Skip to content
TILT Handbook

Decision log

Standing decisions and the reasoning behind them. Add to the top as new ones are made. This prevents re-arguing settled questions and explains non-obvious choices to a future reader.

Business & positioning

Section titled “Business & positioning”
  • Consulting is positioned as “for complex businesses,” not “for real estate firms.” Real leads (a clean-tech CFO, an 8-figure operator) confirmed the audience is industry-agnostic. RE Analyst is the one exception (RE-bound).
  • May 2026 consulting launches are a revenue-recapture move, not a pivot. Follow-on revenue share fell from ~89% to ~19%; CFO/FP&A/RE Analyst/DICE/ Municipal/Startup exist to win it back. Judge them on a 6-12 month horizon, not launch-month lead counts.
  • RE pages stay primary. RE is 96% of the funnel; do not deprioritize it to chase the new lines.
  • Pricing stays off all public surfaces. Addressed via a no-numbers posture line and the free consultation.

Marketing

Section titled “Marketing”
  • The page split (focused variant pages) was the right call. 2024’s conversion crash was low-quality ad traffic on a catch-all; tighter pages reduced mismatched leads. Do not undo it.
  • Reduce Google Ads dependence rather than scale it. The channel works but is the dependence to reduce; growth levers are research, email, co-marketing, podcasts.
  • No competitor named publicly. Multiple competitor audits concluded TILT wins by pivoting to its own strengths.

Technical

Section titled “Technical”
  • Website deploy stays manual paste-into-Divi. Deliberate trade for direct control over a git-to-prod pipeline. See Deploy process.
  • Rank Math schema on model pages moved off Product type. Product schema caused Google Search Console Merchant/Product errors; corrected toward WebPage. If GSC Product/Merchant errors reappear, start at notes/archive/rank-math-page-config-2026-05-13.md.
  • QUIC.cloud abandoned. Performance work settled on LiteSpeed + Imagify; QUIC.cloud was evaluated and dropped. History in notes/current/audits/.
  • Rent tool capped at 45 RentCast calls/month (free tier is 50) and caches aggressively, so it cannot surprise-bill; it degrades to the free HUD figure.
  • Customer bot: user-managed surfaces first, website widget deferred. Lower risk; the website widget, when built, uses a custom Anthropic API path (~$1-11/mo) over a $40+/mo SaaS because Megan can build it.

This handbook

Section titled “This handbook”
  • Markdown-in-repo is the single source of truth, published privately to Cloudflare Pages behind Cloudflare Access (free, email-PIN, on the existing stack). Keeps it AI-readable and continues to feed the bot and llms.txt. Editing is Claude Code + github.dev; no new app. Rationale and alternatives considered: the approved plan at /home/megan/.claude/plans/i-d-like-to-work-quizzical-panda.md.