Three weeks, fixed fee. We walk your operation, find the work that is high in volume and thin in judgement, and hand you the first three agentic applications worth building, each with a value attached and an honest account of what has to be true for it to run.
A licence gets bought, a pilot gets built, and eighteen months later nobody can say what changed in the business. The reason is almost always the same: the work was never picked on purpose. We start at the other end, with the processes that carry real volume, and we say which of them an agent can take, what that is worth, and what it depends on.
The work that repeats hundreds or thousands of times a month. Volume is what makes an agent worth building and what makes the number defensible afterwards.
We mark the points in each process where a human has to stay. Those become gates. Knowing them in advance is what stops a build turning into a rollback.
Every application rests on data someone can vouch for and a process someone owns. We state those conditions up front rather than discovering them in week six.
One document your leadership team can fund from. Three candidate applications, ranked, with the case for each written the way a budget request is written.
Named against a real process, with the volume it carries, the people involved today, and where the agent takes over. Specific enough to brief a build team from.
The value in hours, throughput, cycle time or error rate, calculated from your own figures. If a candidate cannot be quantified, we say so and it drops.
The data it needs and who owns it, the gates a person keeps, the systems it has to reach, and the obligations that apply to it. Per application.
Which one goes first and why. Usually the one with the shortest path to a number, because the first number is what funds the second application.
| Candidate | Volume | Gate a person keeps | Depends on |
|---|---|---|---|
| Invoice exception handling | ~2,400 / month | Credit notes above threshold | ERP is source of truth |
| Tender response drafting | ~40 / month | Final submission | Document store is contested |
| Field report triage | ~9,000 / month | Safety escalations | No owner for asset data |
| Supplier onboarding checks | ~180 / month | Approval to trade | Registry access in place |
Six to ten conversations with the people who do the work, not only the people who manage it. We are looking for the tasks everyone complains about and nobody has counted.
Volumes from your own systems, the judgement points inside each process, and a first read on whether the data underneath will carry an agent. Candidates that cannot be quantified come off the list here.
We write it up, walk your leadership through it, and leave you the document. It is yours whether or not we build anything.
Our working paper calls that a principal company: the people who decide what good looks like stay in charge, agents do the volume, and a record shows what happened. It is not a philosophy exercise. It is the reason one company gets three agents into production while another is still running a pilot.
The Agent Case is the first step towards operating that way, and it is deliberately a small one. You get a costed plan for three applications before you commit to building any of them.
Three routes are normal. One of them is doing nothing further with us.
You have a costed plan and your own team builds from it. The fixed fee bought a deliverable and no obligation.
Four weeks, fixed fee, the top candidate running in production with the record on from the first day.
Where the case surfaced a data problem big enough to block all three candidates, this is the engagement that sizes it.
No maturity model, no five-year roadmap. Three applications, three numbers, one order to build them in.
We do not arrive with a shortlist of vendors. The question is which work is worth automating, not what to buy.
Obligations appear as conditions on a build, in the place where they change what you do about it.
Three weeks, fixed fee, three costed applications. Most clients use it to decide what to fund next.