Job search app, a design exercise

I take complex products and make them feel simple. Eight years across trading, security, maps, food delivery and AI learning. Now healthcare.

At Huma I design Atlas, a UK healthcare product, and build the AI workflows our design team runs on, from task audits to developer handoff. I work in code with Claude, not in Figma.

Before that I led design for Andarz, Wallex’s AI learning product. I built its design team from zero to four and developed the M3 framework for how an AI tutor should behave. I write about it, and about what else changes when products start to think.

--:-- in Yerevan

The business problem

Deals were everywhere, and nowhere

Promotions at Snappfood come in many forms: cashback, eco food, food packs, coupons, discounts. They exist to attract people who want to pay less. But they were fragmented across the app, so the people who cared most about them had the hardest time finding them.

Vendors couldn’t showcase offers

There was no place at the start of the journey to put a promotion in front of customers.

Promotions lacked impressions

Low visibility meant low engagement, so vendors saw little return on their offers.

Deal-seekers took too many steps

Users looking specifically for promotions couldn’t reach those vendors in a few taps.

Build different discovery journeys for different user segments, so each one finds the promotions it actually wants.

Prioritizing

Three ways in. We picked one.

We could grow the business impact of promotions in three ways. Customer experience was where the existing problems hurt most, and it would teach us how the system really behaved before we touched the vendor side.

More variety of promotions

Increase the kinds of offers available.

Better vendor tools

Improve promotion creation and reporting for vendors.

We started here03

More promotion use by customers

Help customers find and redeem promotions.

Data gathering

What the funnel told us

We started with a few questions. Do users understand what a promotion is worth? Is behavior consistent across promotion types? Does it matter how many promotions a user engages with, or how many get redeemed overall? To answer them, we analyzed user and business data along the product funnel.

Average order value rose about 3% in a year

It came from higher prices, not from changed behavior.

Some promotions lost money

Cashback in particular had become less profitable.

High cognitive load

Many promotion types meant more to process and longer loading times.

It looked like spam

Too many promotions with too little explanation felt like noise.

The old version

Three systems, three entry points

Two kinds: a percentage of the order credited back, or a fixed amount regardless of order size. Users met cashback in their profile (balance and history) and at checkout (what they’d earn).

Percentage and fixed cashback — Right: a percentage of the paid amount. Left: a fixed amount.

Right: access from checkout. Left: access from profile.

Coupons unlock when a condition is met: free shipping over 890,000 tomans, 30% off a first order, a free side with a specific burger. They appeared on vendor cards, filters, menus and the basket. Users rarely filtered by coupon; most activated one from the menu once their basket qualified.

Coupon access through vendor card and filters — Through the vendor card, filter suggestion and filter modal.

Coupon section in menu and basket — Through the coupon section in menu and basket.

Parties run on four services (food, fruit, sweets and meat) and sit on Snappfood’s home page. They discount selected items, and drive about 4.7% of Snappfood’s sales.

Party types: food, fruit, sweets and meat.

Target users & metrics

Who it’s for, and how we’d know

Active cashback users

Know the benefits and actively look for lower spending.

Price-sensitive users

Search for discounts and promotions to save on every order.

Food Pack visitors

Already aware of Food Pack and come to learn what it offers.

Higher micro CR from discovery to menu for vendors with promotions.
Less time spent on discovery.
More use of displayed promotions: active cashback, discounted vendors, personal discounts, coupons, parties, buy-one-get-one.

Total micro CR from discovery to menu, starting at 79%.
Drop-off rate during discovery.
RPD (revenue per delivery) as a key indicator.

The risk we named up front: redirecting a big share of homepage traffic (for example, half of it) to a new page we didn’t yet understand could pull discovery-to-menu conversion down.

Hypothesis

A dedicated page for deal-seekers

Give price-sensitive users one discovery page that features vendors with promotions, and make its impressions and conversion match the home page and listings. The page needed three ways in.

A prominent place in the app’s navigation menu.

A dedicated section or filter inside the home page and vendor listings.

Promotion Center entry on home and listings

Promotional campaigns and notifications that drive awareness and bring people back.

Campaigns & notifications

Promotional campaigns and notifications point people to the Promotion Center to build awareness and engagement.

Design process

What others do with deals

We reviewed the promotion, reward and offer sections of competitors for first and last impressions, and for patterns that could work for single promotions and for bundles of products or vendors.

Benchmark of first and last impression — Reviewing first and last impressions.

Design patterns for promotions — Design patterns that could carry a promotion.

Patterns for bunches of products and vendors — Patterns for product bundles and vendor collections.

Borrowed from the App Store and Netflix

The App Store’s Today tab raised installs from browsing by showcasing the best apps every day as a feed. Netflix’s home screen and Previews drive discovery the same way. Both pointed at a dynamic, personal, content-led Promotion Center.

App Store Today tab — App Store, Today tab.

Share of App Store installs from browse, before and after iOS 11. Source: Sensor Tower.

Today tab features median weekly downloads — Median weekly downloads for apps featured in Today. Source: Sensor Tower.

Final solution

One center, five ways to save

The full Promotion Center flow. Click to zoom.

Food partyDiscount storesShopping bonusFoodProDiscount codes

Outcome

Longer sessions, better conversion

+62%

Pages per session

Longer sessions, not shorter

We aimed for less time in discovery. Sessions that included Promotion Center were longer instead: 11.67 pages, against 7.18 in sessions without it.

+5 pts

Sessions that reached a basket

People who used it converted better

Across three periods, sessions that used Promotion Center reached a basket about 5 points more often (36% against 31%) and reached checkout 0.5 to 2 points more often. It compares people who chose to use it with those who didn’t; it wasn’t an A/B test.

+20%

Total party orders

A new home for party orders

Party orders from the home page and Promotion Center together rose 20%, and Promotion Center’s share kept growing.

Pages per session: 7.18 in sessions without Promotion Center, 11.67 in sessions with it — Pages per session: 7.18 without Promotion Center, 11.67 with it.

Tables of basket and checkout rates for sessions that used Promotion Center and sessions that didn’t, over three periods — Basket and checkout rates, Promotion Center users against everyone else, over three periods.

Daily party orders from the home page and from Promotion Center, October 2023 to February 2024 — Daily party orders from the home page and Promotion Center, Oct 2023 – Feb 2024.

Challenges & lessons

What I’d tell myself before starting

Tight deadline

One cycle, our biggest initiative, little room for research. We had to make the right calls fast.

Innovating inside a mature product

Every new idea had to keep the existing experience fully intact while we ran experiments we could learn from.

Defining the MVP

I kept widening the scope. I learned to hold an MVP line and park my ideas as next steps in the design doc.

Personalized and dynamic content — Personalized, dynamic content

Flash sales, built on Snappfood’s Promotion Maker

Next case study

Balad Maps: Search & Discovery →

Approach

Four phases, in order

Here’s how I take a design task from a blank page to a shippable UI.

Understand the client’s requirements, the target audience and the market, through interviews, surveys and research. Here: requirement analysis and competitor analysis.

Define goals, objectives and design strategy with stakeholders, aligning the product vision and key features. Here: personas and problem statements.

Organize and structure the content for intuitive navigation, using wireframes and prototypes to set the hierarchy. Here: feature prioritization and information architecture.

Create on-brand high-fidelity mockups and interactive prototypes, then iterate on feedback and usability testing. Here: UI design and design system.

Phase 1 · Understand

Requirement analysis

Who are the target users?

Mobile job seekers at every level, from entry-level to executive. Usually looking for a new job, sometimes researching careers or companies.

What are their needs and pain points?

Finding jobs that fit their skills, and applying quickly. The biggest pain is time: hours on job boards and filling in online applications.

Which features are a must?

Search by location, title and other criteria
Job details: description, salary range, company info
Apply online or with a résumé
Save jobs for later
Track applications

What’s the user journey?

Create profileSearchView detailsApplyTrack status

What does the market look like?

Crowded, with Indeed, LinkedIn and Glassdoor holding large user bases. The trend is mobile-first, so the app has to work beautifully on a small screen.

Limited experience, competing with stronger candidates, and finding room to grow.

Phase 2 · Define

The problem, in 5W1H

Job platforms lack personalization, deep company insights, real networking and good application management, so job seekers struggle to find the right opportunities.

Job seekers, wading through huge numbers of listings with little company information, few connections and messy applications.

Across the job search app industry, where current platforms fail to offer a seamless, user-centered experience.

Right now. People feel frustrated and inefficient during their search because of these gaps.

Missing personalization, insights, networking and application tools lower satisfaction and make it harder to land the job they want.

An app with personal recommendations, deep company insights, strong networking and efficient application management that closes the gap between people and the jobs they want.

Problem statement board — The problem statement board in FigJam.

Phase 3 · Conceive

Red routes first

The red route technique ranks features by the critical paths users take most. It keeps the work user-centered, spends limited resources on what matters, validates the core early and gets a useful product to launch sooner.

Foundation and core features. The scope of this task.

Registration & profile

Name, contact details and experience, so the search can be personal.

Job listings

A clear list with filters for location, industry and experience level.

Advanced filters

Salary range, job type and company size to narrow results down.

Personas served: Sarah (profile and saved jobs), John (entry-level listings and filters) and Lisa (internships through search and filters).

Personalization and connection.

Networking

Profiles, connections, messaging, industry communities and forums.

Recommendations

AI suggestions from preferences, behavior and saved searches.

Company insights

Culture, employee reviews, benefits, values and growth.

Streamlining the application process.

Résumé builder

Custom templates for skills, experience and education.

Application tracking

Status, reminders for interviews and deadlines.

Interview prep

Tips, sample questions and best practices.

Phase 3 · Conceive

Information architecture

Focused on iteration 1’s critical paths: search, read a listing, apply, and keep track.

Home3

Personal recommendations
Quick search bar
Browse by category

Job details3

Title, company, location, salary, responsibilities
Apply in the app
Save for later

Search & filters3

Location, salary, type, experience
Sort by relevance or date
Clear preference input

Profile3

Preferences and contact info
Résumé upload
Saved jobs and history

Notifications2

New jobs and status updates
Custom frequency and types

Settings2

Account, privacy, notifications
Help, FAQs and support

Phase 4 · Visualize

Interface and UI kit

A palette that says professional, growing and energetic. Click a swatch to copy its hex.

UI designEvery screen of iteration 1Open in Figma ↗ Interactive prototypeTap through the core flowOpen prototype ↗ Design systemColors, buttons, tags, job cardsOpen UI kit ↗

Next case study

Work

Promotion Center

Search & Discovery

Graph Vault

Job Search Application

Workflow

Experience

AI Design Engineer

AI Product Design Manager

Product Design Manager

Senior Product Designer, Promotions

Senior Product Designer, Search & Discovery

Product Design Team Lead

Product Design Consultant

Senior Product Designer

Product Design Team Lead

Product Designer

Product Designer

Product Designer

Lead UI/UX Designer

UI/UX Designer

UI/UX Designer

UX Designer

UX Designer

UX Designer

UI Designer

Graphic Designer

Designer

B.Sc. in Software Engineering

Medium articles

Why EdTech AI is a behavioral problem, not a UI problem

When AI builds the screens, what do designers build?

Designed for delusion: when engagement-focused UX creates “AI psychosis”

How AI can understand learners, not just teach them

Say hello.