Beli: Food for thought

Breaking down how a restaurant-ranking app works shows us the general principles of vertical social networks

If I was a venture capitalist interested in consumer tech and I found out that a new app out there was being used by every college student and young professional, it would pique my interest. The reason for my interest would be simple: pattern matching. The 18-25 year old demographic is coveted within the consumer category because the three most important social networking companies of the last two decades — Facebook, Snapchat, and Instagram — all started their stronghold on the broader public by first saturating college campuses.

So when I saw all the interns at my day job using a new app that I had never heard of, I had to download Beli for myself. 

On its face, Beli is a simple app that users can use to rate and rank restaurants, create lists, and find good places to eat around town. And as the weeks have passed by, I’ve gotten many “[Person I know] just joined Beli!” notifications on my phone.

This anecdotal experience matches the data. As of March 31st, 2023, Beli shared with its users that it had more than 6 million rankings on the platform. At a rate of 5-10 rankings/user, that gives us 600,000 to 1.2 million users of the platform. More importantly, that is up from 2.5 million rankings at the end of November, 2022. Assuming those same average usage numbers¹, the comapany had 250,000 to 500,000 users just a few months ago².

This strong growth trend makes Beli worth diving deeper into — what are the factors that are animating this growth, how can Beli sustain it, and what are the challenges that are up ahead for this app?

And by identifying the specific factors that are behind Beli and the challenges ahead of it, we can explore on a few aspects of what makes consumer apps, specially vertical social networks, tick more broadly.

So, let’s start with how the app is used.

Proof of work: The Restaurant Rating

The fundamental unit of interaction or proof of work³ in Beli is the restaurant ranking. After you go to a restaurant, you can search for its name and add it. This brings up a list of comparisons against other restaurants you have previously rated. You choose which restaurant you liked more and after a couple of taps, you have placed the restaurant in a list of restaurants you’ve ever visited.

It’s a chess-like Elo ranking with custom parameters. This means that every restaurant is rated not on an absolute scale, like on Yelp or other review sites, but on a relative basis. For the cinephiles in the group, this is similar to the Facemash scene in The Social Network, bringing an end to the Facebook references in this piece.

This has one distinct advantage over the rest of the reviews available. It is personalized to you. This is generally a good thing: while nobody has totally distinct tastes, nobody has totally overlapping ones either. It also allows Beli to predict how much you’ll like a new restaurant that you have never been to, but might be considering.

Note that this by itself is not a defensible idea: it took me about 30 minutes using a little ChatGPT magic⁴ to create a rudimentary version in Google sheets. I know people who maintain robust spreadsheets that mimic Beli (see screenshot below.)

A friend’s NYC food spots spreadsheet

Global rankings are also limited in context. For example, one of my highest rated places on Beli is Bouchon Bistro in Yountville, California. So every time I rate any restaurant highly, it asks me for a comparison with Bouchon. The problem is that I’ve only been to that restaurant twice and while I’m there, I’ve always been on vacation — those experiences are very, very different from a work dinner or a brunch with friends. Rating all meals in the same way flattens the experience. While this issue is not exclusive to Beli (all ratings are doing this implicitly; you could argue that the goal of ratings is to standardize the experience), it becomes more important when the explicit promise to the customer is that of personalized ratings.

Lastly, there is currently a level of friction involved in this experience that is not ideal for building a sticky product. Every time you go to a restaurant, you have to remember to put in an entry afterwards. As the network of users becomes large enough for you to check the app on its own (consider how you might check Twitter or Instagram even if you don’t want to post anything there), this could become less of a problem. But it is a real one right now.  

Still, this proof of work works well and power users like it. It may be a bit excessive for the casual user, but for now, Beli can afford to and should ignore them. What all users need though is an app that encourages creating more ratings. To do this, all the elements of the app need to work together.

App design: It’s probabilities all the way down

Every pixel on our little black mirrors is precious. So what a team puts on the homepage is a strong reflection of their priorities and of what they believe to drive engagement and growth. To that end, other parts of the Beli app include restaurant lists with featured restaurants in the area, ratings from your friends you’re following on the app, and quick filters that you can apply to the feed.

More broadly, one way to look at an app design is to identify what parts of the user experience is deterministic vs probabilistic. For example, a feed like Twitter’s “For You” page is entirely probabilistic — tweets are ranked and presented to you in the order of how likely you are to engage with them. Higher probability means the tweet is closer to the top of your feed. On the other hand, its “Following” page is more deterministic — tweets are ranked chronologically (more recent tweets show up first) and only include people you follow.

Think of the extremes here: TikTok is entirely probabilistic; email has nothing but sweet, sweet determinism. 

If we look at the Beli app, there is a good balance of the two. That is a good thing. While most algorithm-driven content has a higher engagement rate, Beli is a relatively young app. Its users probably just want to see content from their friends as your friends are more likely to be in the same geographic area as you and have a similar willingness to explore new places. 

But the algorithmic angels should not be ignored for too long. There is a real social cache component to eating out and looking at people beyond our current circles can be pivotal in growing the app. This brings us to the exploration where we look beyond the mechanisms of the app to the deeper psychology driving usage.

Job(s)-to-be-done: Balance of utilities

If you spend any time on Instagram, you’ll definitely come across a photo of someone’s meal with a location tag of the restaurant. The phone eats first. 

There are two things being signaled here: you’re (1) signaling that you have the financial capital to be able to go to fancy restaurants, and (2) if it’s a lesser known place, that you’re “in the know” of what’s good. There are no Instagram stories at Wendy’s⁵.

It’s also generally true that no social network stays “cool” for too long. It would have been really cool to have Facebook before March 2004, when it was only available to Harvard students. As more people joined, it became less cool to a point that today, I only use it to send occasional messages to friends from high school who I have lost touch with. Take this to the extreme — do you really want to be on the same social network as your parents? — and this principle makes general sense.

But as the social cache of a network decreases, its practical utility often increases. If you were a Facebook user in March 2004, it would have been really nice to be a part of the “in crowd” but if you wanted to send a message to a high school friend who went to another university, you couldn’t. Once the Harvard restriction was lifted, you could. That’s a marked increase in practical utility.

Borrowing from Eugene Wei’s fantastic Status as a Service essay, we can think about this along two axes — with social capital on one axis and utility on another.

Most social networks can be placed on these axes. Importantly, people want to have social capital with their perceived in-group. Intuitively, status games that a high school teenage girl plays versus what a middle-aged man plays are very different.

So, where does Beli fall right now? The app has practical value in that you can choose where you want to eat based on the recommendation of your friends and strangers in the know. And it’s definitely a “cool” app that many people are signing up for — it’s a way for you to see where your friends are eating and show your friends where you are eating. I’d put it somewhere along the medium positive utility axes and a low social capital capital axes. This could definitely change in the future.

Hype behind a product is generally a good thing, but too much social capital too fast can be a bad thing. From the excellent piece on The Hype Subsidy by Sarah Travel:

Imagine building a marketplace where a hoard of people outside the company are deciding how much of a subsidy you should provide for transactions every week, and to you, it’s a blackbox. Whether it’s 20% one week and 80% the next, all you know is it’s there, but you don’t know how much.

To say the least, it would be difficult to optimize a marketplace under these conditions. People are using your marketplace for a broader set of use cases than you had anticipated, but would they happen without the subsidy? People are referring their friends, but would that happen without the subsidy? Etc.

This is what building in the midst of hype is like. Everything works better (and differently) than it would without the hype subsidy.

It can be tough to distinguish between real growth and what’s being driven by the hype subsidy. To an outsider observer like me, Beli’s growth has been fairly organic — they haven’t pulled any outsized stunts to shock-and-awe users or investors. All this is good: Beli has sustainable user growth (shared above) which means that they are likely to get stable user churn.

All things considered, it is difficult to make something that users want and Beli has done that. Its growth will always be driven by a combination of practical and social utilities and while one side or the other might dominate temporarily, it doesn’t seem like the company is leaning one way or the other. In doing so, it has remained true to its users’ needs.

But as it fights to onboard and retain those users, Beli has to be aware of the competition it faces on multiple fronts.

Competition: Chipping away at the Goliaths

Beli competes along both two axes of its utility — the practical and the social.

On the practical side, it competes with the likes of OpenTable, Resy, Tock, and frankly, just a simple “best places to eat in [City X]” Google search. These are all ways to find out where you want to eat. Beli might be better at these with its friend-driven recommendations and personalized recommendation scores, but the alternatives are not materially worse. 

Along the social utility dimension, Instagram is the most important competitor for Beli. People are already on Instagram and it's the de facto “let me tell you how cool my life is” platform, so using it to tell people that you are going to eat at a nice place is a common use case. The one possible counterpoint here is that Instagram’s pivot away from photos and stories to be more Reels-focused could change this, but for now, it’s a real competitor for Beli. 

Both the practical and the social point competitive forces point towards a power user-focused strategy for Beli. While there might not be a strong habit component for most people who go out to eat casually 1-2x a week, having a separate place that aggregates all your visits to all restaurants, creating themed lists, writing down your favorite items, etc. is helpful if you’re doing this a lot. On the social side, someone like a food influencer might want to keep all of their favorite restaurants and ratings in one place. They can share it with their followers and have them follow along. It might have some social utility for the casual user, but it would be limited.

With this understanding of Beli’s relative utility for its different user segments, its paths to monetization become clearer.

Business model options: Worship your power users

Monetization for a niche app like Beli can be challenging. Right now, the app is free to use. But with hundreds of thousands of users, they’ve likely started to think about how to generate revenue. There are some established and some not-so-established ways they can do this:   

• Display ads: Restaurants could purchase the ability to be featured on lists and to show up first in searches. This would be similar to the type of advertising that Google Maps does. Display ads would be easy to implement since they can be fully self-serve on the restaurant side and would require no tuning of algorithms on an ongoing basis. For the restaurants that would be advertising, they would know exactly what they are buying. On the user side, they are generally okay with ads if it means keeping the app free.     

• Subscription revenue: Beli can also ask for power users to pay a regular amount for more advanced features like turning off ads, ability to create more than a certain number of lists, to be featured on leaderboards, etc. If it can get a critical mass, they might be able to have enough paying users who are willing to pay to just support the existence and development of the app (and not necessarily to get any extra features.)

• Marketplace: A more interesting business model is one that Beli seems to be heading in the direction of. It currently has a feature where you can post a reservation that you have for a restaurant that you can no longer make. You post the reservation and presumably, some other Beli users can take it off your hands. This can be especially useful for cities where getting a dinner reservation is harder than getting into college and there are onerous cancellation fees. A marketplace like this would work in large urban cities and likely has large overlaps with current power users. It’s unclear if Beli currently takes a cut to facilitate transactions like this, but it could do that. This business model would require a large network of power users, so I’m skeptical in the absence of data — but it would certainly provide a useful service that some people would be willing to be pay some amount for.   

As Beli navigates this, its north pole should be its power users. The kind of person who is hunting for reservations at nice restaurants and who obsessively keeps track of all of them is likely to find enough value to either pay a small monthly subscription and engage in the marketplace of reservations. Advertising on the other hand can be a business model that supports a broader usage profile across the user base.

As it navigates this, it could be helpful to see what one version of success could look like in this space.

Lessons from a success: Letterboxd

The movie rating app, Letterboxd is an example of a vertical social network⁶ that Beli can learn a few lessons from⁷. It has carved out a specific niche among the die-hard fans of cult movies and the general movie going public alike. In doing so, it developed a business model that pairs ads (for free users) and subscriptions (for premium subscribers.)

Letterboxd has a similar premise as Beli — you can rate movies you’ve watched, add things to your watchlist, leave comments, and create lists. Given that, there are a few things that Beli can learn from Letterboxd:

1. Cultivate an authentic voice: Food can be made fairly pretentious⁸. So can movies. Something that is unique about Letterboxd is that reviews on the site have their own unique voice. The litmus test for this is that you can usually tell by reading a review that’s outside Letterboxd that it would be something that would be posted on Letterboxd. As one user shared in the New York Times’ profile of the app:

   If I’m writing a professional review, I’m writing for a general audience,” he said in a recent phone call. “Whereas on Letterboxd, I don’t worry about pro forma things like plot synopsis. I make jokes and references you would have to have a fairly deep film knowledge to understand. I find it much more liberating.”

   If Beli can encourage this kind of a “liberating” voice on its platform, then there is a whole new genre of food critics that could be enabled through the app.

2. Lean in on the patronage model: Letterboxd has a similar subscription model to what I outlined as a potential for Beli above. For $19.99/year, you get a handful of pro features. These features are nice, but none of them are critical to the usage of the app. Users primarily pay to support the development of the app and to enable a community that they like. I only pay for 3 app subscriptions (Bear, Paper, and Letterboxd), and Letterboxd is one of them because of the last lesson that Beli can learn.

3. Right sized ambition: A key element of Letterboxd's success is that there is no world domination plan. It’s a simple app that does what it says it does. It’s delightful to use. It’s fast. It didn’t try to pivot to becoming a Web3 app last year and it’s not going to try to become an AI app this year. People and companies struggle when they try to become something they are not. Letterboxd limits its scope but executes perfectly on it.

The comparison is by no means a perfect one. First, movies are generally not supply-constrained — if you want to watch a movie you can do so through any number of streaming services, independent theaters, and chain cinema houses. Food is not like that. There is also the case of the back catalog. Food is not like that either.

But the same general principles of being a niche app apply: can you provide a sufficiently useful and delightful experience for a particular activity that people are willingly to give you their attention and money?

Go-to-market: A Gen Z app at heart

Once you consider the factors of tough competition and a need to hone in on its power users as it broadens the user base, Beli’s go-to-market strategy becomes apparent. 

Most of its efforts have been focused on TikTok and Instagram reels. Restaurant features and video lists of the “Top 10 places to get a Bagel in LA” variety are the primarily content present on these platforms. This is particularly appealing to food influencers targeting the late Millenial and GenZ crowd.

This is not surprising at all. For my generation, a workout that is not tracked on your Apple Watch might as well have not happened. A weekend with friends that is not shared on your Instagram might as well have been spent lounging around on your couch. A thought that is not turned turned into a Substack might as well have never entered your brain⁹.

Analogues for this kind of an app exist beyond the food space. Kit.co is an example of a website where you can create a page that lists out the equipment you use that works through affiliate marketing and is popular in the podcaster/YouTuber crowd. There are other services like this across various verticals.

Overall, Beli is doing what it should be when it comes to their go-to-market strategy. They have staked out two platforms — TikTok and Instagram Reels — and are focused on executing on them. It aligns with their target audience and the broader tailwinds of these platforms shifting towards video-first¹⁰. 

Conclusion: What’s Beli’s scale of ambition?

Today, Beli is an app on the ascendancy with growth numbers that would make any up-and-coming founder jealous. Users are generally engaged and the team continues to add new features rapidly. Even in its current form, the app and go-to-market strategy mesh together to serve its power users. The competition from food-specific platforms and more general social networking apps is stiff, but Beli has been able to carve out a successful niche so far.

As it considers next steps, it needs to answer some key questions. Do they want to raise millions of dollars, hire a large team, and try to dominate every cultural conservation about food? And do the problems they want to solve require that?

Or they want to build something more limited in scope, but ultimately in greater service to a smaller number of people who are very passionate about food? It would still be a success, just not a flashy one.

There is nothing wrong with either option, but they do have real tradeoffs. The limited scope option likely eliminates large exit outcomes like an acquisition by a larger brand. On the other hand, momentum and growth like this can always be a flash in the pan and fortifying yourself with cash is an attractive option. Ultimately, founders have to ask themselves the question of what type of company do they want to build?

That perhaps, is food for thought for them to chew on.

1. Usage numbers are likely lower as cohort quality generally degrades over time, but let’s keep it simple in the absence of concrete data.

2. One callout I will make right at the top is that Beli has an invite system that you don’t actually need to use to get access to the app, but that it will seem like you need. It seems to be a remnant from the time that it was actually an invite-only app. Right now, it just leaves a bad taste in my mouth.

3. Every user-generated content based social network has a “proof of work” mechanism. Ex: Snapchat: the story, Facebook: the post, Twitter: the tweet, etc.

4. One of the most important things that ChatGPT does is increase personal productivity factors by creating something like this on the fly. I remember a summer internship in college during which I spent 3 days perfecting a cohort analysis spreadsheet that had to be presented to VCs. Today, that could be recreated in 30 minutes with multiple bells and whistles using ChatGPT.

5. Unless done ironically. Moves towards more “authentic” sharing of moments like BeReal, etc. play on similar dynamics, but incentivize subtler signals.

6. I’ve previously written about Letterboxd in 2020.

7.  There have, of course, been more than just one successful vertical social network. One of my other favorites is Goodreads. But it’s been acquired and had its soul stripped by Amazon. It’s likely not an example of a social network to emulate. 

8. If you haven’t watched the 2022 movie, The Menu by Mark Mylod, I highly recommend it.

9. I’m not being holier than thou here. I’m admittedly guilty of all of these.

10. Relevant for Instagram which was previously photo-first.