Home/AI Capability/The Method

Methodology

Design one professional workflow you can run again and again.

One documented, repeatable, evidence-generating workflow you own, complete with standards, review points and a clear record of who decided what. The AI-Augmented Workflow Design Method is how you build it: ten stages, run once to design the workflow, after which you operate the workflow the method produced.

Read this before anything else

This is a method for designing or redesigning a workflow. It is not a checklist you run from Stage 1 to Stage 10 every time you do a piece of work. You run it once to build the workflow. Then you operate the workflow it produced.

Figure 1

The Workflow Design Method Map

The relationship between designing a workflow, operating it, and improving it. Most confusion about AI methodology comes from collapsing these three into one activity.

Workflow Design Method Map The ten canonical stages — Define, Recover, Map, Allocate, Engineer, Govern, Test, Validate, Document, Optimise — run once to produce an operating workflow. Stages 1 to 9 design and document the workflow; Stage 10 improves it using evidence gathered during operation. Escalation is not a separate stage: its rules are designed within Engineer and its ownership and evidence are governed within Govern. ARCHITECT, the knowledge recovery and codification standard, enters primarily within Recover. The operating workflow is then run repeatedly, which collects evidence, which feeds optimisation, which loops back into redesign. The method is applied in three contexts: internal operations, product generation, and client field work. 1 · Design the workflow — run once 01 Definetarget output 02 Recoverknowledge 03 Map currentworkflow 04 Allocatehuman & AI 05 Engineer theworkflow 06 Governstandards, gates 07 Test onreal work 08 Validatethe output 09 Documentthe system 10 Optimisewith evidence ARCHITECT enters here Escalation is designed in Engineer and owned in Govern: a control, not a stage. Stages 1–9 design the workflow · Stage 10 improves it from operating evidence. The operating workflow A documented procedure you own, adapt and keep. This is the asset. 2 · Operate, evidence, optimise — repeatedly Operate the workflow Normal professional work, run through the designed process. Collect evidence Judgment applied, AI contribution, validation, output quality. Optimise Improve the workflow because it was measured, not because it felt off. Redesign cycle 3 · Three application contexts — different ownership, different evidence Internal operations We design our own SOPs with it. Evidence: method evidence only. Cannot support client claims. Product generation We build reference workflows with it. Evidence: product evidence only. Cannot support client claims. Client field work A customer builds their own workflow. Evidence: client field-work evidence. The only class that supports outcome claims.

Figure 1 — Workflow Design Method Map. Scroll horizontally on narrow screens.

The design method

Ten stages, run once per workflow.

In four moves: Define the result → design the work → test it on something real → document and improve it. The ten stages below are how each move is done properly.

The ten-stage AI-Augmented Workflow Design Method
StageWhat happens
01 · DefineClarify the professional output, who it is for, the standard it must meet and the condition that counts as success. Most failed AI work fails here.
02 · RecoverCollect your knowledge, evidence, source materials, examples and constraints. Where knowledge is tacit or fragmented, this stage invokes ARCHITECT, the knowledge recovery and codification standard. This is the stage that turns a generic workflow into yours.
03 · MapDocument the current workflow, decisions, handoffs, delays, dependencies, failure modes and tools. You cannot redesign a process you have not described.
04 · AllocateAssign work and authority among the human, AI, deterministic tools, external specialists and independent reviewers. Deliberate allocation, not default.
05 · EngineerConstruct the future workflow: sequence, inputs, outputs, templates, prompts, file structure, version control and the escalation rules: the conditions under which work stops and returns to a named person. See below.
06 · GovernTranslate professional standards into rubrics, acceptance criteria, evidence requirements and quality gates, and assign escalation ownership: who receives returned work, what evidence accompanies it, and who authorises resumption.
07 · TestRun the designed workflow against a real project or a representative work sample, and capture failures, exceptions and support needs. Not a toy example.
08 · ValidateDetermine whether the workflow and its outputs meet their declared criteria — accuracy, originality, professional readiness — within the approved evidence meaning.
09 · DocumentProduce the operating manual, prompt pack, templates, SOPs, responsibility records, decision log and reuse instructions. You now have the asset.
10 · OptimiseImprove the workflow's design using evidence gathered during operation — performance data, reflection, feedback and new tool capabilities — without silently changing its governed version.

A control inside Engineer and Govern

A workflow that cannot stop itself is not a professional workflow.

Escalation is not a separate stage. Its rules are designed within Engineer (Stage 5) and its ownership and evidence are governed within Govern (Stage 6). Designing the stop conditions is as much of the work as designing the steps, and it is the part generic prompting has no concept of, because a model has no way of knowing what it would be professionally negligent to guess at.

Three questions a professional workflow must answer — When does it stop? Who receives the decision? What evidence travels with it? The stop conditions we design in:

  • Insufficient inputs — pauses and requests the missing evidence; does not fill the gap with plausible text.
  • Accountable human judgment — strategy, ethics, credibility and final acceptance stay with a responsible person.
  • Confidential or client-owned material — treated as a design constraint, not remembered later.
  • Rights and licensing — attribution, permissions and third-party assets reviewed before use.
  • Capability or performance claims: no evidence, no claim.
  • Specialist or regulated matters — legal, medical, financial and compliance escalate to a qualified professional.
  • Output quality failure — output that fails the standard goes back through revision or external review; AI output is never final by default.

Why these appear on every product page: a system that only describes what it can do is marketing. One that describes where it stops is a method.

Application contexts

The same method, three very different evidence positions.

We use this method three ways — on our own operations, to build products, and with customers — and those prove very different things. The full breakdown of what each context can and cannot support is on the evidence policy →

Ownership, stated plainly

A workflow you build in one of our programmes is yours. Your source materials, your knowledge and your resulting operating system remain under your control. If we ever want to use any of it as a product example or case study, we ask first, in writing, and we anonymise unless you tell us otherwise. Full terms →

Apply the method

The method is standardised. The workflow is yours.

Choose the route that matches how much support you want: self-study, built with us live, or coached through the difficult decisions.

Register for the Benchmark See programmes