Machine learning for marketing

Recommendation engines are the most visible ML use case in marketing. Amazon attributes 35% of its sales to its recommendation system. They work by analysing user behaviour (what they’ve purchased, viewed, searched) and finding similar patterns in other users to suggest relevant products.

Two main approaches exist: collaborative filtering (users similar to you bought X) and content-based filtering (if you liked this product with these attributes, you’ll like this one). The most advanced systems combine both approaches with deep learning to maximise relevance.

Dynamic pricing adjusts prices in real time based on demand, competition, inventory and customer profile. Airlines and hotels have used it for decades, but ML has democratised access: any ecommerce can now implement sophisticated pricing strategies.

Pricing models analyse historical sales data, price elasticity by product and segment, competitor prices and seasonality to recommend the optimal price that maximises margin or volume depending on the business objective.

Traditional segmentation groups customers by static demographic data. ML enables segmenting by predictive behaviour: not just who they are, but what they’re going to do. This allows creating audiences based on purchase propensity, churn risk or affinity with a product category.

Meta and Google Ads’ lookalike audiences use ML internally to find users similar to your best customers. But more advanced companies build their own audience models with first-party data, which are more precise and don’t depend on third-party cookies.

ML is already built into major advertising platforms. Google Ads’ Smart Bidding, Meta’s Advantage+ and programmatic DSP algorithms optimise bids, creatives and audiences automatically. The challenge for marketing teams isn’t activating these features, but understanding how they work to feed them correctly.

ML-based attribution models (data-driven attribution) outperform rule-based models (last click, linear) because they capture each channel’s and touchpoint’s real contribution to the conversion journey. Google Analytics 4 and tools like Segment or Mixpanel offer this capability.

Personalisation at scale combines ML with automation to adapt messages, offers and experiences to each individual user. Doing this manually for thousands or millions of users isn’t feasible, but a trained model can personalise in milliseconds.

You don’t need to build models from scratch. Most marketing tools already integrate ML: Smart Bidding campaigns, Klaviyo’s recommendations, HubSpot’s scoring. The first step is activating and understanding these features, not reinventing the wheel.

For more advanced needs, services like Google Cloud AutoML, AWS SageMaker or BigQuery ML let you build custom models without a dedicated ML team. The key is starting with a clear use case, measuring impact and only scaling what works.