I Kept Checking Prices by Hand, So I Built a Bot That Runs Two Price Sites
By Jon Ng, Founder, Novatise · 7 min read · 30 October 2026 · Updated 10 November 2026

A price-comparison bot reads the prices shops publish on their own websites, matches each one to the right product, copies it exactly, checks it makes sense and publishes it with a timestamp. Fetching pages is the easy part; most of the work is trusting the number, which means leaving out anything uncertain. phoneprices.sg and goldprices.sg are built this way. The same pattern suits any business that checks prices, stock or rates by hand every day.
In Singapore, if you want to sell an old phone or buy gold, the honest answer to "where's the best price?" is usually "it depends on the day and the shop."
The problem: prices everywhere, no single view
Shops publish their own price lists on their own websites. Phone shops list trade-in and cash buyback prices for each model and storage size. Jewellers and banks list gold rates per gram. The prices change often, and every shop formats them differently.
So I did what most people do. I opened the shop websites one by one, every morning, and compared them by hand.
It was slow, easy to get wrong, and the moment I finished, some of the prices were already out of date. It was exactly the kind of repetitive job a bot should do.
The fix: a bot that reads the prices for me
I built a bot that does the checking, and it now runs two public sites:
- phoneprices.sg compares phone trade-in, cash buyback and new phone prices from Singapore shops. As of 30 October 2026 it covered 503 phone models from 10 sources, updated daily.
- goldprices.sg compares 916 and 999 gold rates per gram from Singapore jewellers and a bank. As of 30 October 2026 it tracked 8 sources and checked them several times a day.
How the bot works, at a high level
The idea is simple. The care is in the details.
1. Read only what shops publish
The bot visits the price lists each shop publishes on its own website. It doesn't guess and it doesn't use private data. If a shop doesn't publish a price, it isn't on the site.
2. Match every price to the right product
A phone shop might write "iPhone 15 Pro 256GB" while another writes it a slightly different way. The bot matches each listing to one model and storage size, so you are comparing like with like. For gold, it keeps 916 and 999 separate, and keeps selling prices, cash buyback and trade-in rates apart so they are never mixed.
3. Copy prices exactly, never estimate
Prices are copied as shown. Rows that say "WhatsApp us", "Pending" or "Coming soon", or that the bot can't read reliably, are left out rather than filled in. Where a shop only states a flat rule, like a fixed deduction if you don't trade in, the price is worked out from that rule and labelled clearly.
4. Check that the numbers make sense
For gold, every value is checked against the international spot price. If a number looks impossible, it is dropped, not guessed. Each shop's own timestamp is kept where it gives one, so you can see when a shop last updated, and stale rates are flagged.
5. Publish with a clear timestamp
Each update goes live with a "last updated" time, plus a plain explanation of where the prices come from, so visitors know how fresh and how reliable the numbers are.
What I learned building it
- The hard part isn't fetching the page. It's trusting the number. Most of the work went into matching products correctly and refusing to show prices the bot isn't sure about.
- Leaving data out is better than showing wrong data. A missing row is annoying. A wrong price sends someone to the wrong shop.
- Timestamps build trust. Showing each shop's own update time tells visitors more than any claim of accuracy.
- Remind people to confirm. Buyback offers are made after the shop inspects your phone or tests your gold, so both sites tell visitors to confirm with the shop before they go.
If you want to build a price bot: the steps
- List the sources and check each one publishes the data openly on its own site.
- Read each source's terms and robots.txt, and keep request rates low so you never burden their site.
- Define your product list: one master name per product, and the variants that matter (storage size, purity, condition).
- Write a reader per source that pulls only the price table, not the whole page.
- Match listings to your master list, and log every listing that can't be matched for a person to review.
- Validate: types (selling vs buyback), plausible ranges, timestamps and duplicates.
- Publish with a last-updated time and a plain explanation of the method.
- Monitor: alert when a source stops returning data or its layout changes.
Data quality rules worth copying
| Rule | Why it matters |
|---|---|
| Copy prices as shown, never estimate | A guessed price is worse than no price |
| Skip “call us” or unreadable rows | Filling gaps creates numbers the shop never published |
| Keep price types separate | Buyback, trade-in and selling prices aren't comparable |
| Check against a reference | Impossible values get caught before they're shown |
| Keep the source's own timestamp | Visitors can see how fresh each price is |
| Flag stale data | A price that hasn't changed in a long time may no longer be valid |
Keeping the bot running
Shops redesign their websites with no warning. When that happens a reader may return nothing, or worse, the wrong column. A few safeguards:
- Alert when a source returns far fewer rows than usual.
- Alert when prices from one source jump or drop sharply in one update.
- Keep yesterday's data so a broken update can be rolled back.
- Review unmatched listings weekly, new models and new formats show up there first.
Other jobs the same pattern fits
- Watching competitors' published prices for your own products.
- Checking supplier stock or lead times each morning.
- Collecting exchange rates, commodity prices or published fees into one sheet.
- Building a daily internal report from several systems that don't talk to each other.
The habits are the same, every time: read from the source, check the numbers, and show clearly where the data came from. For how we connect steps like these, see business process automation with n8n, Claude and Supabase.
Why this matters for your business
Almost every business has a version of my morning price check: a report someone builds by hand, data copied between systems, or a website someone has to update every day. At Novatise we build these kinds of bots and automations for clients, using the same habits: read from the source, check the numbers, and show people clearly where the data came from.
Have a task you still do by hand every day? Tell us about it at novatise.com and we can usually show you a working prototype within 24 to 48 hours.
Frequently asked questions
The bot only reads prices that shops publish openly on their own websites. Anyone building a similar tool should read each site’s terms and robots.txt, keep request rates low, and take legal advice for their situation. This is general information, not legal advice.
As of 30 October 2026, phoneprices.sg was updated daily and goldprices.sg checked its sources several times a day. Each site shows a last-updated time, and you should confirm the final price with the shop.
Yes. We build bots and automations that read published data, check it and deliver it as a website, report or alert. We can usually show a working prototype within 24 to 48 hours.
Related field notes
- Building Business Process Automation with n8n, Claude API, and Supabase
- I Run a Fleet of AI Bots and Became a Human Message Bus Between Them
- Why 24-Hour Rapid Prototyping Beats Months of Requirement Specifications
- A Singapore phone retailer: website and back office, rebuilt.
- All articles and field notes