Matching Buyers to Sellers in an Online Marketplace: Using Machine Learning to Optimise Price, Volume and Decongest Trade
This post was written by Alex Ioannides, Director of Data Science & Market Design at Perfect Channel.
You don’t need to be an auction theorist to know that getting the right people in front of the right inventory, at the right time, is the key to trading your goods with the people that value them the most – any trader can tell you this from their own experiences.
Similarly, Auction Theory shows us explicitly how price is optimised by maximising the number of bidders participating in the price discovery mechanism – if we maximise participation we are more likely to trade goods with the bidders that value them most.
These are easy concepts to grasp, but harder to put-to-work systematically on a big online marketplace.
Buyer-to-Seller matching is the process of signaling to the buyers of goods, whom the sellers with the inventory of most interest are; and likewise, the process of signaling to sellers, whom the most appropriate buyers for their goods are likely to be.
The aim is to solve the problem of finding the best mix of market participants to invite to an online ‘trading room’, in order to achieve the best possible prices through the sale of goods to those that value them the most.
This problem is acute when then are 100s of buyers that are largely unknown to a community of 100s of sellers, each of which has 100s of items of inventory for sale. This situation often leads to a marketplace where there are 100s of listings available for sale at any one point in time, requiring considerable time and effort to browse and surface the listings of greatest interest or value. Ultimately, this leads to a marketplace that is ‘congested’ with potential trading opportunities, which leads to fewer bids, sub-optimal pricing and weary traders.
Perfect Channel applies machine learning to historical trade data to build intelligent algorithms that match the most appropriate buyers to sellers with goods for sale that will be of greatest interest and value to them. More specifically:
- Our matching algorithms can work alongside any ‘hard coded’ business logic that needs to be present. For example, buyers and sellers can be filtered for eligibility by jurisdiction, credit worthiness, etc. – so only those counterparties who are actually able to trade need be notified of an opportunity;
- For any good (or goods) available for sale we produce a list of eligible buyers that are most likely to place a bid. Likewise, for any good that is desired by buyers we can produce a list of sellers that are most likely to be able to fulfil the order. These lists can be used to automatically invite market participants to participate in a trade, or for actively recommending inventory within a web application (i.e. as a recommendation engine); and,
- The number of buyers and sellers invited to participate in a trade, can be automatically chosen, such that there is no expected increase in final price from notifying any additional participants. The idea is to strike a balance between achieving optimal price discovery for a single trade listing, and ‘spamming’ all market participants, saturating their attention and congesting the marketplace at the detriment of all other trade listings. This strategy is based on classic results from Auction Theory as well as our experience across many marketplaces, so it has been baked-in to our matching algorithms.
Getting the right people in front of the right inventory, at the right time, is the key to trading your goods with the people that value them the most, and optimising the price discovery process. As marketplaces move online the number of trading opportunities and signals are multiplying at an exponential rate, making this seemingly simple process prone to the ‘inefficiencies of scale’. Perfect Channel’s intelligent marketplace technology provides the solution by automating the process of matching buyer-to-sellers.