An 87% conversion rate masked a critical segmentation failure. 90% of users were registering as the wrong account type, flooding sales with manual reclassification work and polluting the CRM. The fix wasn't more clarity. It was better architecture.
Tree-Nation is a B2B and B2C platform where companies gift trees to employees or customers, who then register to claim and plant them. On paper, the funnel was healthy: an 87% registration conversion rate. Strong numbers, no obvious red flags.
But underneath the surface, 90% of users were selecting the wrong account type. Most were employees who had received a gift tree from their company. When registering to claim it, they selected "Company account" because their mental model was simple: this came from my company, so I pick Company.
“The issue wasn’t conversion. It was data quality and segmentation failure.”
Core diagnosisThe system technically knew users were arriving from a gift email, but since gifts could also be sent to companies, automatic account type assignment wasn't possible. The misclassification cascaded through the entire operation: sales manually reclassifying hundreds to nearly two thousand users per month, a CRM corrupted with bad data, and users landing in business-oriented dashboards when they expected a personal experience.
Volume varied with company campaigns and seasons. At peak, nearly two thousand users per month entered sales' reclassification queue, a task that had become part of the team's daily routine.
Reclassification wasn't a one-off. It was a persistent drain: daily manual effort that had normalized into the sales workflow, making the underlying design problem invisible to anyone not looking at the data.
We approached this as a diagnostic problem first. Funnel analysis with tracked events (traffic source, account type selected, post-registration tree planting), combined with Hotjar session recordings, journey mapping, and competitive benchmarking.
Tracked events from entry point to post-registration activation revealed the drop-off pattern and confirmed the scale of misclassification. Traffic source data was key: the majority of wrong selections came from gift email entries, users whose goal was singular and immediate.
Hotjar recordings showed users scanning quickly and selecting “Company” without reading the explanatory copy. This was the first signal that clarity-based solutions would have limited effect. The cognitive load of “which type am I?” was being resolved by environmental cues, not by what the form said.
Mapping the full gift recipient journey exposed the mental model collision: the user's context was "my company gave me this" and the registration UI presented an equal-weight choice between Citizen and Company. The architecture was creating a false choice where there was none.
They had one objective: claim the tree. Account type selection was an obstacle, not a meaningful decision for them. This meant that no amount of explanatory copy would reliably guide behavior; users weren't reading because the question didn't feel relevant to their goal. The intervention needed to be structural, not communicative.
We worked closely with the PM and CEO throughout. The CEO was initially hesitant to make deep changes: the form had required significant prior investment and was converting well. Getting from Iteration 1 to Iteration 2 required three months of data and a shared shift in diagnosis: the problem wasn't clarity, it was hierarchy.
Our first hypothesis was that the misclassification came from insufficient explanation. We restructured the form to surface the account type decision earlier, added clearer copy, and introduced friction for the business path to slow down users who might be making the wrong choice instinctively.
The 37% improvement was meaningful, but misclassification remained too high. Session recordings from this period confirmed the core issue: users still weren't reading. The intervention had improved the form, but the underlying architecture still treated both account types as equally valid choices for every user arriving at the page.
After three months of post-Iteration 1 data, and a candid conversation with the PM and CEO about what the recordings were telling us, we shifted strategy. Instead of explaining more, we changed hierarchy.
The insight was that segmentation doesn't need to be technically detected when it can be behaviorally guided. 90% of users arriving at registration were citizens. We designed for that reality.
This was a deliberate choice to guide segmentation through architecture rather than explanation. Business users who needed a company account could still find the option, but they would need to actively seek it, which is appropriate behavior for someone with a genuine business intent.
The final iteration delivered a 70% improvement in account classification accuracy, measured over three months against the pre-intervention baseline. Overall conversion performance remained strong throughout both iterations.
Beyond the headline number: CRM data quality improved significantly, post-registration experience alignment improved (users landing in the dashboard that matched their actual context), and sales recovered operational bandwidth that had been consumed by a problem the team had normalized over time.
The form had required significant prior investment and was still converting well at the surface level. Proposing deeper architectural changes (ones that deliberately reduced the visibility of a whole account type) felt risky from a product ownership standpoint.
What got us through it was data. Three months of post-Iteration 1 recordings and classification metrics made the case that clarity-based interventions had a ceiling. Once the PM and I reframed the question from “how do we explain the choice better?” to “should this even be an equal choice?”, the path forward became clear, and the CEO came with us.
This is the kind of stakeholder work that doesn't show up in a Figma file. Learning to build the case for a structural intervention (not just a visual one) is a different skill than design craft, and one I'm actively developing as I move toward Product Lead roles.
Running Iteration 1 before proposing the more aggressive Iteration 2 turned out to be the right sequencing. It gave us three months of evidence that the incremental approach had a ceiling, which made the behavioral reprioritization feel like a logical next step, not a risky gamble.
Looking back, the session recordings from before Iteration 1 already showed users not reading copy. I could have formed the “architecture over explanation” hypothesis sooner. The three-month gap between iterations was useful for stakeholder alignment, but the design insight was probably available earlier.
The most interesting thing about this project is that it wasn't really about the form. It was about understanding that when user intent is dominant and goal-driven, trying to teach users something mid-task is a losing battle. The right move is to design for the behavior you actually observe, not the behavior you wish users had.
Deliberately reducing the visibility of an option felt counterintuitive at first. But it was the most honest design decision we could make: one that served the 90% without eliminating the path for the 10%. That's the kind of thinking I'm most proud of here. Not the UI. The argument.