ChildBlossoms started as a straightforward brief: help parents in Ghana find and compare schools. The existing process was a combination of word-of-mouth, WhatsApp groups, and hours of research that still left families uncertain about whether they'd found the right fit.
The product we ended up building was significantly more complex than the brief suggested — and the most interesting problems weren't the technical ones.
What we were asked to build
A platform where parents could search verified schools by location, curriculum, fees, and other criteria. Schools would have profiles. Parents would filter, compare, and contact.
Straightforward directory product. We've built variations of this before.
What made ChildBlossoms different was the matching requirement: parents shouldn't have to understand the Ghanaian school system to find a good match. The platform needed to ask the right questions and surface the right options, not just return a filtered list.
The matching problem
The naive implementation — filter by curriculum, location, fees — returns too many results, in the wrong order, for parents who don't know what "IB curriculum" means or why the school 8km away might be better than the one 2km away.
We built a structured intake that asks questions parents can actually answer: age of the child, what matters most (academic, arts, sports, mixed), how far they're willing to travel, rough budget. From these inputs, we rank schools against a weighted criteria model.
The model isn't AI in the machine learning sense. It's a deterministic scoring system with coefficients calibrated against feedback from the schools themselves and from early beta parents. We considered a more sophisticated approach but concluded that explainability mattered more than marginal accuracy gains. Parents wanted to understand why a school was recommended, not just that it was.
What broke in production
School data quality. Schools submitted their own profiles. The variance in data quality was significant — some schools had detailed, accurate information; others had outdated fees, wrong curriculum labels, or missing location data. The matching algorithm is only as good as the data it ranks against.
We built a moderation layer and a structured submission process that forced schools to provide specific fields in specific formats. Existing profiles had to be verified manually before going live. This added two weeks to the timeline and was not in the original brief.
The anxious parent problem. Early testing revealed that parents who got a match they weren't expecting — even a clearly good one — didn't trust it. They wanted to understand the reasoning. We added a "why this school" explainer to each recommendation, surfacing the two or three criteria where the school scored strongly for that parent's inputs. Conversion from recommendation to contact improved significantly after this change.
Performance under school admissions season. The platform launched just before Ghana's peak school admissions period. Traffic was higher than projected. We'd built for steady-state load; the seasonal spike required infrastructure changes in the first month that we hadn't planned for.
What we'd do differently
The data quality problem was predictable. Any platform that depends on third-party-submitted data needs a content pipeline built before the product launches, not after. We should have been more insistent on this in the scoping phase.
The trust problem — parents not believing the recommendation — was harder to anticipate. It's a lesson about the gap between what a recommendation system optimises for (matching quality) and what users actually need to act on a recommendation (confidence and comprehension). Those are different things, and building for one doesn't automatically give you the other.
Where it is now
ChildBlossoms has over 100 verified schools live on the platform, across Greater Accra and beyond. The matching system is running in production. The school submission and verification pipeline is automated. School profiles are updated on a schedule rather than manually.
The most meaningful metric: parents who use the matching flow contact schools at a significantly higher rate than parents who browse the directory without it. The structure of the product — asking questions before showing results — changes the quality of the match and the confidence of the person acting on it.