Skip to content
Built by SamirTalk to Samir →

How I run a project

The Built Method

Most technology projects fail for the same reason: nobody agreed in advance on what success would look like, so everything gets declared a success.

Five steps. The fourth one is the one that costs me money.

That's the step where technology has to earn its place. A system survives it only if the number it was built to move actually moved.

The Built Method, five stepsA vertical flow diagram of five steps: Find, Quantify, Build, Validate, and Scale. Validate is drawn larger than the other steps to show it is the differentiator. Two paths lead out of Validate: one continues down to Scale, the other branches right to a stop, showing that a Validate step can end the engagement instead of continuing it.01Find02Quantify03Build04Validateone number, agreed in writing,before build startsProceeds05ScaleSTOPStops —in writing

1. Find

The bottleneck you describe is rarely the bottleneck you have. I interview your people, watch the actual workflow, and review the systems before recommending anything.

2. Quantify

Every problem gets a number: dollars, hours, revenue at risk, compliance exposure. If a problem can't be quantified, it doesn't move forward. That alone eliminates most "AI opportunities."

3. Build

The smallest thing that could plausibly move that number. Weeks, not quarters. Real system, real data, real users — not a demo.

The differentiator

4. Validate

Before any build work starts, we agree in writing on a single number that defines success, how it will be measured, and over what period.

When the period ends, I write down whether it hit. If it didn't, my recommendation is to stop — in writing.

That's a successful engagement. You paid to find out, you found out, and you didn't spend the next eighteen months funding something that doesn't work.

5. Scale

Only systems that proved out earn more investment. Then we roll it across locations, harden it, and hand it over — you own the code, the accounts, and the documentation. You're not renting it back from me.

Evidence

Why you can believe the fourth step

It stayed suppressed.

I run this on my own work.

I built a predictive model, ran it for months against real outcomes, and it looked strong. Then I found a defect in how outcomes were paired to predictions. I fixed it, re-scored every prediction honestly, and the result inverted: the model had been wrong the entire time I was reading a record that said otherwise. I turned it off and left it off.

Separately, I ran a second model in shadow mode across 125 real outcomes before letting it touch anything. It came in at 52.1%, below the threshold I had written down before I knew the answer. It stayed suppressed.

Nobody was paying me to be honest about either one. That's the point.

The validation chartA line chart with no numeric values. A line trends confidently upward across most of the run, well above a baseline. At a marked point labeled "Defect found," the trajectory is reinterpreted: a corrected line drops from that same point, crosses below the baseline, and ends lower than where it started, marked "Turned off." The chart illustrates that the honest, corrected result inverted the original, favorable-looking one.baselineDEFECT FOUNDTURNED OFFMeasured performancePredictions scored