0
Orientation
Who are we working with and what are the ground rules?
Scope, compliance constraints, data access, stakeholder map, communication strategy. Nothing proceeds without this foundation.
Framework
Framework.
A structured seven-phase framework from orientation through expansion. One workflow proven before the next begins. The right tool selected for each problem.
Phases
The seven phases.
Each phase has a question it answers and an artifact it produces. The next phase does not start until the current one closes.
1
Workflow Discovery
What does the team actually do?
Interviews across leadership, team leads, and individual contributors — separately. The real process rarely matches the documented one. Shadow systems, workarounds, and single points of failure surface here.
2
Assessment
Which workflows are worth targeting?
Every discovered workflow scored across eight dimensions: labor intensity, current cost, return potential, frequency, AI suitability, implementation complexity, data availability, and error tolerance. The matrix drives the recommendation. The real insight comes from the conversation around divergent scores.
3
Selection & Prescription
What are we building, and who owns it?
One workflow selected. The right technical approach prescribed from the full range — simple automation, ML, LLM, or hybrid. A Process Champion assigned from inside the organization. Success criteria agreed before a line of code is written.
4
PRD
What exactly are we building?
A product requirements document detailed enough to build from without ambiguity. Current state, future state, user stories with acceptance criteria, data requirements, integration points, edge cases, failure modes. The PRD is also the onboarding document for the tool — it specifies how the AI is trained for this workflow.
5
Build — Test — Iterate
Does it work?
Three structured cycles: internal validation, champion review with real data, then live use. An iteration log tracks every change — what changed, why, who decided, and what happened. Prompt changes, threshold adjustments, process refinements — all versioned. MVP acceptance is a defined gate, not a feeling.
6
Expand
Where do we go next?
Retrospective on the completed workflow. Lessons documented. Assessment matrix revisited with new information. Next workflow selected. Parallel tracks assessed when the team is ready for them — they are earned, not assumed.
Principles
Three keys to success.
The framework is the shape of an engagement. The principles are why it works.
01
First-Win Principle
One workflow. Proven. Then expand. A system in production by week ten lands inside the business in a way no slide deck does. The first win earns the second engagement and makes the third one routine.
This is not a phased rollout. It is a single proven instance the rest of the program builds on top of. We have watched the alternative play out enough times to refuse it.
What goes wrong without it
A program that buys eighteen months of strategy, launches in three places at once, and exits before any of them measurably moved. The team is left with documentation no one reads and tools no one logs into.
02
Process Champion
Every workflow has an internal owner from day one: deep knowledge of how the work actually runs, authority to change it, accountability for the outcome. The Champion carries the system into the business and keeps it answering to the people running the work.
The Champion is the internal owner the system answers to day-to-day. They co-author the rules the model operates under, sign off on review gates, and decide when an exception is a bug versus a policy choice. Without one, adoption stalls and the system rots quietly.
03
Right Tool for the Job
We work across a range of techniques: deterministic automation, machine learning, LLM copilot, LLM structured automation, autonomous agent. The workflow decides the technique.
A framework that defaults to one technique is a vendor with a product, not a firm doing the work. We choose the tool after we understand the workflow, never before.
Anti-promises
The framework is defined by what it refuses.
If an engagement requires any of the below, the shape is wrong and we say so before scoping it.
We do not deploy generic chatbots.
A horizontal assistant bolted onto a business is the failure mode we built the framework to prevent. The work lives in workflows, not chat windows.
We do not build without a baseline.
If we cannot measure the workflow before we touch it, the engagement does not start. Without a baseline there is nothing to defend later.
We do not write strategy decks without a path to production.
A roadmap that does not name a shipping system in the next ninety days is slideware. We exit those engagements at the scoping call.
We do not automate judgement without an owner.
Every system has a named human accountable for what it produces and authorised to change how it works. No owner, no system.
We do not expand before the first workflow works.
Parallel tracks before the first one is in production is how programs collapse. The First-Win Principle is not negotiable.
We do not default to an LLM.
Half the workflows we look at are better served by a Python script and a scheduled job. The framework picks the tool based on the workflow.
See the framework in the work.
Send it over. We'll tell you whether it's a fit and what a first step looks like.