The two white papers set out the principles. This section is about what happens when they meet a real firm on a Tuesday, with a partner who disagrees and a paralegal who has already built a workaround. Six phases, what each one is for, and accounts of how particular engagements actually went.
The order is not cosmetic. Each phase consumes the artefact the previous one produced, which is why firms that jump straight to choosing a tool have nothing to point at when it does not work.
The method opens with a choice, and it is not which tool to buy. It is which workflow to examine first. Firms almost always pick wrong here, because they pick the one that annoys the loudest partner rather than the one that will actually carry a result. It is a scored decision, made by one accountable person rather than a committee.
Produces: one chosen workflow, and the reason it beat the others
Not the way it appears in a training manual, and not the way a partner remembers it. The way it runs today, with every handoff, queue and piece of rework visible. This is the phase that produces the uncomfortable conversation, because the map and the firm's self-image rarely agree.
Produces: a step-level map of how the work actually moves
Every step on the map gets exactly one of three answers: Automate, Augment, or Abstain. The value is not in the labels, which are simple. It is in the discipline of the questioning that produces them, and in writing the reasoning down at the moment the call is made rather than reconstructing it later under pressure.
Produces: a classification for every step, with its reasoning on the record
Short enough that people actually read it. Not a strategy deck, not a memo that gets filed and forgotten, and not a statement of values. An operational standard with named owners and a defined line, so the boundary is still enforceable after the consultant leaves.
Produces: a written AI policy your bar counsel can read
The order is the whole thing. Change the wrong step first and you spend credibility you have not earned yet, and the programme stalls no matter how good the analysis was. Sequencing is what separates a policy that ships from one that is admired and ignored.
Produces: shipped changes, in a defensible order
Classifications are living decisions. Tools change, vendors change, and the regulatory ground moves underneath a policy written in good faith last year. Reviews are triggered by what happens outside the firm, not by a calendar reminder nobody honours. The tools will change; the way you decide where AI belongs is what endures.
Produces: a review that fires when something actually changes
These are narratives rather than tutorials. What the firm believed about itself, what the work revealed instead, which decision was argued about hardest, and what changed once it shipped. Enough that an owner recognises their own firm in it. Not enough to run the method from, because that is the engagement rather than an article.
Two reasons this section fills slowly
Consent. Nothing about a client firm appears here without that firm's written permission. Not the name, not the numbers, not a detail specific enough to identify them. That rules out the fastest way to fill a page like this, and it is the right trade: a firm that hands over its operations should not find them in someone's marketing.
Substance. The principles behind the work are published in full and free to read. The instruments that operationalise them are not, because a scorecard read without the judgment behind it produces false confidence rather than a decision. What you will find here is what the work felt like and what it changed, which is the part worth reading anyway.
Most owners can name the workflow that is costing them. Far fewer can say what it is costing. That gap is usually the first few minutes of the call.
