A subscription model for Twitter would make sense, and I would pay for it. In fact, if well-executed, it would solve the ad-revenue driven misinformation problem. Here’s why.
I work at TransferWise, which is a fintech company. As with any other financial services company, it has to deal with fraud. It’s an inherent by-product of a convenient and low-cost way of moving money.
A recent internal blog post about fraud by one of our product leads advised a different way to view fraud, which got me to thinking. It said, paraphrasing, that instead of seeing fraud as a by-product, we could see it as a product-market fit in a market you really don’t want (fraudsters). Just like the success of a product can be defined by how well of a product market fit it is for the customer base you want to address, the success of fraud prevention can be defined by how well of a product market fit we have for regular customers, while having atrocious product-market fit for fraudsters. The way to do this is pretty much the inverse of what makes a product great: instead of removing friction from the customer experience, you could increase it. You make the product expensive, rather than cheaper. The trick is to identify points where you could do this effectively, so that the customer base you want to address is minimally impacted, while your fraudster base encounters more and more pain in their usecases.
This is very similar to how you would approach security in a product. In fact, a lot of algorithms for encryption work precisely like this: They don’t make it impossible to decrypt a message, just infinitely difficult with the current technology. But I digress.
Since my post-uni applications for an engineering position, I’ve been thinking about how impactful it would be to solve the misinformation problem Twitter faces, almost as a very infrequent side hobby. So, this blog post simmered in the back of my mind for a little bit, as I thought through it in the context of an “social information” product (which, I define, as any product that provides a way for anyone to publish their thoughts online to an audience, like Twitter or Medium).
Let’s consider the case of Twitter. Twitter has had a great product market fit in the market of information fraudsters. A financial fraudster aims to make more money than they invested by abusing a financial product. An information fraudster, aims to accumulate more power by creating an audience it can influence by means of misinformation. The end goal could be more money, but also could be more election-rigging power.
How do you reach a terrible product market fit for information fraudsters, while keeping a great product-market fit on the rest of your well-behaved audience?
You could start by fact-checking statements made on the platform, but it’ll only get you so far, as Twitter is a platform for voicing opinions, not just facts. And it’s easy to argue “free speech” while pushing forwards fraudulent information disguised as, or within, an opinion.
So, what other ways could you add friction to the product?
You could, indeed, charge for it. A cost that is small enough for the markets you want to address, while making it infinitely more costly for abusers to use your platform. Or, at least, costly enough that they jump off to a different one with less cost :)
How could you charge in such a way? Let’s look at some ways a regular customer base differentiates from a fraudulent customer base for Twitter (hypothetically, as I don’t work there or have enough data):
- a regular user has 1-2 accounts, whereas a misinformation campaign has many many more accounts,
- a regular user, I assume, reads more than posts, whereas a misinformation campaign posts a lot more than reads, and posts a lot more frequently than a regular user
- misinformation campaigns heavily retweet posts from accounts that are also part of said campaign, both to give credibility to the posts, and to hide away, by layering, the fact that the original info was fraudulent (just like money laundering).
As long as your charging model punishes information fraudster behaviour significantly more than regular user behaviour via charging differently across these differentiators, you’ll give your users minimal pain while solving 80% of your fraud problem.
(As I type this, I am acutely aware of how some of the VC twitter would fall bang into the information fraudster bucket. But then again, either I am right and there’s a ton of survivorship bias we could use less of on the platform, or VC’s believe in themselves and the value they’re creating for the world enough to pay up and reach said audience).
The big gotcha here is that, just like any financial fraud model, your information fraud model (and the associated behaviours) will have false positives. It will de-platform some individuals that should have a voice. The onus is still on you to keep making your model better. Your job will never be done, and the goalpost will be ever-moving. But I’m happy to pay to try it out.