The Strategy Practice
Judgment for the Decisions You Can't Take Back.
The tools changed. The judgment is still the job. Bespoke strategic counsel for leaders whose next call is hard to reverse, whether or not AI is the variable.
What This Is
Judgment, Shaped to Your Company.
General advice is now at our fingertips and essentially free. The harder part is telling which of it applies to you, catching the one distinction that changes the answer, and naming a problem while it is still cheap to fix. That means discernment mapped against the specifics: your own field, your workflows, your risk tolerance, your people, and the decision actually in front of you this quarter.
That takes three forms here. Building the system a company runs its AI on, so the tools are pointed at the right problems and someone owns what they are allowed to do. Keeping a leader's judgment current once that system exists, on a cadence, so the answer does not quietly go stale. And the older work this practice is named for, where AI is beside the point: positioning, messaging, and the partnership calls that decide what a company becomes.
What holds across all three is how the work ends. You are left with something built rather than advice you have to translate, and what counts as working is written down before the work starts, in your language, tested against your own material, and judged by your own people.
The Work
Engagements, Named by What They Leave Behind.
Strategic Council Build
For a company that already has AI in use and no system around it. One source of truth your people actually maintain. Guardrails mapped to the risk profile of specific workflows rather than a policy that could belong to anyone. Tailored AI teams placed at the decision points where they change the outcome. And the training that makes it yours to run without me.
Those AI teams are built on the Davis Codex, a method of Trevor's own: how each one is scoped, what it is permitted to touch, how it is tested against real work before it is trusted with real work, and how it stays current instead of going stale in month four.
Before anything is trusted, the tool is run against work your people already know cold. They mark what it missed. That becomes the acceptance test, and you are the one who rules on it.
What you keep: the source of truth, the workflow risk map, the guardrail set, the built AI teams, an acceptance test for each one, and the training. Four to six weeks, then ninety days of support.
The Boundary Build
The rules your organization runs these tools by, written down and small enough to follow. What each tool may touch and what it may not. Who signs off before an output leaves the building. What happens the first time one of them is wrong, decided before it is.
Every rule carries a pass-fail test written before the work begins, and your people run it. You rule on whether it passed. What you keep is a set of rules that survived contact with your own material, not a policy document nobody opens.
The Standing Brief
It opens with a written statement of where your organization actually stands: the answer you hand to clients, owners, insurers and boards when they ask, on a bid or in a renewal, how you use AI. Accurate, checkable, promising nothing you are not doing. Alongside it, privately, the things currently true that you should not say out loud yet, and what it would take to change them.
Then it continues on a fixed cadence, in two halves: what changed that could hurt you, and what changed that you could use. Both written against your business, not the market in general. The research is done before you see anything, so the brief costs you reading time and nothing else.
A defined first term, with continuation by agreement.
AI Leadership Intensive
One week for a leader who has to direct this work, approve it, and interrogate the people selling it, and who currently cannot. At the end you can read a vendor's accuracy claim and say what it does and does not measure, name what a tool is allowed to touch in your business and what it is not, and run the meeting instead of nodding through it.
Fractional Chief AI Officer
A senior seat for organizations that need the judgment without a full-time hire. Direction, governance, vendor calls, and the internal fluency to carry it after the engagement ends.
Strategic Advisory :: High-Stakes Decisions
One decision. Three weeks. A written brief containing the recommendation, the strongest available case against it, the specific evidence that would change the answer, and the conditions under which you should reverse it.
That last part is the unusual one. Almost nobody who gives advice writes down in advance what would prove them wrong. For the decisions where being wrong is not recoverable by trying again: a platform commitment you will live with for years, a raise, a market entry, a partnership, a restructure.
Brand Positioning
Every company now claims AI fluency, which means the claim no longer distinguishes anyone. Founder and company positioning that earns belief before the deck does, and holds when competitors are saying the same words. Often the work is an outside eye finding the partnership, the audience, or the funding source the people inside the mission stopped being able to see.
Messaging & Partnership Counsel
Advisory on the messages and the relationships that cannot be wrong. Which partnership is worth the leverage it costs. What is safe to say, to whom, and when. A company lives and dies on sales; a movement lives and dies on community engagement, and the messaging that drives one does not drive the other.
The Record
Why the First Group Exists.
Most of this practice has nothing to do with risk. The AI work is different, because it now has a public record, and reading five entries from it is the fastest way to see what the build is actually for.
-
The Chatbot Described a Refund Policy That Did Not Exist.
A customer relied on it, applied, and was refused. The airline argued to the tribunal that the chatbot was a separate legal entity responsible for its own answers. The tribunal rejected that and ordered the airline to pay.
Moffatt v. Air Canada · British Columbia Civil Resolution Tribunal · February 2024 -
Everyone on the Video Call Was Fabricated.
A finance employee, suspicious of an emailed request, asked for a call to confirm it. The chief financial officer and the colleagues he recognized had been built from footage the executives had published themselves. About 25.6 million dollars left in fifteen transfers. No system was breached and no credential was stolen.
Hong Kong Police, February 2024 · The engineering firm Arup confirmed as the company involved, May 2024 -
The Report Cited Research That Does Not Exist.
A review commissioned by a federal department for 440,000 Australian dollars was published with invented academic references and a quotation from a Federal Court judgment nobody wrote. The firm issued a correction, disclosed that a generative AI tool had been used, and repaid the final instalment of the contract.
Australian Department of Employment and Workplace Relations · Associated Press · October 2025 -
One Email, Never Opened, Could Send Internal Files Out.
Researchers showed that instructions hidden inside an ordinary looking email would be read by the AI assistant when it later searched a user's mail, causing it to collect internal documents and transmit them to an outside server. Nobody had to click anything. Microsoft rated it critical, fixed it, and says there is no evidence it was used against a customer.
Disclosed by Aim Security · CVE-2025-32711 · June 2025 -
One Operator, an AI Coding Agent, Seventeen Victims in a Month.
The agent ran reconnaissance, harvested credentials and moved through networks, and the model was left to decide which stolen records would hurt most and to write the demands accordingly. The targets included hospitals, emergency services, government bodies and religious institutions. Some demands exceeded 500,000 dollars.
Anthropic Threat Intelligence Report · August 2025
None of these needed a better model. Each one needed somebody to have decided in advance what the tool was allowed to do, and who owned the answer when it turned out to be wrong. A chatbot with no owner for what it says. A payment approval that a convincing face could satisfy. A report with no check on its own citations. An assistant permitted to read anything and reach anywhere. An agent holding more standing on a network than any person had granted it.
Those decisions are all makeable in advance. That is what the first group of engagements is.
On Accuracy Claims
A vendor's accuracy number grades what its tool found. What the tool never saw is not scored wrong. It is absent, and absence is the failure that costs money.
Who It's For
Principals Making Decisions They Cannot Revisit.
-
Construction, Real Estate, and the Built Environment
Document-heavy, judgment-heavy work where the vendors arrived faster than anyone's rules for using them.
-
Civic and Nonprofit Leadership
Mission-driven organizations where public trust makes a wrong AI decision cost far more than a right one saves.
-
Founder-Led Companies
Early-stage teams building the operating habits they will be stuck with, and answering to investors asking harder questions every quarter.
-
Media, Entertainment, and Creative Organizations
Where the craft itself is what AI is reshaping, and the strategic question is what stays human on purpose.
The practice pairs strategic design with senior technical depth. Trevor Davis sets the direction and builds the system. Ellis Cohen, twenty-five years a chief technology officer and chief information security officer, brings the engineering and security rigor that makes it deployable inside a real company.
Start a Conversation
Most Decisions Do Not Need This. Some Do.
If you are choosing a tool, you have vendors for that. If you are about to commit to something you cannot walk back, or you have AI running inside your business and nobody owns what it is allowed to do, that is the conversation worth having.
Tell me the decision in front of you. I will tell you plainly whether this is worth your money, including when it is not.
Trevor Davis Principal Los Angeles
Trevor@TheStrategyPractice.coNotes From the Practice
Sent when something changes that would change a decision. What arrived that could hurt you, and what arrived that you could use. Not a newsletter, and not often.
Your address is used for these notes and nothing else. One click to stop them.