AI Price Extractor

What is the AI Price Extractor and how does it work?

The AI Price Extractor uses a custom machine learning model we've developed to extract product prices (both the current price and the referral price in case of a discount) from any online stores without having to prepare a custom scraper. You can use it to easily monitor product prices across tens, hundreds or even thousands of websites with ease.

The only restrictions to usage are that the extractor currently only works for online shops using american dollars, british pounds, euros or any currency used in EU countries, and that the extract prices can sometimes be wrong if the underlying machine learning model makes a mistake. Read the "How accurate is the extractor" section of this readme for more concrete information on how accurate the extractor is.

How should the input look?

The extractor input is very simple, you only need to specify two things:

1. Product URLs - a list of URLs leading to the product webpages
2. Proxy configuration - whether to use proxies and what kind

The actor page provides a GUI to make specifying the input easier. If you instead want to use the JSON editor to specify the input, it might look like this:

1{
2  "proxyConfig": {
3    "useApifyProxy": true,
4    "apifyProxyGroups": [],
5    "apifyProxyCountry": "CZ"
6  },
7  "urls": [
8    "https://www.sainsburys.co.uk/gol-ui/product/diet-coke-12x330ml-7517073-p-44",
9    "https://www.alza.cz/samsung-galaxy-smarttag2-baleni-4-ks-black-2-white-2-d7899023.htm",
10    "https://www.pointcarre.be/fr/olymp-chemises-manches-longues-noir-200864-68-noir"
11  ]
12}

What will the results look like and where can you find them?

The results will be stored in the default dataset of the Actor run, accessible through the run page in Apify Console. You can download them both manually and through an API in a variety of formats including JSON, CSV, and Excel.

The format of the results is very simple, they contain up to four attributes:

1. url - url of the product
2. currentPrice - the price the product is currently sold for
3. referralPrice - the price the product was sold for before the discount, only present if there is a discount active
4. currency - the currency of the extracted prices

How accurate is the extractor?

We have striven to make the extractor as accurate as we can, manually annotating product webpages from thousands of online shops and using them to train the model. As is the common practice, we’ve also prepared a separate dataset of websites that the model never saw during the training process and then estimated the model's performance using this data.

Using this dataset, we have measured that the extractor was right in more than 85% of products, and our practical experience is that the extractor correctly extracts every product we tried from two thirds of online shops. The reason for the difference between accuracy for products and for entire shops can mostly be attributed to loyalty programs and store cards, because we don't count discounts that need either of those as proper discounts, but the extractor often extracts them as such. If that's not the case for your use-case, the results could work even better for you.

Of course, since you will be providing your own URLs which will probably be coming from different online stores than we tested, the performance you see might differ, even though we tried to make the extractor as general as possible. For that reason, we recommend you do your own investigation of the accuracy of the extractor's results before you use the results for your use case.

How much will it cost?

The extractor is a paid actor, costing a fixed amount of money each month as subscription to use (which you can find at the top of its Apify store page), and you also pay for any platform and proxy usage you incur, just like any other scraper. For the purposes of trying the actor, you can use a 7-days-long free trial to try it without the need to pay the monthly subscription.