BAD Marketing is in active recovery — today is day 1 of 6
Turbo Tint bookings fell from ~20+/week to 1–2/week on flat ad spend. The client threatened to fire DAB on a Jul 16 call and gave a week and a half to show recovery, with two weeks of contract runway left. The "Code Red" plan started this morning — see the checklist on that tab.
Oasis Housing
Progress
Every tracked metric for this bot, in one place — the breakdown for each of these lives in its own section further down.
Bot 1 of 1
Claim & Landlord Bot
Ingests insurance-relocation claims from email, logs them to the claim board, searches Furnished Finder (and Zillow/Redfin as backup) for matching rentals within budget and a 5-mile radius, sends the landlord-required initial outreach message, answers basic logistics questions (pets, timing, availability) over Quo, and generates a proposed "family sheet" with Oasis's markup and vendor fees applied for Ana/Roeben to approve before it goes to the insurance rep. Hard-flags to a human the instant pricing or negotiation comes up.
Action Items Need Access To Start
Check off as each is delivered — the progress bar above tracks this list.
Recommended build & rollout sequence
DAB engineering judgment applying the playbook's automation/bot/human principles to its own dependency structure — each phase gates the next. Check off items as they close.
Resolve all 10 blockers before building anything client-facing — each one underpins a whole category of downstream steps, not just one.
Every deterministic step, no LLM judgment needed — lowest risk, and it directly hits Oasis's own pain point: claims take ~2 minutes by hand today and "speed is key" was their own phrase on the kickoff call.
Language/judgment steps, but every one drafts-then-a-human-sends — no autonomous sends yet. Validates difficulty scoring, priority setting, and drafting quality against real human review.
Run the bot against sample or historical claims before it ever touches a live family or landlord. The playbook flags this phase as entirely unscoped (D-13) — the four items below are a starting point DAB would still need to define with Oasis, not sourced from the document.
Remove human-in-the-loop only on individually-approved, low-risk steps first. Every hard guardrail stays human-gated indefinitely, regardless of phase, until each is separately re-approved with the client.
BAD Marketing — Turbo Tint
Progress
Every tracked metric for this recovery, in one place — the breakdown for each lives in its own section further down.
What has to be true by then
- Bookings visibly climbing, bot + setter combined, posted every weekday since Jul 20
- Zero conflicting-message reports in the 5 working days after the automation shutoff
- No new "doesn't sound human" complaints since the Claude switch
- Wins landing in the client's own pipeline automatically
The consequence
- Two weeks from the Jul 16 call is when the client decides on installers and the contract
- If conversions haven't recovered by Jul 27, this is when it's decided — the 3 days after are notice, not recovery time
Turbo Tint — Aurora Bot
5.9 / 10DM-setter for the Aurora, CO shop: qualifies inbound leads over SMS/DM, quotes pricing, sends the $29/30 deposit link, confirms payment, books in Shopmonkey.
- Core flow exists: intake, qualification, pricing, deposit link, Shopmonkey booking.
- Confirmed bug: collects a $29 deposit against a contracted $30.
- No
deposit_resume.pyrecovery path (Arvada has one) — stalled bookings aren't reclaimed. - Payment/booking flow not verified end-to-end on live traffic.
Turbo Tint — Arvada Bot
6.5 / 10DM-setter for the Arvada, CO shop: same flow as Aurora, plus a script that resumes stalled bookings once a deposit link goes cold.
- Same core flow, plus a
deposit_resume.pyrecovery path Aurora lacks. - Same $29-vs-$30 deposit bug confirmed here too.
- Slack escalation channel is hardcoded to Aurora's channel ID — cross-contamination risk.
get_pricing.pyhas 12–16 undifferentiated backup files — ongoing pricing instability.
Live bookings — the metric the client is actually watching, not the 15-item scope score above.
Needed for 100% (beyond surviving Jul 27)
- Ship every Phase 1–4 item on schedule and clear the Jul 27 checkpoint criteria above.
- Fix the confirmed $29-vs-$30 deposit bug on both Aurora and Arvada.
- Fix the Slack channel cross-contamination (Arvada's fallback hardcoded to Aurora's channel).
- Get the payment / Shopmonkey booking flow verified end-to-end on live traffic, not just the two bookings seen so far.
- Close out the remaining amber/red items on the original 15-item scope checklist for both locations.
- Taper the human setter back out once the bot's conversion rate holds on its own for a week.
Recovery checklist
Sourced from DAB's internal "Code Red" tracker, built off the Jul 16 client call. All dates/times are US Eastern; weekends don't count as working time. Check items off as they ship — this drives the progress bar above.
Nothing in later phases starts until this ships — tone and the automation cleanup are the two things the client can see and feel immediately.
The client wants conversions now, not a bot on principle. A human DM setter buys the runway the Phase 4 rebuild needs.
The old stack tagged every converted lead "MindFlowed AI Win" — a free, pre-labelled dataset of what actually worked on this account.
Working theory: the previous agency's edge was a thick automation stack around a thin bot, not a better bot. Phase 1 stripped that layer — this puts DAB's own back under it.