AI use cases: from idea to go/no-go
Run the whole screen, and let the honest answer be no when it is no. That is the entire discipline of this track compressed into one sentence. Every page so far has given you one instrument: gates that test whether an idea is real, a definition that separates assistants from agents, a ladder that prices autonomy, an ordering that keeps you from building what you could buy, and a formula that keeps the value claim honest. None of them is the point on its own. The point is what they produce when you run them in sequence: a decision you can defend, written down, reached before real money moved.
Most organisations skip the sequence and go straight from enthusiasm to procurement. The idea sounds good in the meeting, a pilot gets funded, and the questions this track asks get answered later, in production, at the most expensive possible moment. The screen exists to move those answers to the cheapest moment instead: before the build, while a “no” still costs days rather than quarters.
The sequence
Section titled “The sequence”Each step feeds the next. Answer them out of order and you end up justifying a build with the autonomy tier, or picking a vendor before you know whether the use case is real. In order, each step either kills the idea cheaply or hands the next step something concrete to work with.
-
Run the three gates. Name the grounding (which system holds the truth), the number (what moves if this works, readable from a system today), and the owner (who decides). An idea that fails one gate in principle stops here, and stopping here is the cheapest exit in the whole sequence. See the three gates.
-
Name what you’re building. Instructions, knowledge, and at least one action make an agent. Knowledge but no action makes an assistant, a smaller and cheaper build. Neither knowledge nor an action makes a demo, and a demo goes back to step one, not forward. See assistant or agent?.
-
Pick the autonomy tier. Choose the lowest tier that captures the value, and plan to go live one tier below it. Every rung up the ladder is a real governance cost line, so the tier belongs in the decision, not in the fine print after it. See autonomy tiers.
-
Ask buy, extend, or build, in that order. A finished product that fits ends the evaluation with a purchase order, and that is a successful outcome. Building is the exception you defend with total cost of ownership, not the reflex. See buy, extend, or build.
-
Freeze the measurement. Write down the baseline, the formula, and where the number will be read from, before anything is built. A value claim without a pre-build baseline is testimony, not evidence, and by then it is too late to collect any. See measuring the value honestly.
-
Decide, in writing. Go, No-Go, or Park with a named condition. Whatever the verdict, it goes into a document the owner from step one signs, because a decision nobody owns is not a decision, it is a mood.
What a decision-ready go/no-go contains
Section titled “What a decision-ready go/no-go contains”The output of the sequence is a short document, not a slide of enthusiasm. If you cannot fill in every line, you have not finished the screen; you have paused it.
| Line | What it says |
|---|---|
| Scope, in and out | What the system does, and explicitly what it does not do. The out-list prevents the quiet scope creep that sinks pilots. |
| The number | The metric that moves, its current baseline, where it is read from, and the formula that will compute the verified value. Your numbers, from your systems. |
| Buy, extend, or build | The recommendation and the total-cost reasoning behind it, including the products evaluated and why they did or did not fit. |
| Target autonomy tier | The tier, the tier below it where the system goes live, and the governance each tier requires: approval steps, audit log, rollback, monitoring. |
| Owner and decision | Who owns the process, and the verdict: Go, No-Go, or Park with the condition that would reopen it. |
A document with these five lines can be challenged, budgeted, and revisited in a year. A pilot approved on a demo can only be regretted.
No-Go is a result, not a failure
Section titled “No-Go is a result, not a failure”Here is the part most teams get backwards. A screen that only ever says Go is not a screen; it is a formality on the way to a decision already made. The sequence earns its keep precisely when the answer is no: no grounding exists anywhere, the value genuinely cannot be measured, or a finished product already does the job for a subscription fee. Each of those is a real answer, and reaching it after a week of gate-work instead of after two quarters of building is the difference between a decision made cheaply and a decision made expensively. The same idea can come back when the condition changes, and the Park verdict with a named condition is how it does so on evidence instead of on renewed enthusiasm.
Everything on this page you can run yourself, on your own case, with your own numbers, and you should. Running it on a client’s specific case, with their systems and their baseline, is the work machtsinn does.