Why Adtech needs to know what the product record is on each product page at each store

AdTech systems can generate more relevant ads for products if the contents of each store page is known. Knowing the product record on a page and the records (pages) at other stores that contain the same product leads to higher conversions. It is of paramount importance for AdTech companies to have quality matched product data because bad data results in poorly targeted advertising, hence lower revenue.

There are two obstacles for AdTech that prevent more relevant brand and direct advertising being delivered based on user visits to product pages. The first is that AdTech systems don’t have a comprehensive product database. The AdTech companies, along with most large firms, have been unable to reverse engineer the online product databases, match the same record at different stores, and identify the bad and missing data. Once the AdTech systems have access to a product database which contains matched records can build bidding systems. If a user is looking at a product at one site then all other sites which carry the product can then be asked to bid on the product.

AdTech companies need clean product databases with matched records in order to deliver more relevant ads and increase conversions. This will give AdTech companies an advantage over Facebook with their undifferentiated content which serve ads based on a limited understanding of the product graph on the web. A deep understanding of the product graph based on a universal product database of web product offers is the key to creating systems which serve targeted brand and direct advertising.

Google has a limited focus on products as evidenced by their not so deep commitment to Google shopping which has its data populated by paid customers. Google search page results for products show paid placement, Google shopping, and ranked results for the product which may include blog posts, reviews, and product pages. Adsense is based on keyword matching and not product data records for serving ads. Google has not demonstrated that they have a strong grasp of product data records. All ad systems track the page that users have visited and then if the user does not buy will reserve the same product at the same store that the user visited in subsequent ads. Ads for the same product at different stores are not served. Users are not shown different options for the same product. This results in users feeling that they have been tracked and no value is added by the tracking ads.

No company has demonstrated at scale that they have Quality Product Data. It is relatively easy to find brand ads, model numbers, and UPC’s on product pages. Reverse engineering the product record and then finding matching records is required in order to create a normalized consumer product database. There are a few startups that have product matching technology. None of the startups have demonstrated that they have product cleaning and bad and missing data identification and data suggestions and corrections for bad and missing data. People in the ecommerce industry recognize the need for a universal product database with clean and complete data. The database should be normalized. The bad data should be identified and either fixed or removed. The missing data should be inserted into the product dat records to augment the record.

Identification of bad data is absolutely necessary to prevent bad matching. If the data record contains an incorrect UPC which belongs to another product and a correct brnd name and model number and the wrong UPC is not identified a matching system is going to join the two different groups. The two different groups will become one group. It is critical to separate the groups correctly. It is not possible to get all groups separated correctly. But it is possible to have a very high percentage of groups which are correctly grouped and do not have bogus joins between groups which contain different products.

Each store displays their products for sale. Offers are normally generated from a database. The database records for brand records are imported from the manufacturers data set or entered manually. Additional information is added to the records. Descriptions are modified and specifications may have specification attribute names, values, and metrics changed to conform to the naming convention at the site. Matching the same product, including variants, at different stores is a difficult task. Matching products with bad data is much more difficult.

Criteo wants retailers to share data on users to help retailers take on Amazon. But if the retailers have bad andincomplete data how will their products be identified, indexed/stored/matched/merged, so that the retailers offerings can be targeted? No mention is made in this article about a normalized consumer product database and the benefits to retailers in terms of advertising and Quality Product Data which will result in increaed conversions. Amazon’s data has issues to say the least. Amazon does not seem to be making an effort to clean up the data. There is bad data, missing data, duplicate (spamming) product records for the same product at the same store with different product names from merchants (Amazon Associates), variants are not eliminated/matched when the matching gets difficult (e.g. no UPC’s), … There are lots of problems with Amazon data that could be detected and fixed but are not. The same problems exist on other large sites such as Newegg and Rakuten. Quality Product Data is a competitive advantage for retailers. And clean matched records are a way forwards for retailers who want an edge.

AdTech companies that want to serve more relevant ads to users and increase conversions need Quality Product Data. AdTech needs a product graph.