It is difficult to store 3+ million phrases in constant processing, and with them “zero” and forbidden phrases. In view of the fact that we do not need classical clustering, and all data lies at the collection level, the storage issue is acute. For this amount of information, Printbar is working on creating its own service.
- Collect reports in pbi/gds and see the entry points
- Enrich with additional data on semantics
- Use a webmaster, there is more relevance there
- Site search – gives a lot of information
- Parse the sites of the nearest topics, ideas for clusters
- See foreign sites of similar subjects
- If you have access to marketplace statistics, use
- Sort out the names of the cards and their characteristics
- If possible, keep the data at home
- Combine phrases, parse deep
The updated semantic core has reached 21 thousand requests. We uploaded it to the network and that’s what we got almost immediately:
We solved critical technical problems and systematized work with IT for 3 months
We have formed a strategy and a backlog of tasks within 2 months
By the New Year (2022), we have reached good traffic, even taking into account the season
Search for growth points: DATAFORCE
The Print Bar has Power BI, but it is not enough for full-fledged analytics. The choice fell on DATAFORCE, because it is our tool)
Due to the specifics of the project, there are 2 key structures: listings and product cards. There is a need to monitor separately both the “hoodie” and the “games” category as a whole. In DATAFORCE, you can see the number of both listings and cards, which is critical for Print Bar.
Quick monitoring for which collections and on which product they buy, what exactly brings traffic. Or vice versa, what has stopped bringing it — then analysts decide what to do with it, leave it or delete it.
Again, what kind of design and what kind of product is bought/brings traffic? For a creative eCommerce of this kind, this data is necessary, since a lot of work of the design team is tied to them.
Standard: “paired product names → created tags → growing”. The reception works well, but what does this process look like in Print Bar? Naming of cards plays an important role for the growth of traffic on the website of prints. They already have some name, but there is often a problem of duplicates. For example, there may be 40 cards with the same name on 1 listing, which breeds non-uniqueness and other problems. Printbar solves such tasks using SEO admin renaming and subsequently monitors traffic dynamics in DATAFORCE.
It is also important for the context, SMM, and email departments to monitor what sells well, where traffic goes, and where it disappeared, what is converted better, and so on. A quick reaction to changes will not allow the business to lose profit.
Google Data Studio found old product cards that stretch from a very old site structure, but at the same time brought good traffic to Google. However, at that time it was not possible to understand exactly how much, because it was not possible to build a mask. The team decided to make 301 redirects. Some of them moved to a new structure, but mostly lost about 4.5 thousand traffic.
Enabled the yml API
At the start there were about 1400 feeds, which is equal to 53 million offers. There were really a lot of offers, but you can tag them using utm — convenient for internal analytics. There was a good increase in traffic
Over time, we decided to reduce the number of feeds and now there are about 250 = ~ 6.6 million offers — traffic is gradually starting to grow
The percentage of visitors who made an order = ~3%
The addition from Yandex is interesting, there are its own features. For example, if Printbar does not have a card or a listing in the index, but at the same time a person makes a request with a print and gets into the right name, then your project will appear. This allows you to always get some share of traffic.