AI was counting chairs in restaurants

Challenge

The provider of payment terminals wanted to increase its share in the Austrian restaurant market in 12 months. The problem was that they did not know which restaurants had the highest likelihood of changing providers and how to approach them correctly.

Main Challenges

• Low market share - they wanted to grow by 270 new restaurants per year

• 9,000 restaurants in the market - they don't know which ones to approach first

• Lacking data on restaurant size (number of seats, revenue)

• They could not identify hot leads - wasting time on unsuitable prospects

• Generic offers did not work - low conversion rate

Solution

We used AI analysis of 500,000 customer reviews and 40,000 restaurant photos to estimate size (number of seats) and customer satisfaction. The CNN model identified tables and chairs in photos, while NLP extracted sentiment from reviews. We created a lead scoring system from 0 to 100 and approached the top 1,500 restaurants with personalized offers.

How We Addressed This

We analyzed 500,000+ customer reviews using NLP algorithms. We extracted sentiment, mentions of payment options, and dissatisfaction with the current terminal. We knew which restaurants were looking for better solutions.

The CNN model analyzed 40,000 photos from Google Maps and Instagram. AI counted tables and chairs and estimated restaurant capacity with 85% accuracy. This way, we could predict revenue even without financial data.

We created a lead scoring model with 88% accuracy that combined restaurant size, sentiment from reviews, location, and type of cuisine. The top 1,500 restaurants received the highest scores and were approached first.

Each restaurant received a personalized approach via email, LinkedIn, and phone call. The message included specific benefits for their type of business (e.g., 'for restaurants with 60+ guests like yours, we have...').

Segmentation into 3 groups: Tech Innovators (early adopters, want the latest solutions), Growth-Oriented (growing rapidly, need scalability), Traditionalists (reliability and long-term partnerships). Each group received different messaging.

Technologies used: Convolutional Neural Networks (CNN), NLP (NLTK, spaCy), Gradient Boosting Machine (GBM), TensorFlow & Keras, Multi-channel outreach

Results

1. 270 contracted restaurants: Significant increase in market share

2. Significant market share increase: High conversion rate from 1,500 approached restaurants

3. New annual revenue: €2M+ ARR from new clients

Wow Factor

The AI model was able to estimate the number of seats in a restaurant just from photos on Instagram and Google Maps. AI analyzed interior photos, counted tables and chairs, and estimated capacity with 85% accuracy. This way, we could predict revenue and identify the most valuable leads without the need for a phone call.