The method
How a Workflow Build Session compounds your firm's capability.
The deep dive behind the Workflow Build Session. The goal is that the high-value work that used to take a week now takes hours, because we replace generic AI tools with your encoded expertise. This shift allows the AI to handle the labor-heavy first draft: the deep research, the model building, the writing in your firm's voice. Your people spend their time where their judgment matters most.
The recovered capacity goes to deeper analysis per engagement, more clients served, or prototypes your clients can try instead of imagining. Every draft is scored against your firm's prior best work, raising the ceiling and the floor at once. The result is Level 3 codified capability the firm owns: a documented workflow, an AI workspace, SKILLS.md files, and an evaluation rubric tuned to your standard. This is what practical applied AI looks like.
Destination: Level 3 codified capability.
What encoding means
From generic AI to work your team can use
Encoding your expertise is creating reusable context files, skills and references for your AI agent (e.g. a Claude Project) that allow it to express your expertise: which workflow it is running, what your firm brings to it, what good looks like in your hands, and how to check its own draft against that standard. All of this lives inside your firm's own AI account; nothing trains a public model. The act of encoding your expertise for the AI has the benefit of making it readily distributable across your team and drives alignment on your firm's standards.
The four steps
Build Your AI-Enhanced Workflow
We document how you do the work today, and build text files so the AI can understand and execute it under your supervision. This is your own AI-agent, applied in practice.
- 1
Pick the workflow to encode
What it is
The recurring work that is key to the value your firm creates. The signal that something is the right workflow: your team does it on most engagements, it carries your firm's expertise, and it requires research, analysis, and writing.
What gets done
We review the workflow your team runs and break it into component parts, the inputs, the outputs, the senior review process, typical time spent and what the deliverable looks like. Even for bespoke workflows, there are components that are repeatable and can be encoded.
What you walk away with
Workflow Insights: a documented workflow the firm owns. Even if you never built another thing in AI, this artifact changes how new hires learn the work and is the enabler for everything that follows.
- 2
Encode your firm's standing context
What it is
Generic AI gives generic answers. To make this workflow uniquely yours, we document your firm, your people, and the workflow's specific context by assembling content from your website, prior documents, and direct AI interviews with the team. This provides the foundation for the AI to work from your firm's unique perspective. Then each new client engagement becomes its own AI workspace (e.g. a Claude Project) that inherits your firm's standing context so it can be reused on every engagement.
What gets done
We capture six things and put them in the workspace:
Identity (per-person: your role, how you communicate, what you prefer), Firm context (the shared knowledge: team, products, terminology, tools, key decisions), Work process (the workflow from Step 1, in the form AI can run), Client-engagement goal (what this engagement needs to achieve), Tools (deep research, specific databases, valuation models, template structures, comp libraries), and The workspace itself (the container that holds all of the above, ready for a new engagement to inherit on day one).
What you walk away with
A reusable knowledge base of context the whole firm can use, so every new engagement inherits your firm's expertise instantaneously. New hires inherit it in their first week instead of having to rebuild it manually. This encoding of your firm's knowledge and expertise is what enables AI to work like you were doing the work yourself.
- 3
Encode your definition of good
What it is
Pull three to five of your firm's best human-only deliverables from past engagements, read them carefully, and document what makes them good, both in structure and substance.
What gets done
We sit with the deliverables and find the commonalities. We turn the attributes your firm believes are good into a scoring rubric AI can apply to its own drafts. The rubric is yours. Your dimensions, your weights, your standard.
What you walk away with
SKILLS.md files versioned in Git, including the scoring rubric. Every AI draft is scored against your firm's standard before it reaches the senior reviewer. The rubric improves as your firm closes new exceptional work. This is the quality control Level 3 names.
Two examples of what "good" looks like, in two very different firms:
M&A boutique
- confidence in the valuation range
- sharpness of the value-driver insight
- ability to surface the buyers with the highest willingness to pay
- jargon discipline
- internal consistency between the financial summary and the BOV
Litigation firm
- citation accuracy
- persuasive sequencing
- anticipation of the opposing argument
- plain-language enforcement on terms of art
- 4
Run it on real work and improve
What it is
Setup isn't the finish line. The finish line is AI producing drafts your team takes to a real client engagement.
What gets done
We run the encoded workflow on a current engagement and compare the AI output against the rubric from Step 3. Then we tune. The context might be missing a reference document. The workflow might need an extra stage. The rubric might be over-weighting the wrong dimension. The first run usually surfaces three or four things to adjust. By the third run, it's producing good results.
What you walk away with
A trial run completed. A workflow you trust enough to use, because you watched it produce a draft against your own definition of good. The compounding effect starts here: you've created reusable context files, learned how to use AI tools to execute complex workflows, and learned how to use AI to evaluate output against SKILLS.md files that can be shared by everyone. Now you have a process that can apply to any workflow and start to accelerate the entire company.
FAQ
Questions that come up
How long does the Build Session take?
A half-day in your conference room. The four steps above are what we walk through together. Pre-session research happens beforehand so the time on site goes to the work, not the setup.
How do you handle confidentiality and sensitive data?
All of our sessions are 100% hands-on keyboard work, so any sensitive work runs inside your firm's own Claude Team or Enterprise account, with no model training on submitted data. If your firm already has an AI data policy, the workspace inherits it. If not, putting one in place is part of the engagement at no extra cost.
What if my team is already using AI without an evaluation system?
This is the most common state for companies using AI today, and would place you at Level 2. Your power users are producing real work and the rest of the firm isn't benefiting. The power users aren't doing anything wrong, it's that the firm has no way to spread and version the progress. Workflow encoding is the process of compounding your AI gains to firm level. We create context every engagement inherits, and an evaluation rubric that scores every draft against your firm's standard.
Can I do this with my existing AI tools?
Yes. The method is tool-agnostic on purpose. Most engagements run on whatever your firm already has, CoPilot, Claude (Team or Enterprise), or ChatGPT (Team or Business). The Build Session encodes your context and evaluation rubric as artifacts (Projects, SKILLS.md files, prompts) that live inside your existing account, so nothing new gets purchased and nothing your IT team hasn't already approved gets introduced. If your firm doesn't have firm-managed AI accounts yet, the AI Executive Quick Start (2hrs) may be a good first step.
What happens after the Build Session?
You leave with a documented workflow, a reusable AI workspace, SKILLS.md files including the scoring rubric, and a trial run completed on real work. The compounding starts immediately. Every new client engagement on the same workflow is faster because the AI skills you built carry over. The team's skill with AI grows with daily use, and their creativity is unblocked. New hires inherit firm-grade context in week one instead of learning it from scratch. If you want to extend the same method to three to five more workflows and train an internal owner, that is the Workflow Propagation Engagement, one week.
Keep reading
More context, if you want it
Step back to see the broader practice, or look ahead at where a Build Session leads.