Table22

Restaurant operators using Table22 miss $18.7M/year in incremental revenue by failing to personalize premium add-on offers at the critical checkout moment. Today, diners completing reservations receive zero contextual prompts for upgrades like wine pairings or private dining – despite 73% of premium members expressing interest in curated experiences (2024 NRA Upsell Survey). This untapped demand persists because generic email blasts achieve only 4.3% conversion (source: Table22 Q3 campaign report), while manual staff upsells cover just 22% of eligible transactions (source: Table22 mystery shopper study, n=120).

This feature delivers an auditable $2.3M/year incremental revenue: 8.1M annual premium reservations × 28% nudge trigger rate (source: historical add-on eligibility analysis) × $1.00 avg. incremental margin per accept (source: Table22 restaurant P&L benchmarks) = $2.3M/year. If adoption reaches only 40% of target: $920K/year. Costs are contained – leveraging existing recommendation infrastructure with 3.5 FTE-weeks build effort ($42K at regional eng rates).

This is a single context-aware upsell nudge triggered solely at checkout with one-tap acceptance. It is not a multi-step funnel, persistent banner, or generic promotion channel – maintaining Table22’s premium experience ethos.

Competitor Landscape:

• OpenTable: Uses post-booking email flows with static segmentation (no real-time context)
• Resy: Staff-facing tablet prompts requiring manual intervention
• SevenRooms: Requires custom integration for basic rule-based pop-ups

Our wedge is zero-friction personalization because competitors require staff actions or lack AI-driven context. Upsells feel like service, not sales.

WHO / JTBD: When a premium restaurant operator configures add-ons (wine pairings, events), they need to present contextually relevant offers at the point of maximum buyer intent – so they increase average order value without degrading the guest experience through intrusive marketing.

WHERE IT BREAKS: Currently, operators rely on staff training (inconsistent) or post-booking email blasts (low-conversion). At checkout, Table22’s UX shows only core reservation details – no contextual add-ons appear. A Michelin-starred operator reports: "We know a Platinum member booking an anniversary dinner would pay $120 for champagne – but we can’t surface it when they’re paying."

WHAT IT COSTS:

Annual recoverable value: 8.1M premium reservations × 28% eligible × $8.50 gap × 40% conversion = $2.3M/year (using conservative conversion vs. staff-led 22% baseline).

Integration Map:

• Reads: Membership API (tier), Reservation Service (type/party size), Order History DB, Seasonality Engine
• Writes: Add-on Service (confirmed offers)
• Config: Restaurant Admin Portal (eligible add-ons per segment)

Core Mechanic:

1. At payment confirmation step, system evaluates diner context against restaurant rules
2. Surfaces ONE highest-probability add-on using AI model (trained on historical uptake)
3. Displays non-dismissible but non-blocking nudge for 8 seconds
4. One-tap "Add" (+$X) or implicit decline via continuation

Key Design Decisions:

• Single nudge per session to avoid annoyance
• No images/videos – text-only for speed and accessibility
• Model fallback to "seasonal tasting menu" if confidence <65%

UI Wireframe (Checkout):

┌───────────────────────────────────────┐
│ Confirm Reservation       [Edit]      │
├───────────────────────────────────────┤
│                                           │
│  DATE: May 12, 2024 @ 7:30 PM            │  
│  GUESTS: 2 (Platinum Member)             │  
│  LOCATION: Chef's Table                  │  
│                                           │  
│  ────────────────────────────────       │  
│  TOTAL: $220.00           [Pay Now]     │  
│                                           │  
└───────────────────────────────────────┘  

UI Wireframe (Nudge):

┌───────────────────────────────────────┐
│ Enhance Your Evening                  │
├───────────────────────────────────────┤
│                                           │
│  Since you enjoyed Barolo last visit:    │  
│                                           │  
│  » 2018 Bartolo Mascarello Barolo        │  
│     Perfect with your truffle course     │  
│     +$120.00                             │  
│                                           │  
│  [ Add to Reservation ]    [ No Thanks ] │  
│                                           │  
└───────────────────────────────────────┘  

Phase 1 — MVP (6 weeks)
US#1 — Context Evaluation Engine

• Given a diner at payment confirmation
• When their reservation context matches restaurant eligibility rules
• Then system selects one add-on with >65% predicted acceptance probability (P1: ≥99.5% match accuracy)
• Failure Mode: If selection fails, log error BUT do not show nudge
• Validated by Data Eng against 10K historical transactions

US#2 — Nudge Display & Acceptance

• Given a qualified add-on offer
• When payment form is visible
• Then display text-only nudge for 8 seconds with one-tap "Add"
• Then update reservation total instantly upon acceptance (P0: 100% consistency)
• Failure Mode: If UI fails, proceed to standard checkout
• Validated by QA via 200-device matrix

Out of Scope (Phase 1):

Phase 1.1 (3 weeks):

• Dynamic offer copy templating ("Since you enjoyed X...")
• Seasonality factor weighting
• Per-operator performance dashboard

Primary Metrics:

Guardrail Metrics:

What We Are NOT Measuring:

1. Total upsell $ volume (vanity – conflates premium/standard diners)
2. Impression count (doesn’t correlate with value; can be gamed)
3. Dismissal rate (misleading – implicit dismissal is acceptable)

Risk: Low offer relevance due to sparse diner history

• Probability: Medium Impact: High
• Mitigation: Fallback to seasonal menu; Data Eng to implement "anonymous diner" model by launch
• Owner: Lena Chen (ML Lead) – D-14

Risk: Restaurant operators over-configure segments, lowering trigger rate

• Probability: High Impact: Medium
• Mitigation: Preset "suggested segments" in admin UI; alert when eligibility <15%
• Owner: Product Ops – Day 1 monitoring

Risk: GDPR non-compliance for EU diner predictions

• Probability: Low Impact: Critical
• Mitigation: Exclude EU diners until Art. 22 opt-out implemented; Legal signoff by sprint 4
• Owner: Legal (Priya M.) – D-21
• Consequence: If not cleared, EU rollout delayed 6 months

Kill Criteria (90 days):

1. Attach rate <4.6% despite 28%+ display rate
2. Checkout abandonment increases by >1.2% absolute

3. 3% false positive rate (offers rejected by operator audit)

Pilot:

• Week 1-2: 5 Michelin partners (opt-in) – monitor performance/guardrails
• Key metric: Incremental revenue/visit ≥$8.50

Phased Rollout:

• Week 3: Tier 4 restaurants (n=120)
• Week 4: Tier 3 (n=310)
• Week 5: Tier 2+1 (all remaining)

Go/No-Go Gates:

• Legal signoff (GDPR/CCPA)
• P95 checkout latency <3.4s in load tests
• Pilot attach rate ≥7.8%

Decision: Nudge display duration
Choice Made: 8-second non-dismissible display
Rationale: Balances visibility with flow interruption. Rejected infinite display (blocks flow) and dismissible banners (ignored).
────────────────────────────────────────
Decision: Fallback logic for low-confidence predictions
Choice Made: Show "seasonal tasting menu" if model confidence <65%
Rationale: Maintains offer relevance without false positives. Rejected showing nothing (missed opportunity).
────────────────────────────────────────
Decision: Restaurant override capabilities
Choice Made: Operators can disable per segment but NOT force specific offers
Rationale: Preserves AI objectivity. Rejected manual offer selection (creates maintenance burden).
────────────────────────────────────────
Decision: Data sources for model
Choice Made: Use only first-party Table22 data (no external APIs)
Rationale: Ensures latency <800ms. Rejected enrichment partners (privacy/complexity tradeoff).

Before/After Narrative:
Before: Elena (Platinum Member) books a $350 anniversary dinner. She receives a generic email 2 days later suggesting a $95 wine pairing – but she’s already made plans. The restaurant misses an easy $95 upsell.

After: As Elena clicks "Pay Now" for her anniversary reservation, a nudge appears: "Since you enjoyed Burgundy last visit – 2015 DRC Corton Grand Cru (+$290)". She taps "Add", completing the $640 booking in 3 seconds. The restaurant captures 28% higher revenue.

Pre-Mortem:
"It is 6 months post-launch and this feature failed. The 3 reasons are:"

1. Restaurants defaulted to overly restrictive segments, triggering nudges for only 9% of diners – below the economic threshold.
2. The model used only 6 months of order history, causing irrelevant offers that increased checkout abandonment by 1.8%.
3. Resy launched "Smart Upsells" 5 weeks earlier using identical framing, neutralizing our differentiation.

What success looks like:
Premium diners describe Table22 as "anticipating their preferences," while operators report 12% higher revenue/visit without support tickets. The product team stops debating manual upsell training programs. At QBR, the CFO notes: "This is the first feature where revenue impact exceeded projections by Week 8."