A Practical Guide to Marketing Personalization
How marketing personalization works under the hood: rules-based versus algorithmic approaches, the data pipeline behind them and which one to build first.
Marketing personalization gets talked about as one thing, but mechanically it’s two different approaches wearing the same name. Understanding which one you’re building determines how much effort it takes and what kind of results to expect.
Rules-based personalization
This is the simpler, older approach: if a visitor meets a specific condition, show them a specific thing. A returning customer sees a “welcome back” message. Someone who abandoned a cart gets a follow-up email three days later. Someone browsing from a mobile device sees a simplified layout. Every rule is written by hand, which means it’s easy to understand and easy to debug, but it also means it only covers the situations someone thought to write a rule for.
Rules-based personalization is the right starting point for most businesses, because it requires no machine learning infrastructure, just a segmentation tool and a handful of clearly defined conditions. It also fails predictably: a rule either fires or it doesn’t, and when something looks wrong you can trace exactly why.
Algorithmic personalization
This approach uses a model, trained on past behavior across many customers, to predict what a specific individual is likely to want, rather than relying on rules a person wrote in advance. Product recommendation engines are the clearest example: instead of “if you bought X, show Y,” the system learns patterns across the entire customer base and applies them to each visitor individually, including patterns no one on the marketing team would have thought to encode as a rule.
The tradeoff is opacity. When an algorithmic recommendation looks strange, it’s often hard to say exactly why the model produced it, and these systems typically need a meaningful volume of behavioral data before their predictions outperform a well-built set of rules. A small business with modest traffic often gets better results from five carefully considered rules than from an algorithmic engine with too little data to learn from.
The data pipeline behind either approach
Regardless of which method you use, personalization depends on the same basic pipeline:
- Collection. Behavioral data (page views, purchases, email engagement) gets captured, usually through your analytics tool, ecommerce platform and email service provider.
- Unification. That data gets tied to a single customer identity, whether through a login, an email address or a cookie, so behavior across channels connects to one profile instead of sitting in silos.
- Segmentation or scoring. The unified profile gets grouped into a segment (rules-based) or scored by a model (algorithmic).
- Delivery. The relevant experience, an email, a website module, an ad, gets served based on that segment or score, ideally close to real time so the personalization still feels relevant.
Most personalization failures trace back to a break somewhere in this chain, not to the sophistication of the personalization logic itself. A perfectly designed recommendation engine is worthless if customer identity isn’t unified across devices, because it ends up treating the same person as several different strangers.
Real-time versus batch
Some personalization can run on a schedule, updating segments and scores overnight, which is simpler to build and sufficient for things like weekly email content. Other use cases, like an on-site recommendation that should reflect what someone just clicked, need real-time processing, which is more technically demanding and usually the last piece a growing team should build, not the first.
Where to start
Map your customer journey, pick the single moment where a more relevant experience would most plausibly change behavior (often a cart abandonment email or a post-purchase follow-up), and build that one thing well with simple rules before reaching for anything algorithmic. Personalization done narrow and well beats personalization attempted broad and shallow, every time.
More from the Marketing Analytics archive, everything published in 2017 and the Digital Business Marketing Awards.
