AI Strategy · Service 01
A method you cannot inspect is a method you cannot challenge. So here is the whole of it, all five phases: how we inventory your work, how every step is scored, how steps become named agents, how agents break down into buildable pieces, and how the resulting backlog is ranked. If you want to run it yourself, you can. The framework was never the scarce part.
Every workflow is inventoried, scored against four questions, clustered into named agents, decomposed into skills, and ranked into a build backlog you approve before anything is built.
Phase 01
We inventory every step in the workflow as people actually perform it, not as the process document claims it happens. This is done by sitting with the people who do the work rather than by reading the SOP, because the gap between the two is usually where the cost is hiding.
Nothing is judged in this phase and nothing is proposed. The only output is an honest map: every step, who performs it, what it needs before it can start, and what it produces. If the map is wrong, everything downstream of it is wrong, which is why this phase is not compressed to save time.
Phase 02
Four questions per workflow step. The point of scoring is not to find as many candidates as possible. It is to be able to say, with a reason, which steps should be left alone.
Does the step pull from more than one system, file or person before work can begin?
Does it involve comparing against targets, spotting anomalies, or applying consistent rules?
Does the output always take the same shape: a report, alert, summary or update?
Frequency is what turns a small saving into a material one. Daily is stronger still.
Three or four yes answers and the step is a strong candidate. Two and we watch it. Fewer and we leave it alone: most steps in most functions should not become agents, and saying so is the point of scoring them.
Phase 03
Strong candidates do not each become an agent. They are grouped, and a group only earns agent status when it shares all three signals below at once.
The same person needs the output of every step in the cluster. Two decision-makers means two agents, however similar the underlying data.
The steps draw on substantially the same sources. Steps reaching into entirely different systems get evaluated for separation.
The steps serve one outcome. Steps serving unrelated outcomes get separated even when the same person runs both.
An agent is only created when a cluster of steps shares all three signals at once. And it is named for the operational role it performs, never the technology behind it, because an agent nobody on the floor can describe is an agent nobody on the floor will use.
Phase 04
An agent is a role, not a program. Before anything is built, each one is decomposed into the discrete skills it needs: retrieve, analyze, produce an output, take an action.
This is the phase that makes the economics work. Some skills belong to one agent alone. Many are shared, and a shared skill is built once and reused by every agent that needs it. One skill, several agents, built once. Which skills are shared and which are role-specific is established here, before any commitment is made, rather than discovered halfway through a build.
Phase 05
Every item is scored on impact and on build readiness, one to five each, which places it in one of four quadrants. Shared skills are built first, because each one pays back across several agents at once.
Data available, access confirmed, logic understood. Commission immediately.
Worth building, but the data access has to be fixed before anyone writes a line of it.
Cheap to build. Runs alongside the high-impact work rather than instead of it.
Revisited in a later cycle. Naming this out loud is what stops scope drift.
Every skill is scored on impact and build readiness, one to five each. You sign the ranked backlog before anything is built, so the scope is yours, and so is the pace.
An executive briefing walks one of your processes end to end and gives you a written view on where an agent belongs and, just as usefully, where one does not.
Book an executive briefing →