Personalised food marketing, powered by what people already love.
A multi-location restaurant group had a loyalty app sitting at 8% engagement. Their marketing was one-size-fits-all: the same discount, the same message, for every customer. The customer who orders spicy lamb every Friday was getting the same email as the Monday-salad regular. Nobody felt seen.
An AI recommendation engine that ingests POS transaction data and builds a vector embedding profile for each customer. Qdrant powers similarity search to cluster customers by taste preference. OpenAI generates personalised campaign copy for each cluster — so the Friday lamb customer gets a weekend special for lamb dishes, not a generic 10% off anything.
Campaign creation time saved
vs manual copywriting
To generate campaign copy
per segment
Saved per campaign
on average
Manual segmentation
fully automated
POS data pipeline
We built an ETL pipeline that pulls transaction data nightly, cleans and normalises it, and feeds it into the embedding model, turning purchase history into taste profiles.
Vector clustering
Qdrant stores high-dimensional embeddings for each customer. Similarity search clusters them by taste preference - dynamically, as new transactions arrive.
Personalised campaign engine
Campaign managers pick a segment, choose a goal, and the system generates copy variants for each cluster. One click. Reviewed by a human. Sent to thousands.
We take on a handful of projects each quarter.
Let's see if we're the right fit.
we read every message. yes, actually.
