The problem
The article ended. So did the relationship.
A customer would read a post-purchase article, "what to do if your luggage doesn't arrive", "how to declare a minor accident from abroad", and then leave. The article ranked, served its informational job, and the session ended. From the team's dashboards, this looked like success: page views up, average time on page strong, scroll depth healthy. The article was doing exactly what content marketers ask content to do.
But behaviorally, the article was the end of a moment, not the start of one. The customer had just experienced a friction point with the brand (the lost luggage, the small accident). They were emotionally ready to consider, to compare, to evaluate, to engage. And we were closing the tab for them.
McKinsey's update to the Consumer Decision Journey named the opportunity. The classical four-stage CDJ, consider, evaluate, buy, bond, describes one cycle. The compressed CDJ describes what happens after: the most retained customers stop cycling through evaluation altogether. They enter what McKinsey calls the loyalty loop, a short, fast loop between experience and buy, bypassing active consideration because trust is already there.
The shift, in one sentence. The job of post-purchase content isn't to inform the customer. It's to place them back inside the loyalty loop, at the right moment, with the right next thing, before they go evaluate a competitor instead.
The editorial CMS treated articles as standalone documents. There was no infrastructure to route a reader to a next-best piece based on what they had just read, who they were, or where in the loyalty loop they were sitting. Related Articles, where they existed, were tag-based, generic, and identical for every reader. The loop was theoretical. The content stack didn't support it.
The approach
If Netflix can recommend the next episode, content can recommend the next decision.
I rebuilt the post-article surface as a recommendation system, not a tag-based footer. The unit of design was no longer the article, it was the transition between articles. Where Netflix optimizes for "the next thing watched," I optimized for the next thing read inside the loyalty loop.
Four content clusters, mapped to four loop moments.
I split the post-purchase content library into four behavioral clusters, each tied to a moment in McKinsey's loop where a different next-best action becomes available.
- Experience cluster, content read during an active service moment (claim, assistance call, mid-journey). Next-best: a service-completion shortcut, not another article.
- Aftermath cluster, read in the 72 hours after a friction event. Emotionally loaded. Next-best: an empathy-led explainer that closes the loop on what just happened, before any cross-sell.
- Renewal cluster, read near contract anniversaries. Loyalty-loop primary surface. Next-best: a contextual comparison with what the customer already owns, surfaced as "what's new since last year", not a hard upsell.
- Adjacent-need cluster, read by customers exploring something tangential (travel article from a home-insurance customer; family-protection article from a single-life-policy holder). Next-best: a cross-sell, but framed as continuity, not novelty.
Recommendation logic, layered.
The Related Articles block became a three-layer system, each layer overriding the next when its signal was strong enough:
FIG. 02, Recommendation logic
Layer 1, collaborative filtering (what readers like this customer read next, anonymized).
Layer 2, cluster-aware curation (rules per loop moment, set by editorial).
Layer 3, hard regulator overrides (legal / compliance constraints non-negotiable). [logic on file · NDA]
Layer 1, Behavioral, anonymized. A collaborative-filtering signal: readers who finished this article most commonly read X next. Calibrated on rolling 30-day windows. Equivalent to Netflix's "because you watched."
Layer 2, Editorial, cluster-aware. The editorial team set explicit transition rules per loop moment. Aftermath cluster → never serve a cross-sell within 72 hours of a claim event. Renewal cluster → surface "what's new since last year" before any new-product link. These rules overrode Layer 1 when active.
Layer 3, Regulator-safe constraints. Hard overrides set by Legal and Compliance, surfaced through a structured config. Never recommend product X to a reader who hasn't met requirement Y. No exceptions, no soft handling.
UI: minimal, mono-purpose.
The Related Articles block stopped showing three or four options. It showed one, the next-best read, with a quiet "see other suggestions" affordance behind it. The single-choice frame matched the loyalty-loop logic: the next thing isn't an option among many; it's the next move. The interface had to honor that.
Results
Measurable outcomes, reported when shareable.
The system is live on the editorial surface of the pilot markets. Below: structural outcomes, with KPIs reported when they clear NDA review.
4
Content clusters
Mapped to four loop moments, each with explicit transition rules.
3
Recommendation layers
Behavioral · editorial · regulator-safe. Overrides cascade downward.
▲
Next-article CTR
Material lift versus tag-based Related Articles in pilot. Figure pending.
▲
Loop entries
More customers re-entering the loop within 7 days post-friction event.
What changed in practice.
- Editorial briefs now declare a cluster and a loop moment, not just a topic and a tone.
- The CMS taxonomy was rebuilt around behavior, not content type. "News," "FAQ," "Tutorial" became internal metadata, no longer the routing axis.
- Compliance shifted from a blocker to a layer in the system. Their rules are now machine-readable constraints, not late-stage red-pen sessions.
- The team stopped measuring "page views per article" and started measuring session length and loop re-entry rate.
- Cross-sell, when it appears, no longer looks like cross-sell. It looks like the natural next read.
The principle, repeated. Retention in regulated B2C isn't won with a renewal email at month eleven. It's won by being the brand whose post-purchase content closes the loop the customer is already inside.