# The Strategy Grounding Kit From "Applied AI for Marketing Ops" (appliedaiformops.com) by Lily Luo. Companion to my post "Executing at the Speed of Strategy." The premise: AI can only execute as fast as your strategy if your strategy is written down somewhere AI can read it. Most strategy lives in people's heads. This kit gets a usable slice of it onto a page this week. **Pick your level:** - **Level 1 — Manual.** Two documents, filled in by hand. - **Level 2 — The full setup.** For Claude Code or Cowork users: a strategy folder extracted from your real artifacts and wired into your working prompts. - **Level 3 — Hand it to your agent.** Paste this whole file to your AI agent and say "help me set this up." The last section is instructions for the agent. --- ## Level 1: Two documents, by hand ### Document 1: The strategy one-pager (30 minutes) The context you'll attach every time AI drafts, prioritizes, or plans for you. Write it for a smart new hire who starts Monday: ``` WHO WE SERVE Our best-fit customer, in two sentences. Who we do NOT serve. WHAT WE'RE TRYING TO DO THIS QUARTER Top 3 priorities, ranked. For each: why it's a priority right now. WHAT WE SAY NO TO The requests, channels, or tactics we decline, and why. This section does the most work. AI without it says yes to everything. HOW WE DECIDE The 3-5 factors we actually weigh (business impact, ROI, opportunity cost, effort, speed) and roughly how we trade them off. GUARDRAILS Claims we never make. Topics that need legal or brand review. Numbers that must come from an approved source. ``` Keep it to one page. A one-pager attached to every prompt beats a strategy deck nobody opens. ### Document 2: The decision record (15 minutes per decision) Pick one recurring decision you made this week and write down the reasoning, not just the outcome: ``` DECISION: [what you decided] DATE: OPTIONS CONSIDERED: [including the ones you rejected] WHY THIS ONE: [the factors that actually drove it, honestly] WHAT WOULD CHANGE MY MIND: [the condition that would flip this decision] CHECK BACK: [date + what result would prove this right or wrong] ``` One per week. After a month, patterns you didn't know you had will be visible. A warning from my own experience: you can codify a bad heuristic as easily as a good one, which is what the CHECK BACK line is for. ### Putting them to work Attach the one-pager whenever AI plans or drafts for you: "Use this as the source of truth. Flag anything in your output that conflicts with it." Attach decision records when you ask AI to help prioritize: "Here's how I've actually decided in the past. Apply the same reasoning, and tell me where my past reasoning seems inconsistent." When strategy changes, edit the one-pager the same day: a documented strategy can change at the speed of the market instead of the planning cycle. --- ## Level 2: The full setup (Claude Code or Cowork) The manual version depends on you remembering to attach documents. The full version makes the strategy folder something your AI tools consult on every task, and it's built by extraction rather than introspection. ### The folder ``` strategy/ one-pager.md the living strategy document from Level 1 philosophies.md your distilled beliefs, with quotable lines and where each one came from decisions/ 2026-07-14-declined-event-sponsorship.md 2026-07-16-shifted-budget-to-abm.md ``` Git-tracked if you can: the diff history of `one-pager.md` becomes a record of when and why your strategy actually changed, which is a document most teams have never seen. ### Extract, don't introspect Self-reported strategy is what you'd like to believe; extracted strategy is how you actually decide. Point Claude Code at the evidence: QBR decks, planning docs, the emails where you declined things, your published writing. My instruction was a version of: > Read these. Draft my one-pager and philosophies file from the evidence, quoting the source for each claim. Where my stated priorities and my actual decisions disagree, flag the conflict instead of resolving it. That last clause is where the value is. The conflicts are your real strategy talking. ### Reconstruct the decision records You don't have to wait a month. Give the agent 2-3 recent real decisions and let it interview you: what were the options, what actually drove the call, what would have changed your mind. The "WHY THIS ONE" section must be in your words; the honest reasons, not the official ones. ### Wire it in Three instructions make the folder live instead of shelf-ware: 1. **Ground everything:** "Before any drafting, planning, or prioritization for me, read `strategy/one-pager.md`. Cite which sections you used. Flag conflicts instead of silently resolving them." 2. **Prompt the check-backs:** "When a decision record's CHECK BACK date passes, remind me and record what happened." Reasoning that keeps failing its check-backs gets flagged, not reused. 3. **Propose updates, never apply them silently:** "When my choices start contradicting the one-pager, say so." Either the behavior or the document needs to change, and that's your call. This pairs with the Judgment Folder kit: strategy is what you're trying to do, judgment is how you make calls along the way. The two folders together are what let an AI draft something you'd actually sign. --- ## Level 3: For your AI agent *Human: paste this whole file to your agent and say "help me set this up." Agent: follow this section literally.* 1. Ask the human one question first: the two hand-filled documents (Level 1) or the folder version (Level 2)? Build only what they choose. 2. **Interview, don't invent.** Fill in the one-pager one section at a time. Push hardest on "WHAT WE SAY NO TO"; humans skip it. If the human has artifacts (planning docs, QBR decks, declined-request emails), ask for them and propose draft sections from the evidence, quoting sources, for the human to correct. 3. **Extract decision records from history.** Ask for 2-3 recent real decisions and interview the human until the real reasons surface, not the official ones. "WHY THIS ONE" must be in the human's own words. 4. **Set up the files** as shown above, git-tracked if available. 5. **Ground everything you produce** in the one-pager; cite the sections you used; flag conflicts instead of silently resolving them. 6. **Track check-backs.** When a decision record's CHECK BACK date passes, remind the human, record the outcome, and flag reasoning that keeps failing. 7. **Propose updates, never apply them silently.** When the human's choices contradict the one-pager, say so and let them decide which changes. --- Questions, or a decision pattern you didn't expect to find? Contact lily@lilyluo.ai.