Customer Experience

AI Personalisation EngineReal-Time Recommendations

Real-time personalisation using Vertex AI Recommendations AI and BigQuery ML — recommending products, content, or offers based on user behaviour, purchase history, and semantic intent. Sub-100ms recommendations served at scale.

Explore Vertex AI Services

What We Build

We implement a production-grade personalisation engine on Google Cloud — ingesting behavioural events via Pub/Sub, training recommendation models in Vertex AI, storing recommendations in Cloud Spanner for low-latency serving, and delivering results via a lightweight API your frontend calls at render time. A/B testing and model monitoring are included from day one.

Vertex AI Recommendations AIBigQuery MLGemini EmbeddingsPub/SubCloud Spanner

The Problem It Solves

Generic recommendations failing to convert

One-size-fits-all product recommendations ignore individual user behaviour, intent, and history — resulting in irrelevant suggestions that customers simply ignore.

Low click-through rates on product suggestions

Product recommendation widgets deliver poor CTR because they rely on simple popularity metrics rather than semantic understanding of user intent and context.

Revenue leakage from poor cross-sell

Without accurate next-best-offer logic, organisations leave significant cross-sell and upsell revenue untapped — customers never see products they would actually buy.

What You Get

Real-Time Recommendation Scoring

Sub-100ms recommendation scoring powered by Vertex AI Recommendations AI — personalised results served at page load time without perceptible latency.

Multi-Signal Personalisation

Recommendations factor in real-time behaviour, semantic purchase intent from Gemini embeddings, purchase history, and session context simultaneously.

A/B Testing Framework

Built-in experiment framework for testing recommendation strategies, placement, and ranking models — with statistical significance tracking in BigQuery.

Cold-Start Handling for New Users

New users without purchase history receive recommendations based on session behaviour and semantic similarity to cohort profiles — no cold-start gap.

Explainable Recommendation Reasons

Each recommendation can surface a plain-language reason — 'because you browsed X' or 'customers like you also bought Y' — improving trust and conversion.

Business Impact

+31%
CTR on product recommendations
+24%
Average order value uplift
$2.1M
Additional revenue attributed in first 6 months

Frequently Asked Questions

How quickly can the model learn from new user behaviour?

Vertex AI Recommendations AI supports real-time event ingestion via Pub/Sub. User behaviour events (page views, add-to-carts, purchases) are processed within seconds and influence recommendations in near real-time. The model's ranking updates continuously within the same session, so a user who changes their browsing intent mid-session receives contextually updated recommendations.

What product catalogue size does the system support?

Vertex AI Recommendations AI is designed for large-scale catalogues and handles millions of SKUs efficiently. Cloud Spanner provides the low-latency storage layer for serving recommendations at scale. There is no practical upper limit on catalogue size for the recommendation engine — the system is purpose-built for enterprise e-commerce and content catalogues.

How does Gemini Embeddings improve recommendation quality?

Gemini Embeddings creates dense vector representations of product descriptions, attributes, and user query intent. This enables semantic similarity matching — recommending conceptually related products even when there is no explicit co-purchase history. A user browsing 'sustainable outdoor gear' will receive semantically related recommendations even for new catalogue items with no behavioural data.

Can we use this for content recommendations as well as products?

Yes. Vertex AI Recommendations AI supports product, content, and media recommendation use cases with the same underlying infrastructure. We have implemented content recommendation engines for media publishers, learning platforms, and financial services using the same technical approach. The implementation is adapted to your catalogue type during discovery.

Build This for Your Organisation

Turn generic recommendations into personalised experiences that convert. We implement a production-ready Personalisation Engine in 3 weeks.