My last six posts made an argument: AI fails on conditions, not models; capabilities are the unit; AI has three jobs and one it can’t take; transform by slices, never all at once. The most common response I get is some version of: okay, I buy it. What do we actually do on Monday?
Fair question. Here’s what it looks like on the ground: the way we run it, and honestly, the way anyone should run it, including your own internal team. Steal the process. The stages matter more than who executes them, and each one is built to earn the trust the next one spends.
Stage One: Understand, Before Anyone Signs Anything
It starts with conversations, not contracts. The lay of the land: who the players are. Who owns what, who’s carrying the board pressure, who becomes the champion, who might feel threatened. What the goals and objectives actually are, in the sponsor’s own words. The known constraints: regulatory, technical, political, budgetary. Every enterprise has walls that don’t move, and pretending otherwise is how proposals get written that deals die on. And most important: what success looks like, in numbers somebody is willing to write down.
Underneath those questions we’re really testing two things from earlier in the series. Is there a real business problem with money attached, not a use case hunting for a justification? And does the fourth condition exist: is intent governed, is there a sponsor who can say what winning means and kill what isn’t working? If either answer is no, the honest move is to say so and stop. Trust required so far: none. Cost: conversations.
Stage Two: Align on One Room, One Frame
Next, a working session with the leadership and the people who’ll live with what follows. We walk the argument you’ve just read: why pilots die, the four conditions, capabilities versus applications, what AI is actually for, why slices. Not as a pitch, but as a working frame the room applies to their organization in real time.
Remember the warning about automating the artifacts, that the point of a story was never the story but the shared understanding? The workshop is that principle applied to strategy. The deliverable isn’t the deck; it’s collective cognition: a leadership team that can make consistent decisions about this work when nobody from the outside is in the room. The test of a good workshop is the same one I’ve seen convert career transformation leaders: everyone can locate their own work inside the frame. The output is concrete: shared language, and a chosen area of the business where the pain, the value, and the appetite line up.
Stage Three: Hypothesize Where We Think the Seams Are
In that chosen area, we form an extraction hypothesis: which capabilities we believe are trapped in which containers, where the value sits behind which seam. Notice the word. It’s a hypothesis: written down, specific, falsifiable. That’s what directed R&D means, and the next stage exists to test it. The code gets a vote.
Stage Four: Ingest Two to Three Weeks of Truth
Then the navigator goes to work. We ingest the code, and the data around it, and map what’s actually there: domains, subdomains, bounded contexts, the duplicates, the dependency knots. Machines do the archaeology at about 80% accuracy; experienced humans judge the rest. Months of discovery compressed into weeks.
What comes out is the value backlog: the inventory of value propositions worth solving. Each is scored on two axes, how much the slice is worth and how cleanly it cuts, then sequenced economically by weighing value and cost of delay against effort. We are hunting for a specific shape: high-value use cases that can be extracted in ninety-day increments. And it’s a living backlog: the first study seeds it, every delivered slice refreshes it, and the ranking moves as the work teaches us. Alongside it, an end-state picture: if every piece could move freely, this goes to the ERP you already own, this retires, this gets built custom because it’s how we win.
Three honest notes. The depth of the map flexes with the need: some problems can be pulled apart and monetized without a full architectural view, some estates demand the deep map first. The stages don’t change, the depth does. Sometimes the study says not here or not yet, and a few weeks of truth that prevents a multi-year mistake is one of the best purchases an enterprise can make. And notice where we still are: nobody has touched production. Everything to this point sits deliberately below the trust barrier.
Stage Five: Extract the First Ninety Days
Now we cut the seam: the top of the value backlog. Inside the slice, the four conditions get created for real: the capability encapsulated, the data validated at its source, one team that owns it, your people on that team from day one, and intent governed by the number we wrote down in stage one. The navigator maps, the junior engineers build under test-first supervision, humans judge. The slice goes to production. And then the only measurement that matters: the value we promised is the value we produced, in numbers your CFO accepts.
Then the loop turns. Slice two starts easier, because slice one left the conditions behind. Somewhere around slice three, the cadence should be yours, not ours. The goal was never a permanent engagement. The backlog has a bottom: we go only as far as there’s economic value to be achieved, and when the marginal slice stops paying, the work is done. What stays when we leave is a capability that was previously not possible for you.
One structural promise ties all five stages together: every stage is a gate. Stop after any of them and you keep everything of value: the map, the frame, the backlog, the proof. If a partner won’t structure the work that way, remember the tell from the last post: anyone proposing to transform everything before proving anything is asking for trust they haven’t earned.
The whole series was the whole story. This is the first slice. Buy the slice.
This is Part 7 of a seven-part series. Start wirh Part 1 here.
