Case study · 2020 – 2023 · Timișoara

Food ordering where every order pays back

Istigos let people order from local restaurants for delivery, pick-up or dine-in. Each order earned discount and cashback, and part of every restaurant's commission went to the friends who invited the customer. As the sole developer I designed and built the whole system: two store-published apps, the back-office and the services behind them.

React Native · ExpoStrapi · GraphQLNode.jsMongoDBDocker
Customer app home with discount and cashback balances
Customer app restaurant menu
Checkout with discount and cashback sliders

Real screens from the demo stack: wallet, menu, and the discount and cashback sliders at checkout

The idea · #cashbackpebune

A local food app where the commission flows back to people

Delivery platforms take a commission on every order and keep the relationship with the customer. Istigos started in Timișoara with a different deal: part of what restaurants pay goes back to the people who order and to the friends who brought them. The slogan on the flyers was "the new Timișoara app where you get real cashback".

Customers

Save on every order

A discount wallet that lowers the bill at partner restaurants, plus cashback you can spend in the app or withdraw to your bank account. No vouchers.

Friends who invite

Earn for life

Invite friends and you get 10% of the Istigos commission on every order they place, for as long as they order. Their friends bring you another 2%. The more people you invite, the more you earn.

Restaurants

Grow their own regulars

Each partner handed out flyers with its own referral code: new customers got a 5–10% discount there, and the restaurant became part of their network. Restaurants pay a commission per order, with a different rate per channel.

Dine-in was part of the same system. A guest at the table shows a personal QR code, the waiter scans it, and the meal earns discount and cashback like an online order, so walk-in guests could join the network too.

The system at a glance

One product, four moving parts

Customers and restaurants each had their own app. A headless CMS was both the API and the back-office. A small Node.js service handled identity and the emails every order sends.

2,000+app downloads
1,500+orders placed
7partner restaurants in Timișoara
2apps on App Store & Google Play
3sales channels: delivery, pick-up, dine-in
~35klines of application code
Follow one order

From a craving to a confirmed delivery time

Real screens from the running system, in order. Ana orders two pizzas. The restaurant confirms with a delivery time. Ana is emailed and accepts the time in the app.

Scroll sideways →

  1. Customer home screen with discount and cashback balances
    Customer

    Wallet first

    The home screen opens with the two balances: discount (339.88 lei) and cashback (245.56 lei).

  2. Restaurant menu
    Customer

    Menu by category

    Each product has delivery, pick-up and dine-in prices, portions, extras and allergens.

  3. Shopping cart
    Customer

    Cart

    The cart is a snapshot. A later menu change never rewrites the order.

  4. Checkout with totals
    Customer

    Spend the wallet

    Two sliders: discount first, capped by the restaurant (7% here), then cashback. 83 lei becomes 30.88 lei.

  5. Choose payment method
    Customer

    Cash or card

    Card payments go through the bank's 3-D Secure page.

  6. Order waiting for confirmation
    Customer

    Sent Nou

    The order waits for the restaurant.

  7. Restaurant live order queue
    Restaurant

    Live queue

    The tablet app polls the queue and plays an alert sound for new orders.

  8. Order detail in restaurant app
    Restaurant

    Order detail

    Items, notes, discount and cashback used, amount to collect, address.

  9. Confirm with delivery time
    Restaurant

    Confirm with ETA NeConfirmata

    Confirming means promising a time. 30 minutes is picked here.

  10. Order confirmed email
    Email service

    Customer is emailed

    The auth-mail-api sends the branded confirmation email with the order summary.

  11. Customer accepts the delivery time
    Customer

    Accept the time Confirmata

    Ana sees "ready in 30 min" and accepts. Only then does the kitchen start.

Architecture

Headless CMS as both API and back-office

Strapi gave the operations team an admin UI and gave the apps REST and GraphQL from the same schema. Identity and email lived in a separate Express service, so the CMS could be upgraded on its own.

Configuration as code

Roles and the permission matrix are applied at boot by bootstrap.js, so a new database behaves like production. All URLs, keys and secrets come from environment variables.

Business rules as data

Bonus amounts, premium tiers, commission rates and discount caps live in the CMS. The business can change them without releasing a new app version.

OTA delivery

Store builds came from EAS. JavaScript fixes reached every phone in minutes through expo-updates, without store review.

Order lifecycle

A state machine shared by three actors

The customer must accept the restaurant's proposed time before cooking starts. Nobody waits for an order without knowing when it will arrive, and a refusal costs the kitchen nothing.

Noucustomer sent→ NeConfirmatarestaurant proposed ETA→ Confirmatacustomer accepted→ InLivrarewith the driver→ Livrataledger + commission written
AnulataDeClientrefused the ETA AnulataDeRestaurantrefused, or an incident · wallet refunded, reason stored, email sent
Loyalty & referrals

Where the money goes on a 100-lei order

Restaurants pay a commission that depends on the channel. Part of it becomes cashback for the customer's referral chain, up to six levels deep. Every movement is written to a ledger, so any balance can be explained to a customer.

Delivery order, 15% commissionlei
Order total100.00
Platform commission (15%)15.00
→ Level 1 referrer (10% of commission)+1.50
→ Levels 2 – 6 (2% each)+0.30 × 5
→ Whoever brought in the restaurant+ configurable %
Discount the customer may spend here≤ 10.00

Discount caps reward discovery

Each restaurant sets four caps: General, and lower ones if the customer ordered there in the last 1, 3 or 7 days. New customers get the best deal. Regulars still get a deal, and the restaurant keeps its margin.

Three delivery zones

Each zone has its own minimum order, fixed fee and per-km fee. The app computes the zone from the saved address and hides restaurants that are too far away.

Four customer screens: wallet balances, checkout with discount and cashback sliders, cashback statistics, discount history
The wallet in the app: balances, the two checkout sliders, and where the cashback came from (38.60 lei from friends, 7.42 lei from friends of friends, 20.45 lei from a restaurant Ana brought in).
Back-office & email

Operations ran from the CMS

Onboarding a restaurant (menus, photos, hours, fees) happened in the admin UI, without code changes. The GraphQL schema is generated from the same content types.

Restaurant app

Built for a busy counter

Big colour-coded tiles per state, a sound for every new order, a countdown to the promised time, and reports that match the monthly commission statement.

  1. Restaurant login

    Staff login

    A role decides the home screen: manager, waiter or driver.

  2. New orders

    New orders

    Channel, payment method, amount and the "seen by" marker.

  3. Order history by state

    Day history

    Every order of the day, grouped by state.

  4. Reports menu

    Reports

    Per driver, per period, per channel.

  5. Period report

    Period report

    Cash, card, discount, cashback and the platform commission.

In the stores

As customers saw it

The original App Store and Google Play artwork, made by a designer colleague.

Store preview: discounts delivered to your door Store preview 2 Store preview: bigger discounts with every order Store preview 4
Retrospective

What I would build differently today

The product was built fast, by one developer, to prove a business model. Many of these trade-offs were deliberate. This is the plan I would follow now.

ThenRiskToday
Cashback and commissions computed in the appsNot atomic. Clients are trusted with money movementsA backend service owns state transitions and writes order, ledger and commission in one transaction
Public role can write customer recordsAnyone who knows the API can change another customer's dataOne identity provider. The API verifies the JWT and enforces ownership per row
GraphQL built by string interpolationInjection, fragile queriesVariables and typed operations generated with graphql-codegen
Two services share one collectionSilent schema driftOne owner per collection. The other service reads an API or receives events
Card payments started in the appMerchant credentials in the client bundleThe server creates the payment session. A webhook confirms payment
Polling every 10–20 sLatency and wasted loadWebSocket or SSE push, with polling as fallback
Strapi 3, Node 14, Expo 47End of lifeStrapi 5 or NestJS + PostgreSQL. Latest Expo with Expo Router. TypeScript, tests and CI
Try it

The whole system runs locally

The repository has a Docker Compose stack (MongoDB, Strapi, the auth and mail service, and a Mailpit inbox) and a seed script. The seed creates three fictional restaurants, staff accounts, customers with a referral tree, and 40+ orders in every state, all through the public APIs.

ComponentStackPath in the repo
Customer appExpo SDK 47, React Native 0.70, MobX State Treeapps/customer-app
Restaurant appExpo SDK 47, Redux, react-native-paperapps/restaurant-app
CMS / APIStrapi 3.6, GraphQL, MongoDBservices/cms-api
Auth & emailExpress, JWT, Nodemailerservices/auth-mail-api
Demo stackDocker Compose, seed scripts, Mailpitinfra/