B2B SaaS
Recommendation Engine for a B2B SaaS Platform
Client: B2B SaaS Platform
A product analytics company wanted to surface personalised feature recommendations to enterprise customers — but months of vendor effort had delivered a prototype that could not handle production load.
The Challenge
The client had invested months with an external vendor to build a recommendation engine, but the resulting prototype could not scale beyond a small test cohort. Response times exceeded several seconds under moderate load, the model had no cold-start strategy for new customers, and the recommendation quality degraded sharply for enterprise accounts with complex usage patterns. The team needed a production-ready system that could serve recommendations at scale while integrating cleanly with the existing product.
The Solution
We rebuilt the recommendation engine from the ground up on the client's existing infrastructure. The new system combined collaborative filtering with content-based signals to handle both established and new customers. We designed an A/B testing framework to measure recommendation quality against business metrics, giving the product team data-driven confidence in what was being surfaced to customers.
Our Approach
- 1Analysed the failed vendor prototype to identify specific scalability bottlenecks and recommendation quality issues
- 2Designed a hybrid recommendation architecture combining collaborative filtering for established accounts with content-based fallback for cold-start scenarios
- 3Built the recommendation service on the client's existing infrastructure, avoiding new vendor dependencies and reducing operational overhead
- 4Implemented a batch-plus-realtime pipeline: pre-computed recommendations refreshed daily, with real-time adjustments based on session activity
- 5Created an A/B testing framework with configurable experiment allocation, allowing the product team to measure impact on adoption and engagement metrics
- 6Developed an internal dashboard showing recommendation coverage, diversity, and click-through metrics across customer segments
Outcomes
- Production recommendation engine deployed and serving all customer accounts within 10 weeks
- Sub-200ms response times under full production load, down from multi-second latency in the vendor prototype
- Cold-start recommendations available from day one for new customer accounts
- Measurable increase in feature adoption among customers receiving recommendations, validated through A/B testing
- Product team empowered with self-serve A/B testing and recommendation analytics dashboard
Technologies & Capabilities
Have a similar challenge?
Let's discuss how we can help your team ship AI into production.
Book a Strategy Call