Challenge
A SaaS client had a large customer base of over 1M but no insight into customer segments and their Ideal Customer Profile due to a lack of data. This resulted in a very low marketing ROI due to a one-size-fits all approach
Approach
We took a three step approach:
1. Data augmentation - a web scraper was developed to collect publicly available information on customers own online content. This raw data was synthezied into common characteristics using a custom AI pipeline
2. Similarity calculations - a large language model (LLM) was used to calculate customer similarity and cluster customers in high dimensional vector space. A fine-tuning app was developed to further train and improve performance of the model realizing a 90% accuracy when mapping to the segmentation hierarchy.
3. Real time segmentation - customer segmentation model was deployed into production enabling new customers to be segmented in real-time after sign-up.
Impact
2X increase in the number of customers that could be segmented when compared to the previous ways of working. This allowed marketing teams to design highly tailored campaigns around specific industry verticals and geographic regions.