The method
Study the work before you touch the tools.
Five phases, run in order. The discipline is old — field research inside operating companies — and the object of study has changed. It used to be software. Now it is the shape of the work itself.
Discovery
Interviews across every altitude of the function, plus direct observation of the work being done. Executives describe what the function is for; managers describe where it jams; the people doing the work show what actually happens, which is reliably different from both.
- Bottlenecks, queues, and waiting time between steps
- Duplicate entry and reconciliation across systems
- Handoffs, approval chains, and where authority actually sits
- Knowledge silos — what only one person knows how to do
AI opportunity mapping
Every task in the mapped workflow gets classified and sized. This is the phase that separates a real plan from a vendor pitch: the classification is driven by what the work requires, not by what a tool can demo.
- Eliminate — drafting, summarization, meeting notes, routine documentation, repetitive lookup
- Accelerate — research, contract and policy review, technical support, compliance checks, with a human deciding
- Protect — safety decisions, regulatory approvals, final financial signoff, anything requiring accountable human judgment
Workflow redesign
The future state gets drawn as a process, not a tool stack. Steps collapse, handoffs disappear, and the remaining human decisions are positioned where they carry the most weight. Only once the process is agreed does software enter the conversation.
A redesigned intake flow, for example, stops looking like a relay between departments and starts looking like a single path: request, automated triage, retrieval of what is already known, risk scoring, a human review that matters, and documentation that writes itself.
Build and orchestrate
Rarely a build from scratch. Most of what an operation needs already exists and has to be selected, connected, and constrained properly — enterprise assistants, automation platforms, retrieval over your own documents, agents wired into the systems of record you already run.
Your engineers or a delivery partner do the building. My job is the specification, the sequencing, the evaluation criteria, and making sure what ships matches the workflow that was designed.
Measure
Baselines are taken during discovery so that improvement is provable rather than asserted. Reporting uses the vocabulary your board already speaks.
- Hours saved, by task and by team
- Cycle time and first-response time
- Error and rework rates
- Cost avoided, and headcount redeployed rather than removed
Principles
Four commitments that shape every engagement.
Work first, tools second
No platform is selected before the workflow is understood and redesigned. Tools are consequences of decisions, not substitutes for them.
The boundary is a deliverable
What AI must not decide gets written down explicitly, with reasoning, so it survives staff turnover and vendor pressure.
Measured in operations terms
Hours, cycle time, error rates, cost. Never adoption dashboards, model benchmarks, or enthusiasm.
Nothing that needs me to continue
Maps, roadmaps, and specifications are written so your team or another firm can execute them without me in the room.
Next step
The method starts with an assessment.
Phases one and two are packaged as a fixed-fee engagement, so you can see the quality of the thinking before committing to anything larger.