A Singapore phone retailer: website and back office, rebuilt.
Novatise rebuilt an independent Singapore phone retailer's online shop and gave the owner a back office he runs himself: live prices and stock, instant trade-in quotes for 501 phone models, and an AI assistant that answers from the shop's own data.
The client
The client is an independent phone shop in Singapore with a single counter in the city centre. It sells new and used phones, buys phones back for cash or as a trade-in against a new one, and repairs screens and batteries while the customer waits.
Most sales happen at the counter or over WhatsApp. Customers browse online, then message the shop or walk in.
The challenge
The shop had outgrown its website. It ran on an ageing WordPress theme, and the questions customers asked most still had to be answered by hand.
- What is my phone worth? Every trade-in quote came over WhatsApp or the phone, one customer at a time.
- Do you have it, and at what price? Thousands of listings, but finding a model, its storage options and its condition took several clicks.
- Prices changed daily. Updating them meant logging into WordPress and editing products one by one.
- Mobile came second. Most visitors arrive on a phone, and the old pages were built for a desktop.
The owner wanted a site that answers these questions on its own, and one place where he can change prices, stock and trade-in values himself. The owner changes something once in the back office; the storefront and the assistant both read from it, so prices, stock and trade-in values never disagree between them.
The storefront
We rebuilt the shop's website around the way people actually buy a phone: on a phone, comparing a few models, then asking a question.
- 4,019 product pages, every one a real link. Products show as cards on the home page, brand pages and the shop, from one shared design so they never drift apart.
- One page per model, with every option on it. Storage, condition and colour are buttons on the product page. Each still keeps its own address, so no existing link or search ranking was lost.
- Specifications people can read. Long supplier descriptions are tidied into a clean spec table, and the full write-up sits behind a "read more".
- A FAQ on every product. Warranty, local or export set, collection, payment, trade-in and how to order, worded for that product's condition and stock.
- Order the way the shop works. Every product has a one-tap WhatsApp enquiry with the model already typed in, plus reserve online and pay on collection.
- It installs like an app. On a phone, the site can be added to the home screen and opens full screen, like a native app.
Trade-in and repairs
The question the counter heard most, "what is my phone worth?", now gets an answer online in three steps.
- Pick the brand, or search for the model and jump straight to it.
- Pick the model from a photo grid, then New or Used.
- Answer a few questions about storage, screen, body, battery health, accessories and faults. The estimate updates with each answer.
Every one of the shop's 501 buy-back models has its own page and its own address, so each one can be found on Google. The page shows two figures side by side: trading in against a new phone, or selling for cash, which pays a fixed amount less. Before a trade-in is booked, the customer ticks that they have read the shop's terms, and a booked trade-in can be added to the cart alongside a new phone.
Repairs work the same way. The shop's repair price list is published on the site from the same back office, so customers can check a screen or battery price before they walk in.
The shop assistant
A friendly robot mascot sits in the corner of every page. Tap it and it answers questions in plain language: "Do you have the iPhone 17 Pro in stock?", "How much to fix a cracked screen?", "What would you give me for my old Galaxy?"
It is built so the shop can trust it with customers:
- It answers from the shop's own data, not from what an AI model remembers about phones. Each question pulls the matching rows from the live catalogue, the repair price list and the buy-back values, and the assistant answers from those alone.
- It is read-only by design. It can look things up, but it cannot change a price, touch stock, edit a product or place an order. Automated tests check this on every release.
- It keeps no conversations. A chat lives only in the visitor's browser tab and disappears when it closes. Nothing typed is stored or used for training, and the chat window says so.
- The owner shapes it. The back office has a box for its instructions, a test button, and a one-click refresh that re-reads the website.
The back office
We built the shop its own admin: one login and one place to run the online side of the business, with no WordPress training needed. The owner changes something, presses one button, and the site updates.
| Screen | What the owner does there |
|---|---|
| Dashboard | Sees what needs attention: low stock, edits not yet live, posts to approve |
| Stock & price | Edits prices and stock inline, then pushes them live in one click |
| Variants | Prices storage, condition, colour and bundles as differences from a base price |
| Trade-in | Sets buy-back values per model, what each answer takes off the price, the question wording, and the terms customers must accept |
| Repairs | Keeps the repair price list the site publishes |
| Orders | Reads orders alongside the customer's WhatsApp messages |
| Assistant | Sets the assistant's name, greeting and instructions, and tests it |
| Activity | Sees every edit, push and publish, with who made it and when |
The admin stands on its own, with its own database and login. It can be handed over to the shop as it is.
Content and video
The shop already had a library of buying guides and plan comparisons. We carried 200 of those articles over from the old site and connected the back office to Novatise's content engine through a single sign-on. The owner logs in once and moves between his shop admin and content tools without a second password.
We also built a video studio that turns a blog post into short vertical reels for Instagram and Facebook:
- The script comes from the article. An AI writer drafts a 35–40 second voiceover from the post and plans the scenes.
- The numbers become cards. Data caps, contract length and prices appear as on-screen cards, so viewers see the figures as well as hear them, alongside the shop's own product photos.
- A voice for each market. The voiceover is generated, with captions timed word by word.
The reel is checked against the article's own text before rendering, so it never quotes a figure the post doesn't state.
Built to be safe
A shop's website handles customer details and the owner's prices, so we treated safety as part of the build, not an add-on.
- Privacy by default, in line with Singapore's PDPA. Only essential cookies, comments switched off so no commenter data is collected, and an assistant that keeps no chat history.
- Hardened WordPress. Two-factor login for admins, a malware scanner, protection against account-guessing on password resets, and managed security updates.
- Keys kept out of reach. The AI provider key is stored encrypted, and the back office can't be pointed at the live shop by mistake: that guard is in the code, not a setting.
- Safe releases. Every back-office update backs up its databases first, checks the new version is healthy, and rolls itself back if it isn't.
- Every change is logged. Each edit, push and publish is recorded with who made it and when.
Results
The owner now updates prices, stock and trade-in values himself, and customers get a trade-in figure without waiting for a reply. Launch figures will be added once the new site has been live long enough to measure.
Technology
| Part | Built with |
|---|---|
| Storefront | WordPress and WooCommerce with a custom theme, installable as a progressive web app |
| Back office | Python (FastAPI), its own database and login |
| Content engine | Next.js and TypeScript, single sign-on from the back office |
| Shop assistant | A large language model given only the shop's own catalogue, repair and buy-back data |
| Video studio | Python, AI script and scene planning, generated voiceover and captions |
| Security | Cyber Securitise: two-factor login, malware scanning, managed updates |
| Hosting | Docker containers, with nightly off-site backups; back-office releases roll back automatically if unhealthy |
The client's name is withheld.