Start with a useful artifact, a named approver and one handoff that needs continuity; then stay close enough to turn corrections into durable practice.
Learn how the work happens
Follow one process end to end with the people who do it. Begin with the request or event, then trace the evidence it needs, systems it crosses, decisions made, output expected and moments where someone stops, checks or explains.
The useful detail lives in the handoffs: a source in one inbox, a prior decision that needs context, an assumption carried by one operator or a follow-up tied to a deadline. The team’s knowledge is the material that should become durable.
Prioritise together
The customer team and build team agree which workflow goes first. A strong candidate lets a useful artifact reach someone quickly and passes the five tests: repetition, cross-system evidence, a visible artifact, named authority and a next run shaped by what the workflow learns.
Prioritisation sets the boundary. Define what the workflow may read, prepare and recommend, alongside the decisions that remain with a person. A regulatory watch, opportunity brief, event evidence room or research intake can each earn trust because its output and approval point are clear.
Map and connect
Name the evidence, source hierarchy, missing-input rules, objective, artifact, approver and systems crossed. Existing records remain in their authoritative systems while the workflow carries the context needed to prepare a reviewable result.
Define the retained memory: decision and reasoning, accepted evidence, unresolved risks, delivered output, corrections and next action. Status and provenance tell later runs what is current, provisional, superseded or waiting for a person.
Build, run and stay
Build with a dry run before external action. A workflow can start manually, from an event, on a schedule or through a webhook. A critical missing input parks the run and asks. A high-stakes proposed action routes to its named authority. The first useful run is a reviewable artifact.
Changes are versioned, and a boundary widens when evidence supports it. The account team runs beside the customer, watches where the model is incomplete and makes corrections visible. Ongoing support keeps the workflow aligned as sources and objectives change.
The starting delivery cadence is agreed with the team: learn how the customer works, agree the priority, map and connect the workflow, then build it and stay with it. The planned starting cadence is a workflow review in 48 hours, a first useful artifact in 7–10 days and an operating-memory map in 14 days. Scope, access requirements and delivery schedule are agreed with the named account team. Engagement begins when the workflow, owner and access boundary are clear.
What to retain
Measure the run as work. Record frequency, evidence handled, stop points, approver, artifact and what changes on the next run. Fewer repeated explanations, earlier visibility of missing evidence, a more complete decision packet or a follow-up that leaves one inbox can each be described with their actual context.
Keep proof status attached. A delivered client workflow can be anonymised. Customer zero demonstrates use inside the builder’s own business, while independent validation has a separate status. Architecture proof demonstrates controls; a figure becomes a measurement when its baseline, timeframe and sample are named.
The next workflow
Once the first workflow has a stable evidence model, approval model and retained state, its skeleton can guide another team. Each next workflow still receives its own objective, evidence, approver and boundary. Reuse transfers discipline, and each team earns its own permission.
That is how one workflow becomes operating memory: the organisation keeps enough decision, authority, evidence and result to begin related work from a prior position and update that position as the work changes.