Product redesign · Muaban POS — the F&B point-of-sale from my first company

Project

One Order, Three Apps: Redesigning an F&B POS

platform

Mobile-only POS for small and mid-size F&B shops (iOS & Android phones). Three apps around one order: a service app (owner, server, cashier), a kitchen app and an admin app (owner, manager).

Responsibilities

Evaluate the shipped 2020 UI with a heuristic evaluation (Nielsen’s 10 heuristics + WCAG 2.2) and a cognitive walkthrough of the core task, map how one order moves between the three apps, then redesign every flagged issue across ordering, payment, kitchen, menu, staff and reports.

Feature

Ordering by dining table with clear options, send to kitchen with live dish status, three ways to pay with a printed VietQR bill, kitchen tickets with waiting time and sold-out alerts, and an admin that runs menu, staff and reports from a phone.

Overview

The 2020 product worked, but each app was designed on its own. Dining-table status was colour only, paying was one “Print & pay” button, the kitchen hid actions behind swipes and servers never knew a dish was sold out, and every admin module opened with an intro screen. I rebuilt it as one system around a single order lifecycle that all three roles read the same way.

Three apps on one order: service, kitchen and admin

1. Case Study Overview

a. Context & Evaluation

  • Three roles, three apps, one order
  • Heuristic evaluation: 14 screens, 18 issues
  • Cognitive walkthrough: 7 of 16 checks failed
  • Three principles for the redesign

c. Redesign by App

  • Service app: order → kitchen → pay
  • Kitchen app: tickets, late and sold-out
  • Admin: overview, menu, orders, staff, reports
  • Same states and words in all three apps

b. How the System Works

  • One order lifecycle shared by all apps
  • Data flow: service → kitchen → service → admin
  • Notification priorities P1 · P2 · P3
  • Roles and permissions

d. Review & Edge Cases

  • Before → after on five key moments
  • The same walkthrough on the new screens
  • Edge cases and system states

2. Context & Problem

Muaban POS is the product of my first company: a mobile POS for small restaurants and cafés, where the owner often takes orders, runs the cashier and checks the business from the same phone. The 2020 version shipped fast and did its job. In 2026 I went back to redesign it properly, as one system instead of three separate apps.

Brief vs real problem
How I worked: heuristic evaluation and cognitive walkthrough

Who uses it: three roles, three apps

Each role works in a different place with a different device, so each app has its own job, but all of them read and change the same order.

Three roles, three apps

Heuristic evaluation of the shipped UI (2020)

I audited 14 screens from the three apps against Nielsen’s heuristics and WCAG 2.2: 18 issues, 2 Critical and 7 High. The worst ones sit exactly where money and food change hands.

Old screens with issue markers
Heuristic evaluation issue table

Plus 9 Medium and Low issues: placeholder-only labels, repeated “Choose” buttons, 7+ colours without meaning, illustrations taking 30–40% of the screen, red as the admin’s primary colour.

Cognitive walkthrough of the core task

Task: open an empty dining table, order 4 dishes (one with options), send to kitchen, take cash. User: a server in their first week. Q1 knows what to do · Q2 sees the action · Q3 understands it leads to the goal · Q4 knows what just happened.

Cognitive walkthrough of the old UI

7 of 16 checks failed, mostly on Q3 (understanding the action) and Q4 (feedback after it). Each failure maps to a fix in section 4.

Three principles for the redesign

Three principles for the redesign

3. How the System Works

Before redrawing any screen I wrote down how one order travels through the system: who creates it, who changes it and who needs to know. Every screen in every app reads from this one model, which is what makes three apps feel like one product.

Three apps, one order

The service app creates and closes the order, the kitchen moves its dishes, and the admin sets the rules (menu, prices, roles) and reads the results.

System map: three apps, one order

One order lifecycle, read the same way everywhere

Five steps and one branch. The same words, colours and icons appear on the dining-table card, the kitchen ticket and the admin order list. In the old app the same order showed six different colours, and “Ordered” and “Cancelled” were both red.

Order lifecycle

Notification priorities between kitchen and service

Not every kitchen event deserves a banner. I ranked them by what happens to the guest if the server doesn’t know.

Notification priorities

Roles and permissions

Account types and module permissions were two separate lists in the old admin. I merged them into roles with clear rights; staff inherit the role, so nobody toggles switches person by person.

Roles and permissions

The data-flow board in the working file

How “Send to kitchen” reaches the kitchen and how each dish status comes back to the server in real time, drawn with the real screens.

Data flow: send to kitchen, kitchen app, back to service

4. Service App (owner, server, cashier)

Built for one hand and a busy floor. The service app follows an order from the dining table to the payment, and each step answers a failure from the walkthrough in section 2. 19 screens in the working file; the key ones are below.

Order for a dining table

Home splits Dine-in and Takeaway. Takeaway orders use the same card as dining tables, so there is one pattern to learn.

Service app: home, order, options, table in service
Ordering decisions traced to the evaluation

Get paid without mistakes

The old “Print bill & pay” button mixed two actions and never confirmed anything. Now tapping “Pay” always prints the bill with a VietQR code, then the cashier closes it with the method the guest actually used.

Pay by cash, transfer and card
Three ways to pay
Payment decisions traced to the evaluation

Move, merge and know your guests

Moving and merging dining tables start from the table itself, not from a separate menu. Customers are sorted by tier so the owner can run offers by level.

Move table, merge tables, customers by tier

5. Kitchen App

A wall-mounted device, read from a metre away by people with wet hands. Fewer and bigger targets, every action is a labelled button, and the waiting time is always on screen.

Tickets you can read from a metre away

Kitchen app: tickets, ticket detail, sold out, by dish
Kitchen decisions traced to the evaluation

Kitchen ↔ service: late and sold-out

The two moments that used to need shouting across the room now have a defined loop, so servers can answer the guest with real information.

Late reminder, ETA, sold out while ordering, notifications
Late loop between kitchen and service

6. Admin mobile · Owner and manager

The old admin opened every module with an intro screen. Now the owner lands on the numbers, and five tabs cover the daily jobs: Overview · Orders · Menu · Customers · More. Settings and staff live under More as detail pages with a back button.

Know the day in 10 seconds

Revenue and profit sit on the home screen. Details are one tap away, with periods that match how an owner thinks: this month vs last month, then 1, 3, 6 months or a year.

Admin: overview, report, date range, seasonal dishes

Run the whole menu from the phone

The old “Product management” was an intro screen in front of four separate lists (categories, products, units, sold-out). Now it is one tab with three segments, and one form for adding and editing a dish.

Admin: menu tab, add a dish, option group, dishes turned off
Menu decisions traced to the evaluation

Orders, staff and permissions

Admin sees every order including cancelled ones, and staff rights follow roles instead of per-person switches.

Admin: orders, cancel, staff profile, role permissions
Admin decisions traced to the evaluation

7. Before → After

Five moments where the old UI failed most, side by side with the redesign. Each pair keeps the old screen unchanged and notes what changed and which audit issue it fixes.

Dining-table map

C1 · H1: status from colour only → text + colour + icon, readable in grayscale

Before and after: dining-table map

Dish options

H2: radio for multi-select → required / optional, “up to 2”, live price

Before and after: dish options

Payment

C2: one “Print & pay” button → print on Pay, choose the method, success screen

Before and after: payment

Kitchen

H3 · H4: swipe actions and no waiting time → labelled buttons, minutes on every ticket, sold-out alerts

Before and after: kitchen

Permissions

H5: tags with an X → roles with grouped switches and “Admin only” locks

Before and after: permissions

The same walkthrough on the new screens

I repeated the cognitive walkthrough from section 2 on the redesigned screens: same task, same first-week server, same four questions. This is my own inspection, not a test with users.

The same walkthrough on the new screens

Old UI: 9 of 16 checks passed. Redesign: 16 of 16 in my walkthrough. It still needs to be confirmed with real staff before shipping.

8. Edge Cases & System States

A POS is used on unstable Wi-Fi, at peak hours, by people who get interrupted. I designed the unhappy paths before polishing the happy one.

Edge cases

9. Lesson learned

1. How an F&B app actually flows

One order touches three roles. The service app opens and closes it, the kitchen moves its dishes, and the admin sets the rules (menu, prices, roles) and reads the results. Most of the old problems came from designing each app on its own. Mapping the flow across service, kitchen and admin first told me where every screen belongs and what it has to hand over to the next role.

2. Analyse the states before drawing screens

Writing down the order lifecycle (Ordered → Cooking → Served → Paying → Completed, plus Cancelled) and the dish states (Draft, Sold out) decided what each app shows, who can change what, and which events deserve a notification. Edge cases such as a dish sold out after sending, or a transfer that hasn’t arrived, came straight out of that state model.

3. Two inspection methods, two different lenses

Heuristic evaluation showed what was wrong on each screen; the cognitive walkthrough showed where a first-week server gets stuck across the whole task. Together they gave every change a reason I can trace back. They are inspections, not user tests, so watching real staff use it is the honest next step.

077.804.9797