Website personalization: what is practical for a mid-size business

Every rule depends on a signal, and the signals available to an anonymous visitor are fewer than vendors imply. Rank them by reliability.

Reliable on the first visit. Device type and screen size. Approximate location from IP, accurate to the metro area most of the time, less so on mobile networks. Referral source and campaign parameters. Time of day and day of week. Whether the visitor has been here before, if a cookie survives.

Reliable after some behavior. Pages viewed in this session. Products or categories viewed. Cart contents. Search terms used on your site.

Reliable only when the visitor identifies themselves. Email, account, purchase history, lifecycle stage, company. This is the layer where personalization gets interesting and it requires a login, a form fill or a click from an email that carries an identifier.

Not reliable. Third-party data about an anonymous visitor's demographics, interests or employer. Reverse-IP company identification works on corporate networks and fails on home and mobile connections, which is most traffic. Treat these as hints for B2B at most.

Most of the value for a mid-size business comes from the first and third categories. The first is cheap and immediate. The third is what you build toward.

These are the segments that produce different behavior often enough to be worth a rule, in the order most businesses should try them.

1. New versus returning. Returning visitors do not need the explanation; they need the shortcut to what they came back for. Swap the hero for a recently viewed panel, a reorder prompt or a continue-where-you-left-off link.
2. Geography. Show the nearest location, local phone number, regional shipping promise, local inventory or local events. For businesses with a service area, confirm that you serve the visitor's region on the first screen.
3. Device. Mobile visitors on a service business site want a tap-to-call button and hours. Desktop visitors are more likely to read. Personalization by device is really prioritization.
4. Traffic source. A visitor arriving from a paid search ad for a specific product should land on that product, not the homepage. A visitor from a comparison article should see the comparison. Match the page to the promise that brought them.
5. Existing customers. Logged-in or identified by an email click, customers should not see acquisition offers they cannot use. Show account, support and expansion content instead.
6. Cart or intent stage. Someone who has added to cart and returned should see the cart, shipping reassurance and any relevant incentive, rather than the brand story.

What you need depends on how many rules you intend to run.

Built into your platform. WordPress, Drupal, Shopify, Magento and most marketing automation suites can show different blocks based on location, device, logged-in status and, with HubSpot's smart content or similar, lifecycle stage. For the first few rules this is usually enough.

Testing tools with targeting. Optimizely and VWO both let you target experiments and experiences to audiences built from the signals above, and both have moved their personalization features into higher-priced plans this year. Google announced Optimize in September, a free tool in beta that connects to Google Analytics audiences; it is worth requesting an invitation if you already live in Analytics.

Dedicated personalization platforms. Evergage, Dynamic Yield, Monetate, Adobe Target and Qubit build behavioral profiles and recommend content or products automatically. Pricing generally starts in the low thousands of dollars per month and rises with traffic. They earn their cost on ecommerce sites with large catalogs and six-figure monthly traffic, or on content businesses whose revenue depends on time on site. Below that, the money is usually better spent on the rules and the testing.

Recommendation engines. For ecommerce specifically, a product recommendation service such as Nosto or Barilliance covers the highest-value personalization, "people who viewed this bought", at lower cost than a full platform.

Personalization has a measurement trap. A rule shown to returning visitors will always look good compared with new visitors, because returning visitors convert better anyway. The only honest comparison is between members of the same segment who did and did not see the personalized version.

So treat each rule as an experiment. Define the segment. Split it, with 80 to 90 percent seeing the personalized experience and the rest seeing the default. Pick one primary metric that reflects money: conversion rate, revenue per visitor, lead rate, or a step-completion rate for a defined funnel. Run until you have enough conversions in both groups to trust the difference, usually at least a few hundred conversions in total, which for many mid-size sites means two to four weeks per rule. Keep a log of what was tested, for whom, and what happened.

Expect modest, compounding gains rather than a single dramatic lift. Well-run programs typically report improvements in the range of 5 to 15 percent on the targeted metric for the segment in question. A rule that produces nothing after a fair run should be removed, not kept because it seemed sensible.

Personalizing before the default works. If the homepage confuses everyone, showing six confused versions of it does not help. Fix the base experience first.

Too many rules too soon. Every rule is a variant someone has to maintain, check on each release and understand when results are odd. Ten rules you do not test are worse than three you do.

Creepy signals. Greeting an anonymous visitor by the name of their employer, or referencing a location more precisely than a metro area, makes people uneasy. Use what you know to be helpful, not to show off.

Breaking caching and speed. Personalization that runs server-side can defeat page caching; personalization that runs client-side can cause visible content flicker and slow the page. Measure load time before and after, and keep personalized elements small.

Forgetting the crawl. Search engines should see a complete default page. Personalized content should be additive.

No ownership. Rules created during a campaign outlive it. Assign someone to review the active rules every quarter and retire the ones nobody remembers.

We start with discovery: analytics segmentation to find where visitor groups actually behave differently, a review of what signals the site and its systems can reliably provide, and a check of the default experience's performance. From that we plan a short prioritized list of rules with hypotheses and success metrics. Build covers the audience definitions, the content variants, the test setup and any integration needed to carry identity from email or CRM into the site. Then we improve on a rolling cycle, adding rules as earlier ones prove out. The work connects growth, brand, web and technology: the segments come from marketing data, the variants have to stay on brand, the site has to stay fast and the systems have to share what they know about the visitor.

A representative engagement is the ecommerce work we did with a retailer whose homepage was built for first-time visitors while most of its revenue came from repeat customers. We started with four rules: a recently viewed and reorder panel for returning visitors, local shipping terms by region, landing content matched to paid search campaigns, and suppression of first-order offers for identified customers. Each ran with a holdout, and the ones that moved revenue per visitor stayed. Only after that did a recommendation service make sense, and by then the team knew how to judge it.