OpenApply
2026
Lead Scoring: how AI helps admins prioritize students
I rebuilt OpenApply’s Lead Scoring around one score per lead and added AI that suggests factors and surfaces families the rules miss, raising adoption by 160%.
Overview
Why most schools never turned Lead Scoring on
Preface
In 2022, I designed the first version of Lead Scoring. Most schools never turned it on. In 2026, looking back, the problem wasn’t that it measured too many things. It was that it squeezed different things, like how well a family fits the school and how engaged they are, into a single Hot-Warm-Cold ladder. Admins thought they were comparing leads on one scale, but they weren’t.
Background
Admissions teams receive far more enquiries than they can follow up on at once. Lead Scoring is meant to answer one question for them: who should I contact next?
OpenApply already had a Lead Scoring feature, but schools had to build the configuration from scratch, and most never turned it on. When I looked at the setup flow, the deeper issue was the logic itself.
Bands came first. Admins created score bands (Hot, Warm, Cold), then defined separate scoring rules inside each band.
Bands didn’t share a scale. A Hot lead and a Warm lead could be scored on unrelated criteria. The system simply let the higher score win.
The result was hard to explain. Staff couldn’t say why one lead ranked above another, which made the scores hard to trust or act on.
Problem
Most schools never turned on Lead Scoring. Its setup made each score band follow its own rules, so a “Hot” lead and a “Warm” lead were often judged on unrelated criteria.
What I did
Rebuilt the scoring model around one score per lead, backed each scoring factor with conversion data, and designed AI features that suggest factors and surface families the rules miss, with staff making the final call.
15
schools interviewed
160%
adoption rate increased
6
usability tests
Design
Rebuilding the model around one score
1. One score per lead, bands as ranges
Admins now define scoring factors first, and every lead gets one total score. Hot, Warm and Cold become simple ranges on that score, shown on the settings page as “One score per lead. Bands are ranges on that score.”
Each band card shows how many leads fall in it, so admins see the effect of a threshold change right away.
2. Points backed by evidence, not guesswork
Each factor row shows its conversion rate and how many leads currently match it. Instead of guessing how many points a factor deserves, admins can see which signals actually lead to enrollment. Negative factors, such as unsubscribing from an email, are marked in red.
AI
AI that assists, and admins decide
Challenges
Challenges and solutions
Problem 1
Years ago, we decided to implement scoring in a multi-factor way, which led to low user adoption and made it hard to iterate. With little background on the situation, we didn’t know how to deal with the legacy design.
Solution 1
I conducted competitive research and came up with a clear diagram to discuss with the PM. We decided to adopt the new way of scoring.
Problem 2
Even with the new design, we found that some schools still struggled to understand what a lead score is and what its value is.
Solution 2
After listening to user feedback, we created templates for users to refer to, and received positive feedback on them.
Lessons
What I learned
Learning from failure
I designed the first version of Lead Scoring, and I got the model wrong. I focused on giving admins flexibility, so each band could have its own rules. I didn’t test whether admins could understand how those rules related to each other. They couldn’t, and most schools never turned the feature on.
Back then, lead scoring was a new concept for school advancement teams. We weren’t able to recruit enough users, and we only conducted 2 user interviews to decide the direction.
The lesson I took from it:
Rather than pursuing flexibility, a better solution might be to find a method that makes it easier for users to get started before iterating.
OpenApply users sometimes find it difficult to accept new features due to being accustomed to long-standing operating models, so finding ways to make new features easy for them to learn is likely more important.
