The homepage keeps a list of what most AI pitches sell you. It is short: a chatbot nobody asked for, a new dashboard to learn, a six-month transformation program, and speeches about the post-human future.
You have heard the speeches. This piece is about the moment after them, when someone slides a proposal across the table and waits for a signature.
You didn't get into this work to run software. Whatever you run, a mission, a kitchen, a craft, a newsroom, the software was supposed to stay in the background. A pitch that leads with the software has the subject backwards.
The test every pitch has to pass
Does the pitch start from how you already make decisions, or from what the tool can do?
That is the whole test. Everything else is decoration.
It works because it is checkable in one conversation. A pitch built around your decisions will be full of questions about your week: who plans what, when the signals arrive, where the work piles up. A pitch built around the tool will be full of adjectives about the tool.
Four questions to run any vendor through
1. Does it start from my routines, or from a blank slate?
A serious engagement begins by learning how you already think and work. If nobody has asked about your routines, nobody can aim a tool at them. Ask the vendor to describe your week back to you before they describe their product.
2. Where does the data already live?
The strongest AI work connects systems you already run instead of replacing them. At Oliver's, the Chicago restaurant in our case studies, the foot-traffic forecast came out of Toast, OpenTable, a weather feed, and a hotel-occupancy tracker. There was no new dashboard to learn: the GM plans staffing and events inside the routine he already had. If a pitch requires ripping out what you run every day, the price on the last slide is not the real price.
3. What decision gets better, and who feels it?
Outcomes, not deliverables. Calmer mornings, easier routines, decisions made earlier, more time for the work you're good at. If the answer is "efficiency" and nobody can name the person whose week gets lighter, nothing has been promised yet.
4. What do I walk away with if I stop?
A tool that only works inside someone else's retainer is a lock. Ask what stays yours either way: a map of your tools and workflows, insights about how to build new systems for yourself, and the files to rebuild later should survive any ending. If nothing leaves the vendor's hands, you are not buying a tool; you are renting an exit.
What a good pitch sounds like instead
Quiet, mostly. A defined first project against one nagging problem. A scope you can read in one sitting. A defined end date. Results you can feel before anyone mentions a roadmap.
That shape is not modesty. A contained, paid first build is the cheapest way for both sides to find out whether the fit is real, and it is the only kind of first project this studio offers.
The starting point is never the model, either. It is the same test the demo failed: start from how you already make decisions, then aim a small, well-chosen slice of the model at the problem eating your week. That argument, and how to find your slice in an afternoon, is in 5% of the model is enough.
Run the four questions on the next pitch you hear. If the answers hold, the tool might be worth your Tuesday. If they wobble, a paid discovery maps where AI would fit first.