James Delivery · 2021
James Delivery: a search that understands what users are looking for
One week to diagnose why a super app's search returned so little — and redesign it end to end, from research to prototype.
- My role
- Product Designer
Case study - Company
- James Delivery
- Year
- 2021
Summary
A case developed in one week for iFood's Product Designer selection process. The brief: pick a real problem in an existing digital product and solve it, showing the whole process from discovery to interface.
I chose search in James Delivery, an app spanning everything from groceries to services like electricians. In a catalog that wide, search is the shortest path between a person and what they want — but James' search returned little: only stores, no filters, no sorting, no history.
The Double Diamond guided the work: research and benchmarking in Discover, hypotheses and scoping in Define, flow, components and prototype in Develop. Since the project never shipped, the case honestly stops at the third stage.
The problem
Search was the app's front door, and it delivered little. The diagnosis crossed four sources: mapping the current flow, a six-player benchmark, an analysis of 2020 store reviews, and user interviews.
- 3.4 rating on Google Play, with search among the recurring complaints in 2020 reviews.
- Search returned only stores and restaurants: items, dishes, and products were invisible.
- Zero filters and zero sorting criteria, against five and six on iFood.
- No search history and no empty-state feedback: the screen simply went blank.
Challenge
Deeply understand the current search flow, map evidence, and turn it into hypotheses strong enough to support a new proposal — helping users reach relevant results faster.
Constraints
A one-week deadline, a third-party product with no access to internal data, and a commitment to stay consistent with the app's existing components. No market release: the case ends at the prototype.
Context
Experience analysis
I compared James' search with the biggest delivery players in Latin America, axis by axis: what the search offers, what each result shows, what can be searched, filters, sorting, and history.
The most complete search in the benchmark: filters by delivery mode, distance, fee, and payment, six sorting criteria, history with shortcuts, and results by dish and grocery item.
→ It became the redesign's completeness bar: autocomplete, history, and item-level results made the first-phase cut.
Results in distinct formats per category: store, grocery item, and dish each get their own presentation, with the information that matters.
→ It inspired the per-category result formats in the proposed search screen.
Search history that keeps even no-result queries, plus sorting by rating, delivery time, and popularity.
→ It reinforced history as a repetition shortcut — a pain mentioned directly in the interviews.
The only player in the benchmark that let users delete items and fully clear their search history.
→ User control over their own history entered the proposal from the first sketch.
Grocery-focused: item search with autocomplete and price sorting, but no filters and no history.
→ It showed that grocery item search is category table stakes, not a luxury — and James, used mostly for groceries, didn't have it.
Of the six players, James was the only one with no filter, no sorting, and no history at the same time. It wasn't a polish problem: these were basic category capabilities that simply didn't exist.
Approach
Best practices before opinion
I started with desk research across sources like NN/g and Baymard, distilled into a checklist of eighteen mobile search best practices: from a visible magnifying glass to autocomplete from the third character, down to the empty-result state.
Mapping the current flow
I mapped the real search flow screen by screen. The map exposed suggestions with no visual distinction between categories, results limited to stores, inconsistent tags across screens, and zero feedback when a search found nothing.
Listening to stores and users
I classified 2020 reviews from Google Play and the App Store by theme: item search, filtering, wrong results. Then five remote interviews went deeper into the pains. One quote sums it up: 'In this app I couldn't find where to filter the search. Either it doesn't exist, or it isn't easy.'
Hypotheses anchored in evidence
Nine hypotheses connected the evidence to business value, from a more visible search field to the effect of ratings on decisions. From them came the first-phase cut: seven deliverables, from the redesigned home to the filter screen.
Sketch before designing
Hand-drawn flows to explore paths while mistakes were still cheap, and a technical feasibility conversation before any screen: negotiating what ships now and what waits is part of the design.
Consistency with what already exists
I mapped the app's existing components to set the redesign's visual path. The proposal had to look like James — and be plausible to build with what the team already had.
The proposed flow
The ideal flow connects the proposal's four fronts: a home with search in evidence, an initial search screen with history and categories, autocomplete, and a results page with refinement.
What would come next
The deadline ended the case at the prototype. The next steps were already named: stress use-case scenarios, define success metrics, run usability tests, and refine technically with the development team.
Interactive prototype
Walk the proposed flow: from home to search, from autocomplete to sorted results. Tap the screen to start.
Layout & changes
Every screen decision is anchored in research evidence: best practices, the benchmark, store reviews, or an interview quote.
- 01
Home with search in evidence
Groceries become the default category, reflecting how the app is actually used. The delivery address gains prominence to prevent wrong orders, and search joins the tab bar and every page, making it clear you can look for an item or an establishment.
- 02
Initial search as a shortcut
History personalized by usage, with distinct formats for restaurant, grocery, and category, delivery time and closed status before the tap, and control to delete each entry.
- 03
Autocomplete that builds the query
Suggestions from the third character on, showing where each result lives: across all categories, in a restaurant, or in a grocery store. Less typing, fewer mistakes.
- 04
Results that help decide
Result count, sorting and filtering at hand, and cards with what the interviews asked for: distance, delivery fee, time, rating, and promotions in highlighted tags.
- 05
The dish without opening the restaurant
Most-ordered dishes expand right in the result, with prices, avoiding the second search inside each restaurant page that users described as tiring.
A one-week case that ends at the prototype: with no market numbers, what remains is the method — evidence before hypotheses, hypotheses before screens, and every layout change pointing back to the research that justified it.
Next project
Handshake: reliability with low friction