%20(1).png){kind=small}
Product search is the task of finding exact product matches on various websites. The task itself requires time and a basic understanding of what words mean. Given there is no complex critical thinking involved, this is the perfect task for an AI that never gets tired.
We figured we could do better than a human by searching products with AI, and we tested this by searching supplier products on Amazon. That is, we took lists of thousands of products and identified matching Amazon listings using AI. The AI generates a search query, reviews the search results, generates additional queries if needed, and scores the results as matches or non-matches.
Bots aren't perfect, but they do have a secret weapon - they never get tired.
The Product LLM's AI bot:
In-House Experts (Human):
The 2 minute time constraint represents a real-life constraint of scanning efficiently. 2 minutes per SKU for a typical 1000 SKUs source list is almost a full work week of scanning (33 hours). It's unlikely that a team performing this work manually would spend more than that amount of time when given a large dataset of rows and working efficiently.
We increased the stakes by pitting the AI not just against our own team, but our clients as well.
Across the first two face-off methods, the number of products identified more than doubled. We also saved approximately 100+ hours if we were to run that trial across 6000 SKUs, but we have better things to do!
For the AI vs. in-house expert trial, the in-house experts were limited to 2 minutes per SKU. After reviewing the matches for accuracy with our human quality checking team, we found 2-3x more matches with the AI.
For the trial against our client, we picked a client that was sourcing from a direct relationship with a Proctor and Gamble division. This client hired us to help manage their e-commerce portfolio and list their products on Amazon listings. Before they had access to our product search, they manually searched their product list for matches (without time constraints), and found 0 to 2 ASINs per SKU. After our bots scanned the same list, we provided over 2x more options.
Additionally, the clients' products had an average margin opportunity of 12% for selling those products at the current Amazon listing price. After our bots scanned the same list, our products found 18% in theoretical margin opportunity. Given there were 2x as many options, that meant they could have made more than 2x more profit, assuming they had the investment firepower.
Gone are the days of having a virtual assistant team manually search a product list for matching e-commerce listings on Amazon, Walmart, or eBay. Instead, you can use AI search by accessing our our API, which you can try out for free in our API playground. You simply provide a product description and pick a website to search.
There are three steps to conducting AI search. 1) Use our /search-products API endpoint to search for product candidates. 2) Take the candidates from step 1 and fetch the text from each candidate web page using firecrawl.dev. (Alternatively, get the product description from an API like Amazon's or Walmart's). 3) Use our /match-products endpoint to compare the product descriptions from your searched product vs. the candidate page.
If you need any support, please contact us and we would be happy to assist you!
