First, retailers need to collect as much information as possible about their competitors. For instance, hone in on the pricing strategies of rivals. Additionally, examine the competitive pricing history as well as what is currently trending.
For this, retailers have to get the most accurate competitor data. There are two sequences possible here.
Some problems that retailers face when trying to match items:
If the data matched isn’t accurate, then that can heavily influence decisions in a negative way. For instance, poor assumptions and useless choices could be made, resulting in a decrease in profits. Therefore, the depth of the comparisons, the rate of zero prices, the number of errors, and many other aspects should be taken into account in order to make sure that the quality of the data collected is high.
There are two types of product matching: the automatic one and the visual one. Combining them is the best way to get highly qualitative data.
Automatching speeds up the work of the visual matching team by automatically categorizing the goods. Using deep learning technologies, the product matching software is able to find an item on competitors web sources quickly and with a high degree of accuracy.
First we have to classify an item by examining its title (brand, model, name, etc.), image and some particular keywords in the description. It gives us probabilistic assessment of the item belonging to a particular class.
For example, if we’re looking for an electric kettle and the competitor is not selling kettles on his website, the matching software will know about it.
The matching software will narrow down the funnel and compare all kettles on the website to the one in question basing on its title, price, description, technical characteristics, color, and image (here the neural networks are of great help). First the software sweeps away the kettles which have dissimilar titles and characteristics.
As a result of previous measures, we get the kettles which have the same category and similar titles. Now we need to apply the sophisticated models, e.g. comparing the product titles with the help of neural network, making the evaluation of price proximity (the more the price is different, the less is the probability of match).
When we determine that the prices, the titles are similar, we use deep machine learning to compare the pictures when it’s possible (if the quality of picture is good enough).
For every item, we build metrics to define the probability of match. If one of the chosen items is extremely similar to the product in question, we realize it’s a 99% match and we don’t need it to be checked by humans. Though, some of those items are to be checked in order to make sure the automatching algorithms work fine.
If the match probability is less then 20% we conclude that the item was not found on the competitor’s website.
If out of the chosen products there are 3 similar items, we assign the task to the Product Matching Team.
Increase your revenue and markup with our Price Optimization Platform.Try Competera