advise against building it, whatever the total.
Show the detail
Seven dimensions, each scored one to five with the evidence beside it.
| Dimension | The question we ask |
|---|---|
| Value | What does the work cost today in hours, delays or rework, and who will judge the result? |
| Ownership | Who can approve the project, and who will look after the agent once it runs? We name a person, not a team. |
| Data | Which files and systems would the agent read, who owns them, and how current are they? |
| Risk | Which steps stay with a person, and what do your clients and your contracts allow the agent to do? |
| Evaluation | Can we assemble real cases with known right answers and agree a pass mark? |
| Integration | Can the agent work in the tools your staff already use, without a separate project? |
| Adoption | Will the people who do the work use it, and who collects their feedback? |
The scoring sheet is not secret. The workflow scorecard lets you rate a candidate workflow yourself before you contact us.
Deliverables
What you receive at the end
One written pack in an editable format, readable by someone who was not in the room.
Show the detail
One written pack in an editable format.
- How the work runs today
- The steps, the volumes, the time each takes, and where each figure came from.
- The scored list of candidate workflows
- Each one rated on value, ownership, data, risk, how it will be tested, how it connects and whether staff will use it, with the recommendation.
- The control plan
- What the agent reads and changes, which steps a person approves, and who owns each decision.
- A test plan
- Real cases with known answers, and the result that would count as a pass or a fail.
- A build scope
- What to build first, what to leave out, what you must provide and how it will be accepted. We can build it, or you can hand it to someone else.
How it is sold
Sold through Consulting
The sprint is part of our Consulting work. It suits a business that wants to choose before it builds.
You start with a 30-minute call at no charge. If the sprint fits, we scope it with you in writing and quote it before any work begins. The price depends on how many candidate workflows you have and how many systems and people are involved.
If you would rather go straight to building, our AI team setup has fixed prices and a guarantee. Each skill passes the test you sign, or we fix it free.
Show the detail
What counts as finished, and what the sprint leaves out.
The sprint is finished when
- The five deliverables are in your hands in editable form.
- Every figure in the baseline names its source.
- Each control has a named owner inside your business.
- The test plan has a pass mark you have reviewed.
- The decision-maker has seen the readout and recorded a decision.
Not included
- Any build, integration or agent configuration.
- A legal opinion. We note the exposure for your lawyer to advise on.
- A security audit or certification.
- Company-wide AI strategy.
Next step
Three possible endings
The readout produces one of these, in writing, with the reasons. We are paid the same for all three.
- Build it. The workflow scores well, the controls are settled and you have a scope ready to contract, with us or with someone else.
- Fix a prerequisite first. The workflow is right but something is missing: a clean customer list, an owner, a baseline. You get the list, a name beside each item and the point at which we reopen the decision.
- Leave it alone. The value is not there, the risk is too high or the data cannot support it. You get the reasoning in writing, so you can show later why you chose not to proceed.