The problem
The funnel was a map of a journey nobody took.
Most insurance content was written as if readers moved in a line, awareness, consideration, purchase, retention. Trigger, evaluate, decide, convert. The funnel diagram on the strategy deck said one thing; the analytics said another. Sessions looped. Visitors left to a comparator, returned three days later, opened the same page twice, exited again. The data refused the diagram.
This isn't a Belgian quirk. Google's Decoding Decisions research, run by The Behavioural Architects across 31,000 purchase journeys, named the pattern out loud: between the initial trigger and the final purchase sits the messy middle, a non-linear loop of exploration and evaluation. People expand options, then narrow them, then expand again. They don't move through a funnel. They circle one.
In regulated B2C, insurance, assistance, fintech under MiFID, public sector, the messy middle is longer and more anxious. The decision carries financial or health consequences. The legal language slows reading. The product is intangible. And the audience isn't one persona. It's at least four, all reading the same article for incompatible reasons.
The four invisible clusters. Reconstructed from on-site behavior (Hotjar / FullStory), search-intent data and CRM event timelines, none of them appeared in the official persona deck.
- The risk-averse researcher, multi-session, deep scrollers. Reading guarantees and exclusions before any quote.
- The price-sensitive comparator, short sessions, high quote-tool hits, in parallel tabs from comparators.
- The trigger-event newcomer, single triggering moment (move abroad, new baby, new car). High intent, low product literacy.
- The post-incident reactivator, already a customer. Filed a claim. Now reading the fine print they didn't read at sign-up.
Each cluster carried its own bias profile, its own exit triggers, its own reading rhythm. A single linear funnel served none of them, and the analytics showed it. Bounce on the article was symptomatic; the cause was upstream, in the architecture.
The approach
Stop designing the journey. Design the room they explore in.
I stopped trying to push the four clusters into one path. I designed a hub, a topic-level surface that gathered every angle on a question (e.g. "travel insurance for long stays"), structured so that any cluster could enter, loop, expand, narrow, and find an exit on their own terms. The hub wasn't a landing page. It was an architecture.
Mapping six biases to six components.
Google's research named six cognitive shortcuts that consistently shape decisions inside the messy middle. I treated them as design tokens, not marketing tricks, each bias got an explicit, dedicated hub component, with a single behavioral job to do.
FIG. 01, Hub architecture
Topic surface (one screen) → six bias-mapped components arranged for non-linear entry → four cluster-aware exit paths (quote, save, compare, contact). [diagram on file · NDA]
- Category heuristics → a short-form spec block at the top: 4–6 product attributes in plain language, no marketing copy. Lets the comparator triage in under 15 seconds.
- Power of now → a contextual "what changes today" callout: the immediate consequence of choosing now vs. waiting. Speaks to the trigger-event newcomer.
- Social proof → real numbers (claims paid, average response time) rather than testimonials. Regulator-safe, researcher-credible.
- Scarcity / urgency → carefully bounded. Used only when honest (regulatory window, policy renewal date), never as artificial countdown.
- Authority bias → source citations, regulator references, expert reviewers named on the page. Critical for the risk-averse researcher.
- Power of free → reframed as "included by default", assistance, second-opinion, app access. No "free gift" theatre.
Non-linear entry, cluster-aware exit.
The hub had no obligatory reading order. Each component was self-contained and entry-ready. A reader could land on "exclusions" from a Google search, scroll up into "what's included by default," sidestep into "how to compare with a competitor," and exit through the quote tool or the contact form, all without backtracking. The architecture absorbed the loop; the user didn't experience it as one.
Exits were instrumented per cluster. The risk-averse researcher saw a "save this for later" path that emailed a structured summary. The comparator got a one-click "open quote in new tab" with prefilled inputs. The newcomer got a contextual chat prompt (human-staffed, no bot). The reactivator got direct access to claim-status content. One hub, four exits.
Results
Measurable outcomes, reported when shareable.
The hub framework is currently in rollout across five markets. KPIs below are reported when they clear NDA review, until then, the pattern of impact is consistent enough to describe directionally.
6
Cognitive biases
Mapped to six dedicated hub components, each with one behavioral job.
4
Reader clusters
Reconstructed from on-site behavior, not persona workshops.
▲
Time on hub
Increase observed in pilot market, figure pending NDA clearance.
▲
Quote completion
Lift on quote-tool entries from hub vs. direct funnel.
What changed in practice.
- The funnel diagram came off the wall. The hub map replaced it.
- Editorial briefs now start with a bias profile per cluster, not a tone-of-voice slide.
- Quote-tool entries from hub pages overtook direct-funnel entries in the pilot market within the first quarter.
- Legal-review cycles shortened, because authority and scarcity components were built regulator-safe from the start.
- The team stopped optimizing CTAs in isolation and started optimizing the geometry between components.
What's actually being tested. Not the copy. Not the CTA label. The geometry, which component sits next to which, in what cluster's reading rhythm, is the unit of experimentation now. That's the shift the framework forced.