How we work.
The engagement runs in four stages. This page covers what each one involves and where you can stop or keep going.
Each stage builds on what the one before it put in place.
The diagnosis
We start from the result you want changed.
The first conversation is free: a thirty-minute call on one process that is costing you something, with a short form sent ahead. Within two business days you have a written assessment of where that process stands and what we would do first. That assessment is how we both decide whether the diagnosis is worth doing.
How the process runs
We sit with the people who run the process and trace the workflow as it runs today, task by task: what is working well, where the time goes, and where the mistakes enter.
What the process runs on
The same diagnosis covers what your company knows: where it is kept, which versions conflict, what has gone stale, and what exists only in someone’s head.
Gartner expects organizations to abandon 60% of AI projects unsupported by AI-ready data through 2026. Gartner, 2025
Where each task belongs
The diagnosis also works out the best owner for every task: a person with better tools, a simpler process, an AI agent, or some combination. Some of what we recommend will have no AI in it at all.
Whatever we find comes back scored on five levels, for the process and for what it runs on.
"You have to ask the right person."
Whether it happens at all depends on who is involved and whether they remember.
"It's written down somewhere."
A practice exists in some form, and it works when a particular person pushes. Most companies are at this level after buying a tool.
"We know which version to use."
There is one way it runs, it is written down, a named person owns it, and it is followed most of the time.
"We can tell when it's failing, and we act on it."
The practice produces evidence about itself, and somebody acts on that evidence before anyone has to get burned by it.
"It gets better because people use it."
It survives the person who built it, and use keeps making it better. Very few companies reach this level, and for most it is not the right target.
Expect the scores to be uneven across the rows. The finding that matters is where the process stalls and which link underneath is holding it back. We also mark the level worth reaching on each row, and it is well below the top of the scale. The diagnosis comes back written down and belongs to your company either way; if we keep going, the first engagement is small and scoped, with the outcome agreed up front.
The memory
Giving your company a memory means fixing seven links.
A search tool works on one of the seven links. The other six stay as they were, which is why a company can buy Copilot and still get wrong answers: it searches the stale and conflicting copies along with the good ones.
Capture
It was never recorded anywhere
The decision was made on a call or in a hallway, and the lesson stayed with whoever learned it.
Triage
Everything was kept, or nothing was
Either everything is on a shared drive nobody trusts, or nothing was kept and the answer depends on who you ask.
Ratify
Versions conflict, and nothing marks the current one
Each holder believes theirs is current, and each of them is right about something.
Hold
It exists where nobody else can get to it
The current one is in a personal folder, an inbox, or a copy somebody made before the change.
Retrieve
The right answer cannot be told apart from the rest
The right answer comes back level with twenty plausible ones.
Apply
The answer arrived and did not get used
It arrived after the decision was made, or in a form nobody could act on.
Maintain
It was right and went stale
Prices moved, a job went badly, somebody left, and nothing was watching.
Where it starts
It starts inside the one workflow we scoped and extends from there as the system proves out.
Why it can be trusted
The questions where one answer has to win get approved by a named person, and everything else keeps its source, so an answer always shows where it came from. What gets figured out along the way goes back in.
How things connect
Every piece of the memory stays connected to the customers, jobs, quotes, and decisions it concerns. A question that touches several of them, such as whether you have done a job like this before and how it priced out, gets one answer that cites each piece. Most of these connections are plain facts already in your material, so this part starts building on day one.
MIT’s 2025 study named the core barrier behind stalled AI the learning gap: tools that do not learn, retain context, or adapt to the workflow. MIT, The GenAI Divide, 2025 The discipline itself is established: ISO 30401 sets the requirements for knowledge management systems, governance and roles included. ISO 30401
The improved workflow
Then we improve the workflow, without breaking what already works.
With the diagnosis in hand and the memory taking shape, the changes go in, and the improvements start to appear in the output.
The right hands for each task
Each task goes to wherever it is handled best: AI agents and software for some tasks, and your people with better processes and tools for others. Agents do tasks, and people hold the jobs those tasks roll up to.
In a Harvard Business School field experiment with 758 BCG consultants, AI improved results on tasks inside its frontier and made them worse on tasks outside it. Dell’Acqua et al., 2023
A do-no-harm bias
The business runs on the workflows it has today, so the risks of putting agents into a workflow get weighed fairly against the benefits before we recommend anything.
The caution is standard practice: NIST’s AI Risk Management Framework makes governed risk and human oversight core to trustworthy AI. NIST AI RMF
Mostly existing tools
Very little gets built from scratch. Most of the time the right tool exists, and the job is setting it up to run on what your company knows.
Output that shows its source
Whatever the tool is, its output shows which decision it rested on, so it is easy to check and people learn to trust it.
The ongoing role
From there we run the system as part of your team.
Someone has to own the system for it to keep working, and that is the role we fill. It is fractional, so there is no added headcount.
Running the system
We keep the agents on their tasks and the memory true as the business changes: conflicts get settled, gaps get filled, and anything that needs a fresh look goes to the person who owns that area.
The problems along the way
When someone has to pull what matters out of a thousand PDFs, or needs an analysis nobody has time to build, that comes to us. A request that keeps showing up becomes a permanent part of the system.
Watching the technology
We follow the technology with no loyalty to any tool or platform, and our business model depends on the relationship lasting, so the only thing we have an interest in recommending is what works for your business.
The system runs in your company's own accounts. What we set up belongs to your company, the third-party tools in it stay on the licenses you already hold, and if you decide to stop, leaving takes thirty days.
The engagement
The engagement starts narrow, and it grows as the results earn it.
The first engagement is a specific piece of work with an agreed scope and outcome. You can begin with the diagnosis alone, and whatever has been built belongs to your company if you stop there.
One workflow at a time
The engagement takes one workflow at a time: the one that matters most, and the memory underneath it. Once that is running, the same memory can serve the next part of the business, so each extension is faster to put in place.
How it is priced
Paid work starts as a scoped project at a fixed fee, agreed before anything begins. The ongoing role is a monthly retainer sized to what it covers.
How long it takes
The diagnosis takes a few weeks. The first workflow is designed to be live within a quarter of starting.