Lunara
An astrology app for iOS and Android: natal chart, compatibility, horoscopes, and forecasts in 30 languages. I improved, scaled, and grew a product that was already live.
Role
Product designerthe only designer on the productTraffic
160,000+new users entering the funnel every monthResult
+76% purchaseswith traffic almost doubled: autumn 2026 vs late 2025See the result ↓02 · Context
Lunara is an astrology app for iOS and Android: natal chart analysis, compatibility, horoscopes, and forecasts in 30 languages. I joined a product that already had a funnel, paywalls, and the core app, to improve, scale, and grow it.
Three design systems in the product
The appPro and Lite, each with light and dark themesWeb funnelGeneral One, the mobile funnel system03 · My role and team
I was responsible for the funnels, the app, the paywalls, and the web portal.
04 · Problems
The first screen used a heavy raster image. On a weak connection it loaded badly or not at all, so the first step had the biggest drop-off.
Solved in: 07 · First screen · 09 · Light funnel
Funnel screens had no event tracking, so optimisation was guesswork.
Solved in: 13 · Metrics
Cancelling was a single “Unsubscribe” button with no retention mechanics or offers.
Outcome: 14 · Result: less churn
How many people leave at each funnel step
March–April 2026, all languages. Two drop-off points: the first screen and the time of birth.
05 · Research
I built an autonomous system of agents on Claude AI that monitors competitor funnels, landing pages, and reviews every day. It spotted two competitor products early, Soulmate Sketch and Palm Reading, and I used that to build our own funnels in the same directions ahead of the trend.
06 · Geo segmentation
The same A/B tests showed completely opposite results in different regions. A winner in one country lost in another, so every test is now segmented by geo before any decision.
Based on Lunara’s A/B tests; exact numbers are under NDA. That’s why the paywall variant is chosen per country, not one for everyone.
07 · First screen
Visitors who came from an ad see a unique first screen of the funnel, built around the exact topic of the ad they clicked. The promise in the ad and the first thing they see match.
08 · AI decisions
I collect all A/B test data and hand it to AI. It analyses the results and makes one of three calls: roll out, don’t roll out, or collect more data.
Test queue
How the mechanism works: this is how AI sorts tests into three decisions. The tests in the queue are illustrative, the logic is simplified.
09 · Performance
On Android people buy 1.9 times less often than on iOS, and many of them have old phones. So I made the funnel lighter in two ways: vector illustrations instead of raster, and CSS animations instead of JS.
September–October 2026, purchase conversion by platform.
Lightweight funnel: first-screen illustrations, 9 topics
The vector is 99% lighter than the original raster and half the weight of even a well-compressed WebP. It is also sharp on any screen and can be animated with the same CSS.
Total weight of the 9 illustrations from Figma. WebP: sized for a 2x screen. SVG: gzip-compressed, as the server delivers it.
10 · Key features
A low-effort onboarding asks only for birth data. Results come as simple personal insights rather than complex charts.
Answers “are we a match?” in plain language instead of heavy astrology terms. The main reason people install the app.
Built around topics: love, career, personal growth. Content feels personal, not generic.
An emotional wow moment with a visual result people want to open and share.
A dedicated flow for one of the most common questions: “can we get back together?”
From curiosity to purchase: intro, data, first insight, deeper hooks, then the full report offer.
Screens from Figma: the web funnel (natal chart, compatibility, portrait, ex, email before the paywall) and the app (questions on “Love and marriage”).
11 · UX decisions
In the app the entry points read “When will I meet love?”, not “Synastry”.
A short insight comes first; the full report opens only if the person wants more.
Only three fields about birth; everything else is asked later, when it is needed.
The partner portrait and the “ex” flow are the two screens people share most.
By the paywall the person has received three things and given only birth data.
The store screenshots are real screens, so the first session matches the promise.
12 · Design system
The product has three design systems: one for the app (Pro and Lite), one for the mobile web funnel (General One), and one for the web portal. The web portal system is under NDA, so the first two are shown here.
Pro (gold) and Lite (blue), each in light and dark. One set of 63 semantic tokens covers all four themes: the theme changes the colours, while components, spacing, and radii stay the same.
surface/chatsurface/accentborder/bubblesurface/pilltext/accent-aacta/gradientPlanets: one colour each, in every theme
Radius
1215100Spacing
48162432Touch target
48General One is the system behind the mobile web funnel. It is built on atomic design: foundations, atoms, molecules, organisms. 45 semantic colour tokens and 15 number tokens. Every state is specified, down to the 97% press scale.
Chips: a mutually exclusive pair
Input: default, focus, error
Toggle, checkbox, radio
Badges
bg/main#fefbffbutton/primary/bg#d4c2ffborder/accent#b392ffbg/selector-pressed#e5d9fftext/accent-strong#8a57fftext/primary#1d1b1bprogress/inactive#e4ecf7border/error#ff3737Radius
8101624100Spacing
4812162432Stroke and tap target
124413 · Metrics
14 · Result
Traffic almost doubled: the ad-matched first screens and geo tests brought far more people into the funnel.
Purchases grew 76%: more people reached the paywall, and the paywall itself was reworked.
Conversion held within 7% while traffic almost doubled: many more people, and nearly the same share buys.
Before I joined (late 2025) → September–October 2026, web funnel. Relative changes only.
GoTov.org brings religious, cultural, and social services together on one digital platform.
MoneyDay is a mobile app for applying for short-term loans in just a few minutes.
A volunteer app where three-step booking and a two-way trust system replaced manual visit coordination.
