5 Ways to Harness Your Marketplace Inventory Data for Enhanced Recommendations

Two-way marketplaces exist to create a match between supply and demand. However, these marketplaces are often inundated with new item submissions, which can lead to a couple of problems. Firstly, the data in these item submissions is frequently disorganized or irrelevant, especially from user-generated text fields. Secondly, when a supply-side user submits a new item, how do you determine who it’s relevant for and how they discover it? To successfully pair supply with demand, you need to understand each item the moment it’s submitted.

The case for extracting better marketplace inventory data

Simply put, gathering high-quality inventory data is the baseline for enhanced recommendations and high-quality discovery. The more you know about the specifics of the items that make up your inventory, you can use that data to start proactively recommending it to the right audience. That way,end users will find items they love sooner, leading to a better customer experience and higher conversion rates.

Over a third of consumers stop visiting websites that recommend items they don’t care about. (Source: Salesforce)

The consequences of poor personalization and discovery mechanisms are serious. . As more suppliers and items are added to your marketplace, the likelihood of customer churn increases, due to irrelevant item suggestions. 76% of consumers find it frustrating when a business has mediocre personalization (source: McKinsey), while over a third of consumers say they stop visiting sites that recommend things of no interest to them (source: Salesforce), which can have a devastating impact on conversions.

How can I use data to get enhanced recommendations?

Look at your inventory with a critical eye and parse out the high-quality data tied to each of your items. Think things like size, color, category, and so forth. Once you do this, you can leverage the collaborative filtering technology of a recommendation solution much more efficiently in order to achieve the best results for end users.

Here are some tips to help you understand and prepare your marketplace catalog for enhanced recommendations and faster item discovery.

  1. Identify which pieces of data are crucial for your marketplace to function properly. This will also allow items to become more easily identifiable by consumers. The more unique identifiers or pieces of information tied to a given item, the easier it becomes for a recommendation system to surface relevant items to users.
  2. Extract high-quality data from each item. Defining categories and subcategories is a good start to help users self-filter the types of items they’re interested in. However, technology like Deep Content Extraction enables you to gather insights on each item beyond its structured fields. This lays the foundation to surface more relevant items for any given user.
  3. Know which types of content to extract. Natural Language Processing (NLP) text summaries, image recognition, and video recognition are all elements that you should extract when possible. Image and video in particular can provide valuable signal about what makes a product unique, as well as its similarity to other items. 
  4. Gather insights from user-generated content (UGC) – with some help from data cleanup. UGC has a reputation for being unwieldy because fields generated by users (such as “description,”) can be riddled with typos or irrelevant information. However, by leveraging data cleanup techniques, you can parse valuable inventory data within the UGC that can enable faster item discovery.  
  5. Create a submission process that enables clean data. Avoid a lot of the headaches associated with data cleanup by allowing users to provide rich, valuable data in a more formatted way. For example, a clothing marketplace might have a baseline description field (e.g., “blue jeans”) as well as fields for the item’s condition (“gently used”) or other differentiating properties (“distressed”).

Doing this preparation prior to feeding the data to your recommendation solution will help you enable faster discovery of items that are most relevant and interesting to your users. It’s worth investing some time upfront so you can maximize your conversion potential right out of the gate, as well as prevent negative customer experiences.

Alexandre Robicquet
CEO & Co-Founder

Alexandre earned his first two Master’s degrees in Mathematics and Machine Learning at the age of 23 from ENS Paris Saclay. Alexandre had his work published at the age of 21 and held two 4-year positions as a researcher under Pr. Sebastian Thrun (founder of Google X) and Pr. Silvio Savarese. He received a third Master’s degree in Artificial Intelligence and started on a path for a Ph.D but chose instead to work on building Crossing Minds full-time.