How We Designed Search for a Nigerian Classifieds Site With 11,000+ Active Listings

Search is the heart of any marketplace. Get it wrong and buyers leave; get it right and you have a moat. Building search for a Nigerian classified marketplace surfaced a set of constraints worth sharing for anyone working on commerce products in emerging markets.
The problem: 11,000+ active listings across 23 categories, refreshed daily, accessed primarily on low-end Android devices over LTE. Listings span phones, cars, property, fashion, electronics — items where the buyer's mental query is rarely a clean keyword.
Constraint 1: Mobile network latency. A search response that takes 800ms on a good connection takes 3+ seconds on a Lagos LTE network at rush hour. We learned to cap payload size aggressively (return 20 results, never 50) and pre-compute popular category counts so the filter sidebar never blocks the result list.
Constraint 2: Location is a primary index. "Phones in Lekki" should not be slower than "phones in Nigeria." We treated state and area as first-class index dimensions rather than post-filter conditions. This single change cut median query time roughly in half on location-scoped searches.
Constraint 3: Typo tolerance matters more than relevance tuning. Nigerian users searching for "tokunbo car" and "tokumbo car" and "tokumbo cars" all expect the same results. Fuzzy matching at the tokeniser level returns more value per engineering hour than fancy ranking algorithms.
Constraint 4: Recency beats relevance for classifieds. Unlike Amazon, a classified listing has a life cycle measured in days. A 4-day-old "perfect match" is worth less than a 4-hour-old "good match" because the seller is more likely still reachable. Our default sort is recency, with relevance as a tiebreaker.
Constraint 5: Empty states are the most-viewed screen. When someone searches for a niche item and gets zero results, that screen needs to do more than apologise. We populate it with closely related categories, recent listings in the same area, and a "notify me when posted" CTA — turning a dead end into engagement.
You can see this in production at blinkersnigeria.com — a classified marketplace built on these constraints, operating across all 36 states in Nigeria.
The meta-lesson: marketplace search in emerging markets is not a watered-down version of Amazon search. It is a fundamentally different optimisation problem — driven by network conditions, locality, and listing lifecycles unique to the market.