Stage 3 of 5 · Decide

Determine what is worth doing.

What should we do, and can we defend it?

By this point the work has produced two things: a record of what is true, and a comparison against what the market is doing. Neither of them tells you what to do. That takes judgment about your business, your risk and your timing, and judgment is the part we have deliberately not tried to automate.

The recommendation should be where the work leads, not where someone decided they wanted to go.

This is the sentence the whole model is built to make true. It is easy to say and hard to arrange, because it requires the evidence to exist before the recommendation does, and it requires whoever makes the recommendation to have nothing riding on which way it goes.

NarrowGateX holds relationships across the technology market, and in some transactions a provider pays compensation. That is disclosed to the client. What it never does is determine the recommendation: the evidence decides whether action is warranted, and only then does anything go to market.

Who does what

Three roles, and only one of them is software.

The platform

Preserves. It holds the evidence, the options, the tradeoffs and the reasoning so the decision survives the person who made it and can be examined a year later by someone who was not there.

Your advisor

Interprets and recommends. Which differences matter, what they are worth, what the risk actually is, and what they think you should do. They put their name on it and stay accountable for it.

You

Decide. It is your environment, your budget and your risk. The recommendation is an input to that, not a substitute for it.

Why there is no worker here

Four of the five stages have an AI worker. This one does not.

Atlas reads documents. Wimbl finds gaps. Nadia compares against the market. Vera runs supplier events. Porter watches dates. Each has a narrow job, a visible output and a stated limit, and each does work that is genuinely repeatable.

A recommendation is not repeatable work. It weighs one organization's risk appetite against its timing against its politics against what it can actually absorb this quarter, and it has to be defended afterwards by someone who can be asked why. There is nothing here to automate, and building something that produced recommendations anyway would make the output look more certain than it is.

So the blank in this stage is not a roadmap item. It is the design.

Defensible decisions

The decision is a record, not a meeting.

Most technology decisions survive as a slide and somebody's recollection. When the renewal comes back around, or the person who made the call has moved on, or a board asks why the estate looks like this, the reasoning is gone. The platform keeps it.

The alternatives

Including the ones that were not chosen, and what they would have cost. A recommendation with no visible alternatives is an assertion.

The evidence

What the Baseline established and what the Benchmark compared, with the grade on each fact carried through.

The tradeoffs

What is being given up. Every real decision gives something up, and the ones that appear not to are usually the ones nobody examined.

The reasoning

Why this, and why now. Written down at the time, not reconstructed later from memory.

Financial and operational impact

What changes in cost, and what changes in how the business runs. The second one is the half that gets skipped.

That record is also what makes the next cycle cheaper. A decision you can reconstruct is a decision you do not have to make again from nothing. What makes it reconstructable is that every fact underneath it carriesa grade rather than an assertion, and that theworkers that assembled it state what they will not do.

The cycle

Govern keeps the Baseline current, and the cycle starts again.

← Benchmark Execute →

Bring the decision you are already facing.

A renewal, a rising cost, a proposal on the table. We will show you what the evidence would have to say for the answer to be clear.

Start a Technology Baseline See the whole model