Case Study — Reaching New Users Sooner — Straight Up AI
Case Study · e-commerce · £10m a year

A £10m increase in sales:
serving 2.6 million additional customers

The client's recommender only learned from people with an established history, leaving a large group of newer users on generic results. We halved the threshold to pull them in, without diluting recommendations for everyone already there. The result, 50,000 more customers were served tailored recommendations each week, worth ~ £10m in additional sales per annum.

Client
E-commerce
Service
AI Engineer · Recommendations
Focus
Cold-start coverage
Stack
Python · PyTorch · TorchRec · Databricks · Airflow · SageMaker
The Challenge

The users who needed help most got the least

The client's recommender only learned from people with an established history, leaving a large group of newer users on generic results.

The obvious move — just lower the threshold — carries a real risk. Bringing in users with less history can result in existing, high value users recieving less personalised recommendations. The change had to maintain performance for existing users whilst improving activation for new ones.

The Approach

Widen the funnel, guard the baseline

=
=
+
The Results

New users, activated

£10m Sales
Driven by new users spend from personalised recommendations.
+2.6m Users Reached
New users brought into personalised recommendations.
+12.4% Return Rate
New users are 12.4% more likely to buy again.

Leaving new users on generic results?

See if we're a fit →
Next
Next

Negative Sampling