Roxanne's isn't a normal taqueria — it's a single-unit fast-casual + a 260K-YouTube creator brand + a paying community. Most of the standard restaurant tooling (loyalty, online ordering, POS) is already solved via Toast. What's NOT solved is what makes Roxanne's different: the catering pipeline, Sknny Chef's comment + DM volume, the operator content engine, and the bilingual lunch rush. That's where these 10 modules go.
Roxanne's catering business is real (ezCater 4.9★, dedicated catering page, school fundraiser program) but inquiries flow through 5 channels with no consistent qualification or follow-up. This module catches every inbound catering ask — ezCater, phone, web form, Sknny Chef DMs, Instagram comments — qualifies it (date, headcount, dietary, budget), drafts a quote, and routes warm ones to Josh or Jesse for a 5-minute close call.
Josh's content does 100K+ views per post. The comment volume is unmanageable. The engine reads every comment + DM across YouTube, TikTok, IG, FB, classifies into 4 buckets (qualified catering lead / Inner Circle prospect / content idea / noise), drafts replies in Josh's voice, and surfaces a Monday morning inbox of 8-12 things only Josh can answer. The rest auto-handle with on-brand replies.
The lunch rush phone problem. 60% of 12:00-1:00pm calls go to voicemail because the counter is jammed. AI picks up in 2 rings, takes orders in Spanish or English (38% of the lunch base prefers Spanish), drops the ticket directly into Toast, and texts the customer their receipt. ~$3-7K/mo of currently-missed orders captured. See live dashboard →
Reviews repeat one complaint: "employees are lovely but terrible at assembling takeout orders correctly." Every takeout bag gets verified against the Toast ticket by an AI vision check at the expo station (camera over the bag), flags missing items before they leave, prompts a fix. Won't catch 100% but will catch the obvious ones (missing chips, missing salsa, wrong protein).
40% of one-time taqueria guests never come back. AI spots them by name + phone in Toast, drafts a personal text invite ("Hey Maria, missed seeing you — al pastor's on tonight"), sends from the restaurant's number. Recovers 8-15% of churned guests in 90 days without burning the brand. Cross-references Sknny Chef Inner Circle list so we don't text Inner Circle members like generic customers.
4.3 Google · 4.2 Restaurant Guru · 4.5 DoorDash. Reviews are positive but reply velocity is weak. AI drafts brand-voice replies in the language the review was written in, flags any review needing human attention (1-2 star with substance), and pushes happy customers to leave a Google review via SMS at the 90-min post-meal mark.
Sysco / US Foods / Restaurant Depot / local distributor invoices fed in via email or scan. AI normalizes SKUs, benchmarks against your 90-day baseline + regional market, flags any drift >5%, suggests sourcing alternatives. Taquerias swing harder than most restaurant categories on avocados, queso fresco, masa, cilantro, lime — the watchlist is tuned to that.
Predicts tomorrow's lunch + dinner covers from Toast historicals, weather, Stoughton local events (high-school games, Patriots away schedule), Sknny Chef video drop schedule (his big drops drive in-restaurant pilgrimages). Outputs a prep list with quantities by protein. Saves 6-12% on food waste, kills Saturday-night stockouts.
Pulls 8 weeks of Toast covers + the team's known availability and constraints (Jesse always Mon/Wed/Fri evening, two part-timers school schedule). Outputs a 2-week schedule in 5 minutes. Includes weather + local event adjustments. Edits in real-time when someone calls out.
Cameras in the kitchen + dining room (already there for Sknny content) feed continuously into the engine. AI clips the funny moments, the rush-mode footage, the customer reactions. Drafts captions in Josh's voice for each channel (YouTube Shorts, TikTok, IG Reels, FB). Josh approves the top 3 per day, the rest get archived for the Inner Circle playbook library. Cuts content production from "Josh's only job" to "Josh's 20-minute Monday review."
Each module ships incrementally on top of the Defender foundation. No big-bang. Josh and Jesse keep doing exactly what they're doing today.
| Month | Ships | Phase | Time impact |
|---|---|---|---|
| 1 | 01 Catering Pipeline · 02 Sknny Chef Comment Engine | Phase 1 | Josh's inbox drops from 200/day to 12/day |
| 2 | 03 Phone-Order AI (bilingual) | Phase 1 | Lunch rush capture · ~$200/day recovered |
| 3 | 04 Order Accuracy QA | Phase 2 | #1 review complaint pattern eliminated |
| 4 | 05 Lapsed Guest · 06 Review Engine (ES+EN) | Phase 2 | Map-pack rank lift + 8-15% guest recovery |
| 7 | 07 Food Cost Monitor | Phase 3 | Jesse catches drift before the P&L does |
| 8-9 | 08 Prep Forecasting · 09 Schedule Optimizer | Phase 3 | 4-6 GM hours/wk · 6-12% waste cut |
| 10 | 10 Sknny Chef Content Engine | Phase 3 | Josh's content production becomes a 20-min Monday review |
| 11-12 | Optimization · cross-module tuning · Inner Circle integration | Phase 3 | The whole system runs on autopilot |