The starting point

A field sales app already existed, but it came from a company my employer had acquired: an app built for one customer only. To turn it into a product it had to be rebuilt from the ground up, for three reasons:

  • to update the tech stack;
  • to make it work offline, because a sales rep can’t stop where the signal ends;
  • to run it on Android and iOS tablets, because every customer has their own devices.

So there was nothing to improve: we had to decide what the app should be, from navigation to components.

Constraints

  • Two native apps, built separately: one developer for Android and one for iOS.
  • Tablets in portrait and landscape: every screen has to work in both orientations, 8-inch tablets included.
  • Offline: the app works on a local copy of the data and syncs when the connection is back.
  • One product for many customers: multiple organisations, six languages, configurable colours, features that only appear when the user has the permissions and the data.
  • The existing back-office APIs, which the app had to use as they were.

The app’s architecture

I set the app up as a single product with a repeatable structure:

  • a sidebar on the left, as in the back office, with items that depend on the user’s permissions;
  • a dashboard as the landing page, with targets, configurable quick actions, statistics, upcoming appointments and recent items;
  • the customer record at the centre of the app: from there the rep sees details, contacts, tasks, appointments, documents, payments and statistics, and almost every action starts there;
  • the calendar and tasks, orders, surveys and sync areas.
Wireframe of the app on a tablet in landscape, with made-up content: the sidebar, the customer list and the customer record side by side, the offline banner at the top. The row border shows the customer type.

The app, step by step

  1. A sidebar on the left, as in the back office, with items that depend on the user’s permissions.

  2. List and detail adapt to the screen. In landscape the customer record opens next to the list, to make comparing customers faster.

  3. The customer record is the heart of the app: details, contacts, tasks, appointments, documents, payments, statistics and much more.

  4. When the app is offline, a banner appears at the top of the screen showing when it last synced.

  5. The AI assistant will let reps act straight from the chat, from creating an order to summarising where a customer stands.

Key decisions

1. Native components where they matter, hybrid ones for the rest

Decision
Where users expect their operating system’s behaviour, components are native and follow the Human Interface Guidelines on iOS and Material on Android. Everything else is hybrid: a single component designed to work well on both systems.
Alternative rejected
Identical components on both platforms. A developer asked for it, to cut the work.
Why
A sales rep uses the tablet for everything else too. An app that behaves like its operating system is quicker to learn, and with one developer per platform native components cost less than redesigned ones, while hybrid ones are designed only once.

2. List and detail that adapt to the screen

Decision
The main lists, such as customers and tasks, have three states: list only, list and detail side by side, detail only in full screen. In landscape the list stays alongside; in portrait the detail opens on top.
Why
A tablet is used in both orientations, and the rep needs to move from one customer to the next without losing their place. Detail views adapt too: the button reads “Back” when the detail is full screen and “Close” when it sits next to the list, because it depends on the layout the user has in front of them.

3. Context fills in the fields for the rep

Decision
When an action starts from a customer or a task, the app fills in and locks the fields it already knows. A survey created from the customer record doesn’t ask for the customer; a long press on a task opens an appointment that is already filled in.
Why
Reps work between one visit and the next. Every field they don’t have to fill in saves time and removes a chance for error.

4. Offline without surprises

Decision
A banner at the top of every section says the app is offline and when it last synced. With the product owner we decided that no data can be created or edited while a sync is running.
Why
Reps always need to know whether they are looking at up-to-date data. Blocking edits during sync prevents conflicts instead of having to resolve them later.

5. The customer’s credit always in view

Decision
Remaining credit shows as a green, orange or red progress bar, both in the customer record and while building an order. Credit is checked before an order starts and again when it is released.
Alternative rejected
Two numeric indicators, “overdue credit” and “credit due”.
Why
With numbers, the minus sign was read the wrong way round, as “how far from the limit” instead of “how far over it”. The bar lets reps size the order, and there’s no point making them fill in an order that can’t be released.

The native design system

I created the app’s design system from scratch, before Nile, specifically for the native app. I designed every component myself and kept them aligned with the feature specs.

  • One tablet layout, native system components. Screens are designed once; what users expect from their operating system, such as date pickers, number keyboards and switches, stays native and follows the Human Interface Guidelines on iOS and Material on Android.
  • Hybrid components for everything else: sidebar, lists, customer record, order stepper, delivery status, tags, messages and confirmation dialogs are a single component, designed to work well on both iOS and Android.
  • 64 components and more than 400 variants in the tablet library. The button alone has 168, because it combines importance, meaning (create, confirm, delete, warning), state and size.
  • Colours from tokens shared with the back office, named by role, importance, state and property, plus a few mobile-specific exceptions.
  • Written rules: documentation sheets on one template (anatomy, properties, layout and spacing), rules for dates, times and feedback screens, and fixed tablet layout measurements, from the sidebar to the order summary bar.
Shared tokens Same as the back office Colours by role, importance and state; typography; mobile exceptions
Native components Different per platform Date and time pickers, keyboards, switches: Human Interface Guidelines on iOS, Material on Android
Hybrid components One for both iOS and Android Sidebar, lists, customer record, order stepper, tags, dialogs
How the app’s design system is organised: the same tokens feed the native components, which change with the platform, and the hybrid components, one for both systems.

Ordering, the most complex flow

An order goes through four steps that also work as tabs. A fixed bar at the bottom sums up lines, free items, discounts and remaining credit, without having to reach the summary.

  1. HeaderOrder details, with the first group already open to invite the rep to start
  2. CatalogueOnly the items that customer can order, with their purchase history
  3. LinesBuilt for review, without a side panel
  4. SummaryWarnings, the customer’s signature and release
The four steps of an order. Credit is checked before the header and again at release.

Promotions, blocks and approvals aren’t a separate module: they show up the same way in four places in the flow, with the same wording, so reps understand why an order is blocked wherever they are.

The order flow already has room for AI-assisted ordering, which will become standard in this app: the search field in Lines is deliberately shorter to make room for the import, and items added by the AI get their own badge.

Testing with users

In October 2026 I prepared and ran the user test on promotions: two moderated sessions, with a tablet prototype I built with AI and a customer’s IT staff playing the sales rep. Of five tasks, we assessed four: one was invalidated by a prototype issue.

  • First click right in 4 tasks out of 4: the user always knew where to start.
  • 2 tasks out of 4 completed unaided, the other 2 with help. Only 1 was finished within the target time.
  • The problem wasn’t finding promotions but understanding them: the badge didn’t tell active promotions from applicable ones, and there was no way to find out which promotion gave away a given item.

Each problem became a proposed fix for the product.

Results

  • Live for two customers, on Android since May 2026 and on iOS since September 2026, with more customers interested.
  • One app for every customer, configurable, instead of a custom-built one.
  • Ready for AI: it’s the app where AI-assisted ordering will become standard.
  • Towards Nile: the app’s design system will gradually move to Nile.

What I learned

Who you test with matters as much as what you test. IT staff know the promotion rules well, but they don’t work under the pressure of a rep in front of a customer: the test shows whether the interface makes sense, not whether it holds up in the field. I want to run the next tests with sales reps: this time the customer’s constraints made it impossible.