23 March 2021
I’ve been thinking about marketplaces recently in connection with newsletters. For example, CrowdMagnet is a marketplace I’ve used to find other newsletters to cross-promote with The Sample. There are also various attempts to connect newsletter authors with sponsors. (I’ve sponsored a few newsletters myself). And of course, there are plenty of other non-newsletter-related applications that exist mainly to create connections between people based on some specific kind of need (like Upwork for freelancing or Airbnb for spare rooms).
These kinds of thing are usually hard to start because of the chicken-and-egg problem: they can't provide value until they have lots of users, and you can't get users until you provide value.
So here's a thought that I had for the first time a couple days ago: would it be possible to solve the chicken-and-egg problem once and for all by creating a "meta-marketplace" where people could signal needs of any kind, with a search engine and a recommendation algorithm for making connections?
Take newsletters as an example. I've looked at several solutions for advertising in them, like Find Sponsors, Letterhead and SponsorGap. For cross-promotion, CrowdMagnet is the only thing I'm aware of. If the meta-marketplace existed, I might use it to indicate that I want to:
advertise in newsletters with 1k - 10k subscribers
cross-promote The Sample with other tiny newsletters (150 - 200 subscribers)
Since there are lots and lots of people who want to buy and sell ads, the meta-marketplace might start out providing value to just those people. But as the sponsorship market continues to grow, the number of participants who are also interested in cross-promotion will increase. The cross-promotion market can solve its own chicken-and-egg problem by piggybacking off the sponsorship market.
So the big question here is: could this be taken to the logical extreme? Could you have a general-purpose network that lets you signal any kind of need, not just newsletter-related needs?
A few issues come to mind. One is the practical problem of how to structure the data. Perhaps it would be feasible to use a collaborative tagging system. When you create a need, you choose a set of tags to apply, some of which can have values. For example, you could choose "newsletter", "cross-promotion" and "subscribers: 150". As you type, the UI presents tag suggestions so that (when possible) you can choose tags that are already commonly used.
Next is an easier one. Marketplaces often do more than connect people; they facilitate transactions. Letterhead, besides connecting advertisers with newsletters, provides tools to help people purchase ad slots and upload images and copy. The solution here seems pretty simple: outsource it. When you create a need on your profile, you can write a free-form description to go along with the tags. If you're selling newsletter ads, you just include a link to some other website where people can purchase an ad slot from you.
I've saved the hardest problem for last: trust. Connecting everyone means you connect scammers and other unsavory types. If I'm thinking about buying an ad slot in your newsletter, how can I know that I will in fact get traffic in enough quantity and quality to make it worthwhile? You could try to outsource that too: if you have a bunch of subscribers, include some social proof. Link to tweets and articles that mention you.
That might work if the problem is "given a candidate, find out if they're trustworthy." But that problem is preceded by "given a need, find a list of candidates." If trust isn't built into the search and recommendation algorithms somehow, the meta-marketplace will be overrun with spammers.
My initial thought here is to just use a rating system. You can give any other participant a rating, along with a few tags to provide meaning. The critical thing here is that you don't treat ratings naively like Amazon does. The average rating of something across the entire population is meaningless. You have to take into account the trustworthiness of the ratings themselves. In our case, rather than asking "is Alice trustworthy," you ask "is Alice trusted by Bob."
You could use collaborative filtering for that. Given a bunch of user-to-user ratings, create a model that predicts what rating Bob will give to Alice. Then, when Bob searches for candidates, you rank them based on a combination of their relevance to Bob's search query and their predicted trustworthiness by Bob. Besides that, my recommender systems textbook has an entire chapter on "Social and Trust-Centric Recommender Systems," with such sections as "TidalTrust Algorithm," "MoleTrust" and "TrustWalker." (If I ever get around to reading this chapter, I'll let you know any insights I gain).
Although these concerns need to be dealt with, I don't think any of them are intractable. The main thing is just the good old chicken-and-egg problem: how do you get this rolling? What is the starting angle?