Case studies

Research · worked examples and a sample template

Members are encouraged to carry one case study through the year, starting whenever they join. These are the worked examples so far.

California water rate data

Leads: Patrick Atwater and Maxwell Titsworth · with the California Data Collaborative · ongoing, now in implementation

California's water rates are public but scattered across hundreds of utility sites and PDFs. Machines can now read any rate sheet; finding the current one is the hard part. So the work focused on discoverability (a registry of each utility's rate page and an llms.txt pattern) rather than a new data standard.

Strategy report · Seven precedents · Five cases in data coordination

Construction procurement protocols

Lead: Andre Comeau

Construction procurement redescribed as software development: a Design, a Tender that acts as the API, and construction as the compile. The existing protocols are described as tensions to be managed by the parties involved, and a small app, a fair-and-reasonable price reviewer, tests whether a bot-first procurement protocol could be built with tools available today.

Age of AI, the age of protocols, Andre’s essay on the tool-building side

Protocol Institute brand kit

Marketing · small case

A written brand spec that agents build pages from. It kept producing the same bugs until the kit gained shared components and automated checks. Brand kit

Working smarter with AI

Lead: Robert Peake · with Timber Stinson-Schroff · Knowledge Work · in beta

A shared collection of 228 observations on working with AI, 155 practices and 73 hazards, gathered at the Protocol Institute Symposium 2026 from participants and speakers. AI proposed matches between hazards and the practices that answer them, and people checked each one. Visitors rate the cards and build their own protocols from them. One early pattern echoes the hard core: separate hard walls, which block an action, from soft walls, which flag it for review.

A companion survey asks how people actually co-work with AI: how much effort it takes, how far they trust it, how much control they keep, and where relying on it goes wrong.

Working smarter with AI (PI26 collection) · AI co-working effectiveness survey

Your case study

Pick an organization you can see into: your own, a client's, or one whose records are public. Pick a data operation in it that AI is changing, or should change, and carry it for about a year from whenever you join, sharing progress in a session's show-and-tell.

• Month 1Choose. One paragraph on the organization and where the work gets stuck.
• Months 2–3Brief. One page on the context, the bottleneck, and how the work is done today.
• Months 4–6Intervene. Decide what becomes code, what is left to agents, and what stays with people, and what you will measure.
• Months 7–11Evidence. Run it, measure it, and note what surprised you.
• Month 12Publish. An open write-up, which feeds into the group's thesis.

Sample template

If a structure helps, the worked examples answer these questions. Find your own way if another shape fits your case better.

1. Context: the organization and what the data is for.
2. Bottleneck: where the work is actually stuck.
3. Before: how data was found, checked, and trusted.
4. Intervention: what became code, what went to agents, what stayed with people.
5. Evidence: what you measured, and where the model was wrong.
6. What transfers: what another domain could reuse.

To start, copy the sample template into case-studies/your-name/, or post your case in #protocols-for-business. You can keep a client anonymous.