Ghost does not start with the data

People ask what I mean when I say we analyze a business. They picture a model, a dashboard, a deck full of charts. That is the output sometimes. It is not the work. The work is understanding what is actually going on, and the most reliable way I have ever found to do that is older than any of the tools. Get the right people together, surface what they each know, and align them on what matters most. Do that well and the analysis falls out of it. Skip it and no amount of compute saves you.

This is not a theory I picked up last year. It is how I was trained, and it is most of what I have done for my whole career.

Where I learned it

I trained with the Luma Institute, in human-centered design. The idea underneath it is simple. You solve a problem by understanding the people inside it, not by guessing from a conference room. At Forest Giant, the agency I ran, I made it a requirement that everyone trained in it. Not just the designers. Everyone, down to the office assistant. If you worked there, you knew how to run a session, frame a problem, and pull insight out of a group instead of out of your own head.

I was one of the first people from Louisville to go through a startup accelerator, The Brandery in Cincinnati. That world ran on the Lean Startup, Eric Ries, build measure learn. You do not write the hundred page plan. You find the smallest real test, run it, and let what comes back tell you where to go next.

And I took Elon Musk’s vector idea to heart. The outcome of any company is the sum of the people in it, treated as vectors. If everyone is talented and committed but pointed in different directions, the sum is small. Align the directions and the same people produce a different company. Most businesses are not short on horsepower. They are short on alignment.

Put those three together and you get how I actually work. Start small. Get in front of people. Align the vectors. Then move with priority.

The method, in a room

When I sit down with a business, I am running structured exercises, not a freeform chat. The point of each one is to get what is in people’s heads out onto the wall where everyone can see it and argue with it.

A creative matrix to generate options fast, across the people you serve and the ways you might serve them, so nobody is staring at a blank page. An importance versus difficulty matrix to sort all of it into what is worth doing now and what is noise. A bull’s-eye to force the hard call: what is critical, what is merely important, what is peripheral, because a strategy that will not say no to anything is not a strategy. A round robin so the quietest person in the room contributes as much as the loudest, because the insight you need is usually not held by the person most comfortable talking.

None of this is decoration. Each exercise is a way to take a fuzzy, contested picture of a business and turn it into a shared, prioritized one. By the end, the room agrees on what matters, in what order, and why. That agreement is the asset. It is also genuinely fun, and that matters more than people admit, because people defend what they helped build.

What changed, and what did not

The method did not change. The speed did.

It used to be that the workshop produced a wall of stickies and a synthesis deck a couple of weeks later, and half the energy leaked out in the gap. Now the alignment we build in a room becomes the input to a system that runs. The understanding goes straight into the brain, the place where everything we learn about a business lives and connects. The priorities we set become the goals the work gets graded against. And the loop does the running.

That last part is the piece most people are missing, so let me be plain about it. The room is the front end. The back end is a fleet of agents working against what the room decided. We point them at the prioritized goal, they do the finding and building and checking, and an eval grades every output against the standard we aligned on, not a generic one. The human alignment is what makes the automation aim correctly. Garbage alignment in, expensive slop out, no matter how good the model is. Good alignment in, and the loop compounds in the direction the business actually wanted to go.

So the vectors matter twice now. Once for the people, the way Musk meant it. And once for the machine, because an agent is just another vector, and an unaligned one moves fast in the wrong direction.

The room is where the record starts

Aiming the agents is the easy half of what the room is for. Here is the half I have been underselling, and it is the reason this piece exists.

A system that learns needs something to learn from on day one. Stand one up against an empty record and it has no priors, no sense of what good looks like here, and nothing to grade anything against. It will be fluent, fast, and confidently generic, which is the worst possible combination because generic work that arrives quickly is harder to argue with than generic work that arrives slowly.

Every company already knows what predicts a win. It is sitting in a handful of people, in their notes, in habits nobody has examined in years, and almost none of it has ever been written down. Your best salesperson can tell you inside two minutes on a call whether this one is real. Ask them how and you will get a shrug and a story. That shrug is the most valuable unextracted asset in the building.

The room is how you get it out. Not the data, which is the part everyone reaches for first because it is easy to export. The judgment. What your best people notice, what makes them walk away, what usually happens right before somebody buys, and which of the obvious signals they have learned to ignore.

That is the baseline. Everything the system learns afterward is measured as movement away from it.

Two rules make the extraction work, and both are counterintuitive. Do not introduce your vocabulary until they have already described the thing in theirs, because the moment you hand somebody your word for it they start performing your framework back at you instead of telling you what they actually do. And do not ask people for signals. Ask what they notice, what catches their attention, what tends to happen right before a deal turns. Same information, and you get the real version.

So let me say the strong version plainly. Standing up a learning system without doing this first is not merely harder. It is futile. You will have built an expensive machine for compounding nothing, and it will take you a couple of quarters to find out, because the output will look fine the whole time.

This is why the room is not the preamble to the work. It is the front end of the work, and everything downstream inherits whatever quality it had. The record that compounds has to start somewhere, and it starts with people in a room telling the truth about how they actually decide.

Why this is the durable part

Everyone is about to have the same agents. The models will keep getting better and cheaper and they will not be anyone’s edge for long. What does not commoditize is the ability to walk into a business, get the right people to tell you the truth, and align them on what matters before a single agent runs. That is judgment, and it comes from having done it across a lot of businesses, not from a subscription.

This is a big part of what we offer, and it is the oldest thing we do. The workshops, the prioritization, the alignment. We just stopped losing the two weeks afterward, and we wired the back end so the alignment actually executes instead of sitting in a deck.

Start small. Get the right people in the room, get the truth out of them, and write down what good looks like here before anything runs. Then let the loop go, because now it knows where it is going and what it is being measured against.