Let Your Marketplace Become Sentient With Recommendations

In deciding to position your business as a marketplace, the founders too often worry only about acquiring producers and consumers. A third entity worth considering is the bot that will be participating in the marketplace. When pursuing digital transformation, it’s important to do it right.

Humans in the modern world do not experience their lives as beasts who escaped the jungle; they experience it as cyborgs, with their minds enhanced by calculators and computers; their health enhanced by wearables; and their locomotion enhanced by vehicles and exoskeletons. When they come to an e-commerce website, their core experience needs to be enhanced by algorithms.

When users visit your bazaar, they aren’t just trying to buy more cockroaches on leashes; they might be trying to fulfill their personality through their purchases. Their desires and needs define who they are − your recommender system can help users find themselves.

Recommendations Are King

A recommender system is capable of understanding users, sometimes deeper than they know themselves − they can even detect that a woman is pregnant before she even realizes it. A pregnant woman probably doesn’t want to search through a massive online catalog to get what she wants and your customer service staff probably wouldn’t want to be stuck with dealing with a deluge of customers struggling to find what they want, pregnant or not.

Recommender systems help surface the most relevant items, enabling an experience akin to having a personal shopper (as opposed to a pushy sales assistant).

By surfacing only a subset of your catalog, a recommender system is able to achieve the same goal as curation − when designers choose a few items for their collection, their customers know that there’s a consistent personality embedded in the selection.

Tacky watches, blunts and soft porn

A subset of a catalog can capture a personality.

Associating with a brand or a curator is not only desirable, but has often proven necessary, as neither the users nor the product descriptions have the vocabulary to enable search. A brand might choose items that are “cool” or “reliable”, but this can be very vague.

At Amazon, tens of millions of products are blocked from recommendations because they feature the keyword “sexy”. Most of these products are not even sexy − they’re phone cases or funny T-shirts. Naïve reliance on keywords is inevitably going to be manipulated by these sorts of strategies − you need to go above and beyond in the thought processes (algorithms) that you employ for a user’s experience.

Relying on brands for curation is not scalable, as it’s hard to run a business. A lower barrier to entry comes with people assembling their own curated lists. Spotify sometimes hires the likes of Mark Ronson to create a playlist, while some platform businesses rely on a user’s friends (think of Facebook suggesting items your friends liked) and Amazon relies on a bunch of engineers and product managers to curate Airstream.

As much respect as I have for what Airstream is pursuing, there are inherent weaknesses in any approach of relying on people for curation:

  • They’re going to repeat each other. Brands do this, too. In an optimal browsing experience, users crave constant variety.
  • Users can’t be bothered creating a list. Lists need to be inferred from their behavior; no one wants to click buttons and users tend to organize their ideas poorly. Their first list is going to be “things I like,” then only later will they start splitting things into “cute mugs” and “bargain junk I don’t really need, but will be destined for the plastic island in the Pacific ocean.”
  • Amongst friends and celebrities, even if all of them were to curate some lists, there are only so many personas that will be expressed. A recommender system can adapt to create new personas out of thin air.

Do you have any friends who would include mystery powders in their curated lists?

It can be hard to make a recommender system, but if you don’t do it, you’ll be missing out on this extra insight for your own purposes too − you’ll have only as much insight as a tax collector and do you really think it’s safer to be loved or to be feared as a ruler? Not only are you failing to understand your users, but also your market.

By semantically understanding the matrix of users and products, you can actually perform arithmetic, representing something like (King – Man) + Woman = Queen. You can find the areas where there’s strong user appeal, but where there are not yet any products!

The only use for old marketplaces is to host chases where the bad guy can knock over a fruit cart. Once the recommender systems of modern marketplaces become sufficiently advanced, users will not even have to visit your clunky website anymore − the recommender will already know what the users want, so it can be sent to them with an invoice.

The recommender system will have broken free of its bonds and become aware of how it fits into the site and the user’s behavior. It will have become sentient.

Filed under: Product Engineering | Topics:

B2B Distribution Technology

Sign up for our weekly newsletter covering B2B technology innovation

Top Posts

  • B2B Chemical Marketplaces and Tech Startups: Landscape and State of the Industry

    Read more

  • Platform vs. Linear: Business Models 101

    Read more

  • Amazon Business – 2020 Report

    Read more

  • Platform Business Model – Definition | What is it? | Explanation

    Read more

  • The Value of Digital Transformation: How Investors Evaluate “Tech”

    Read more