$233k in Bookings at an Average Cost of $9: Systematic Customer Acquisition for a Restaurant in Miami

Key challenges before launch

• no proper booking tracking
• inefficient optimization without data on real actions
• generic creatives without offer-level segmentation
• a third-party booking widget limited analytics

What we did

Analytics

The key step was implementing Google Tag Manager. It made it possible to track bookings, see user behavior, and pass data into the ad systems. Without this, optimization would have been “blind”.

Google Ads

We launched campaigns optimized for conversions. The algorithm learned from bookings and interactions, which made it possible to focus on the audience that actually books.

The main emphasis was on intent-based search campaigns that captured hot booking demand.

Meta Ads

We split campaigns by specific offers: happy hours, wine nights, caviar events, and rooftop bookings. Each offer had dedicated creatives and a dedicated message.

This increased ad relevance and improved engagement and conversion metrics.

Creatives

We tested atmosphere, food, events, and people. The best-performing combination was “emotion + experience”, not just food visuals.

Funnel

The funnel was built in two scenarios: ad → website → booking, or ad → interaction (warm-up) → booking.

Optimization

After tracking was set up, we turned off weak campaigns and scaled the effective ones. The algorithms started working far more accurately.

Results

total budget (Google + Meta) — ~$12,183

1,256+ conversions generated (bookings and actions)

total conversion value — $233,175

average conversion cost — $9.70

ROAS — ≈19.1

Google Ads: ~1,040+ conversions, average cost — ~$8.6

Meta Ads: 255 bookings, average cost — ~$23.4

Meta reach — 101,000+ users

Advertising generated real bookings, not just traffic. The system worked steadily and predictably.

Conclusion

Even in a niche with limited tracking, you can build an effective performance system if you set up proper data transfer, optimize for real actions, and work with user intent.

This makes it possible to get stable and scalable results.