Most AI projects don’t fail. They just never end.

You run a pilot. It works well enough to be exciting. Then it sits there, half-wired into your week, useful on Tuesdays, forgotten by Thursday, never quite trusted enough to replace the old way. A forever-pilot. The demo that never graduated.

I’ve watched this happen again and again, and it almost never comes down to the model. The capability was fine. What was missing was a way to move from “this works” to “this is how I do it now.”

SHAPE is the method I use to get one project across that line. Five moves: Situation, Hypothesis, Action, Process, Evidence. I’ll walk each one with a single example, a freelance copywriter who wants to turn client call recordings into first-draft blog posts and keeps not quite finishing the job.

Situation: where you honestly are

Before you change anything, describe where you actually are. Not where you’d like to be.

Our copywriter’s honest situation: she records every client call, means to repurpose them, and has a folder of forty untouched transcripts. She’s tried three AI tools. Each produced a passable draft once, then she drifted back to writing from scratch because the drafts needed as much editing as they saved.

That’s the real starting point. Not “I need an AI writing tool” but “I have a graveyard of good intentions and a trust problem.” Naming it that plainly tells you the constraint isn’t capability. It’s a workflow that never got finished.

Hypothesis: the bet you’re testing

A pilot with no hypothesis runs forever, because there’s no line it can cross. “See if AI helps” can be debated for months. A hypothesis gets settled in a week.

So make it a number. Hers: a single prompt, fed a call transcript, will produce a blog draft she can publish with under twenty minutes of editing, at least four times out of five. Baseline, target, and a threshold that decides it.

The point is that you now have something that can be proven wrong, not precision for its own sake. A bet you can lose is a bet you can win. A vague hope is neither.

Action: the smallest real thing to do

The instinct here is to build the system. Resist it. Do the smallest real version instead.

Not a tool. Not an integration. She takes five actual transcripts, real ones from the graveyard, and runs them through one prompt she’s willing to refine. She times the editing. She checks the result against the hypothesis.

Five real cases beat a beautiful pipeline every time, because they tell you the truth fast. By Friday she knows: three of five were clean, two needed heavy work, and both weak ones were rambling discovery calls. That isn’t failure. That’s the edge of where it works, and now she can see it.

Process: making it repeatable

Something that works once is a trick. Something that works on a tired Tuesday is a system.

So she draws the boundary the pilot found: this prompt is for structured client calls, not open-ended discovery ones. She saves it where she’ll actually reach for it. She writes the three-line checklist she runs each draft through. She decides what happens when it produces a dud, which is that she edits by hand, no drama.

That’s the whole of Process. Not scaling for the sake of it. Just the same input, the same steps, the same result, without her having to remember how she did it last time. The forever-pilot ends right here, the moment the new way becomes the default and the old way becomes the fallback.

Evidence: proving it worked

Two weeks in, she looks at the numbers she promised to look at. Twelve drafts, ten published under the twenty-minute line. Editing time down from ninety minutes to eighteen. The two misses were both discovery calls, the boundary she’d already drawn.

“Feels faster” would never have survived a bad week. The evidence does. And it earns you two more things: it points at the next honest situation, those discovery calls, and it gives you a real number to price on instead of a hunch.

Pick one and run it

None of this is complicated, and that’s the point. The people who stall aren’t short on ideas or tools. They’re short on a way to march one idea from “interesting” to “this is how I work now.”

Pick the project that’s been idling in your own graveyard. Run it through the five moves. You’ll either ship it or learn exactly why it won’t ship, and both beat another forever-pilot.

If you’d like a second pair of eyes on which project to pick, get in touch.