Do I need to be technical to work this way?

No. The point is that you shouldn't have to be. I learn your work; you don't have to learn mine.

What does "5% of the model" mean?

In the AI-sphere of the internet, you'll read about Frontier models and all the capabilities that are yet to come. You don't need that. It is the studio's first principle: a small, well-chosen slice of a model's capability pointed at the right problem does more for a restaurant or a research team than the flashiest demo.

Is 5% a technical benchmark?

It is a working rule of thumb, not a measurement. The point is proportion: the slice of a model's capability that a real problem needs is small, and the care in aiming it is where the value sits.

How does a project start?

With a phone call. Then a paid discovery: you get a clear map of where AI can improve your workflows, based on your routines and the tools you already use.

5% of the model is enough: the only AI strategy a small team needs

Editorial illustration: one small olive-green slice cut from a large dim circle, aimed at a small storefront and a coffee cup on a counter.

Every AI pitch you have ever sat through was about the frontier. What the newest model can do. What it will be able to do next year. How quickly the world is changing.

None of that is the question that matters for your organization. The question that matters is smaller: How can I improve the work I already do, on an ordinary Tuesday?

The principle

You don't need the frontier of what AI can do. A small, well-chosen slice of a model's capability pointed at the right problem does more for a restaurant or a research team than the flashiest demo.

That is the first principle this studio states publicly, and it is deliberately unglamorous. A lot of AI consulting is people geeking out about the future and selling a tool nobody asked for. The alternative is to talk about your Tuesday mornings and show what a new way of working means for your week.

What 5% looks like in practice

At Oliver's, a Chicago restaurant, the general manager didn't need to understand a data pipeline. He needed to know how busy the week was going to be, early enough to staff it and plan around it. The foot-traffic forecast pulls from the tools the restaurant already used: Toast, OpenTable, a weather feed, a hotel-occupancy tracker. The whole story is in the case studies.
At the Usher III Initiative, a rare-disease research nonprofit, the team didn't need a chatbot. They needed a way to run a slow, specialized search again and again, so a compound that can treat more than one disease reaches more patients. I built that as the outside consultant, and the method is theirs to rerun as the science moves.

Neither engagement started with a model. Both started with how the team already makes decisions.

Finding your slice in an afternoon

You do not need a transformation program to find your 5%. You need an afternoon and some honesty about where the week goes.

Name the problem that keeps eating the week. Not the most interesting problem. The recurring one: the schedule you rebuild by hand, the report that always runs late, the answer that always arrives a week after you needed it.
List what you already run. Point of sale, reservations, a weather tab, a spreadsheet, a donor database. At Oliver's, the forecast did not require new software. It required connecting what the operation already ran on every day.
Ask which decision gets better if one signal arrives earlier. That answer is the slice. A model is useful here only if it moves a decision you make.
If there is no nagging problem, the honest answer is "not yet." A tool looking for a problem is how teams end up maintaining software they resent.

A paid discovery runs this same exercise formally, across every workflow you operate, and hands you the map. The afternoon version tells you whether it is worth booking one.

Why not use the whole model?

Because the frontier is a demo, and a demo does not plan your week. The metrics are defined by the outcomes of the tools and workflows we develop: calmer mornings, easier routines, or value generated via time-back and revenue gained. A frontier demo has no metric attached. A workflow that fits how you already work proves its worth every time it runs.

How this shows up in a project

Paid discovery. A clear map of where AI can improve your workflows, based on your routines and the tools you already use.
Project sprint. A solution to a nagging problem: a defined scope, set against a turnaround timeline, that gets you results before you're asked to evaluate a transformation roadmap.
Retainer. If the tool earns its keep, it keeps running quietly in the background and the next piece gets built.

That sequence lives on the Start a project page, and the engagements it produced are in the case studies.

The flashiest demo of the month will not be the thing that changes your week. A small, well-chosen slice, pointed at the right problem, will. And if you are currently sitting through pitches that promise the frontier instead, run them through the four questions in The chatbot nobody asked for.