Quora's $75M Raise and its Journey towards mainstreaming AI
Insights on how Quora wants to cement its position in the AI space.
Quora’s founder, Adam D'Angelo, announced a $75M raise from Andreessen Horowitz last week to accelerate the growth of Poe (short for “Platform for Open Exploration”), its AI chat platform and marketplace. Poe is an AI platform that aims to mainstream the adoption of AI by reducing or removing the infrastructure, cost, and distribution barriers developers face when building an AI chatbot.
Operating an AI chatbot in public is prohibitively expensive, one reason many AI experiments failed. Poe is solving that problem by allowing developers to offload their model inference costs completely when they use the Poe bot query API. More on this later.
Quora’s new valuation
Let’s take a step back and revisit the fundraising. Quora raised $75M at a valuation of $500M. This is about one-third of its valuation during its last fundraising in 2017 and close to its valuation 12 years ago. Interestingly, Quora’s current valuation is approximately the same as Perplexity AI's, a company founded less than a year and a half ago.
Raising money in the current climate is challenging. High-interest rates mean a high cost of capital. As expected, lower interest rates produce high valuations, while higher interest rates produce lower ones. VCs typically value a company based on how they think it will perform in the future, i.e., discounted future cash flow.
When risk-free rates of return increase, the discount rate applied to future cash flows increases, thus cutting the company's net present value (NPV). What we are seeing now is that companies have been overvalued in the past, not only because interest rates were low but also due to unrealistic growth expectations.
If Quora is now valued at almost the same valuation as 12 years ago, it's an indication that the company's growth expectations may not have been met. Founders need to think deeply about this. Your success won’t be measured by your ability to raise new funds but by your ability to deliver on your pitched growth to investors. If you don’t, it will return to hunt your valuations, irrespective of the interest rates and cost of capital.
While Quora is cashflow positive, according to Adam, there are important questions to be answered: If Quora has existed for 14 years, why has it not built up enough cash to fund R&D? What is the rationale for raising new money to fund Poe’s development? Could the new funding be based on Poe’s valuation and not necessarily Quora's valuation, seeing that it's comparable, Perplexity, and is valued at a similar price? The potential conflict of interest due to Adam D'Angelo's position on the Open AI board adds some complexity.
I don’t have these answers. But the good news is that this is still positive news for Quora, and it opens up a new world of possibilities for Quora to deliver on its growth expectations. Artificial intelligence has always been a potential threat to Quora's business model because AI can aggregate subjective human knowledge –the main value of Quora – and cut out Quora’s value proposition.
The future of Quora
I believe that Poe represents the future of Quora (although it is currently a separate project from Quora). Its aggregation and monetization model could help AI cross the chasm and diffuse through the world. In Adam’s words:
Today, nearly one year after launch, Poe has millions of users, some using our free tier and others taking advantage of premium benefits through subscription. There are millions of bots created by Poe users through prompting and by developers through our API…
We see Poe’s role with respect to AI being similar to the role the web browser played in accelerating progress during the development of the early internet. This shapes our two goals:
Be the best way for consumers to chat with a variety of AI products.
Be the easiest way for a developer to build an AI chat product and reach a mass audience. This is true whether they are creating a bot using an existing model with a prompt and any uploaded files, or whether they are training a model themselves.
We hope that by pursuing these goals, we can accelerate the mainstream adoption of AI and all the benefits that it will bring to society, including increased access to knowledge, democratized creative expression, and accelerated economic growth. With more capital, we can build toward these goals faster, and enable a creator economy for AI that works across all technology providers. Our goal is for Poe to enable as many individual developers as possible to make a living, and for as many businesses as possible to operate profitably solely by using the platform.
It is also noteworthy to mention that while Poe is a separate project, it is currently used to answer questions on Quora, meaning there is a lot of potential for AI to improve user experience on Quora.
Tracking Poe’s journey since launch
When Poe was launched in Dec 2022, the goal was to provide users access to different text-generating AI models, including ChatGPT. At launch, it was like a text messaging app for AI models where users could chat with models separately. Instead of going to different apps to chat with AI models, users can explore them in one user interface. At a point, Poe even provided early access to LLMs that people could not access directly.
Since the launch announcement in December 2022, Poe has been rapidly releasing new features to cement its position in the rapidly competitive AI space. The velocity of the engineering team working on Poe is very commendable. Here is a run-through of its journey so far:
How is Poe different from other chatbots?
The Gen AI space is becoming very competitive, and users are swamped by the variety of options available to them either as a separate app (generalistic AI) or embedded in an existing application. The proliferation of AI models is daunting for the average user, and you can expect the law of diminishing returns to set in.
Each company develops its own generalistic AI model and promotes it by providing a UI for consumers and/or APIs for developers. For instance, Open AI (ChatGPT), Anthropic (Claude), Google (Duet AI & Bard), Microsoft (Copilot), Meta (Llama), and so on. This poses two limitations:
Most generalistic AI models are chasing the same market, leading to a zero-sum game.
There won’t be one model to dominate all, meaning users or developers need to switch between multiple platforms or build an aggregation layer at some point.
The need to build an aggregation layer will always come up whenever there are multiple options or vendors to select from, where each does not independently give an entire optimal outcome. This is not a new approach to building software products. For instance, MoneyHash is a payment aggregation platform that can route merchant payments through multiple payment service providers and optimize success rates without the merchants directly connecting to each merchant.
Poe has created a compelling aggregation layer for both text and image AI models, including ChatGPT, GPT-4, DALL-E 3, Claude Instant, Claude 2, Mistral, StableDiffusionXL, Gemini, PaLM from Google, Llama 2, etc. Poe then provides a user interface for users to interact with each model separately and even build bots using their favorite model or a combination of models. Instead of using one platform with one model, users can use Poe and have access to all the models.
Poe is not the only product leveraging on multiple LLMs. Perplexity is powered by AI models like GPT-4 and Claude 2. However, there is only one interface. Users can switch between AI models but cannot simultaneously chat with each model like Poe. Also, Poe has far more integrated models than Perplexity.
How does Poe’s monetization work?
Poe is not just an aggregation platform; it's a marketplace for AI bot developers, which makes it quite interesting. Aside from helping developers build bots –either by plain-text instruction or custom backend –it also helps developers distribute and make money off their hard work. As I have argued here, distribution is a major factor for any AI project that would succeed. Poe built the platform to empower those who want to build AI tools but do not have the resources.
This marketplace platform is not new in software products. Roblox created a gaming infrastructure that lets game developers build games, distribute them to a community of players, and make money. The Roblox Studio and other development resources are free, which means anyone can succeed with enough dedication and planning.
This typically leads to powerful network effects –the more creators build new games, the more games players can find to attract and keep them engaged on the platform. The more players are on the platform, the more developers will be incentivized to build new games. With Poe, anyone can become a developer. Let’s look at the four incentives that drive developers:
Distribution
Poe's bot recommendation system allows developers to get their bot in front of a large audience. Poe has millions of users and the potential to continue to grow. Like every other marketplace, users have optionality, and developers need to figure out what users would like and build a bot for that. The millions of users on Poe does not automatically translate to millions of users for each developer.
Also, there seems to be a marginal difference in the responses of the chatbots if you ask, say, three chatbots the same questions (irrespective of the chatbot name). However, this might change if developers continue to adopt the knowledge base feature to make their bots more differentiated. But without this, most bots may end up doing the same thing, thus reducing distribution upside for developers.
Monetization
Poe allows developers to make money from users across the world. Developers earn money when:
Referral fees are paid when your bot causes a user to subscribe to Poe (active today)
Per-message fees will be paid with every message sent to your bot (not yet available)
One thing to keep in mind is that getting users to subscribe and retaining them is the most difficult task. I have written about subscription fatigue here. According to Poe, the maximum amount a developer can earn in referral fees for a successful monthly subscription is $10, while for an annual subscription, it's $20.
Note that the referral fees are paid once and not on a recurring basis, which does not seem attractive. Also, multiple criteria must be met for a referral to be counted, which makes it more tedious. The per-message fees might be more attractive. But we will get to know when that has been rolled out.
It's hard for me to see how Poe marketplace could become the next “App Store” without giving developers control over monetization. Developers need to own their source of income (not capped by subscription fees) to have the trust to invest their development time. Interestingly, OpenAI recently launched its GPT store and might pursue a different monetization model to pay users based on user engagements.
Monetization can be a game changer in the creator economy. It’s a no-brainer: developers or creators will go to the platform where they can get the most value for their work. I hope that Quora will draw on its experience and lessons in managing monetization programs on the Quora platform to make Poe’s monetization even more attractive.
Costs savings
Operating an AI bot can be prohibitively expensive, and that is why I've argued in the past as one of the reasons most AI experiments failed. On Poe, developers don’t need to worry about API fees. Poe absorbs all the inference costs and gives developers the freedom to develop. This sounds great.
However, considering that Poe is incurring costs using third-party APIs, they need to make revenue to sustain the project. The potential implication for developers is that bots that are very loved and have high revenue upside will continue to get more audience visibility to make more money for Poe. So, there could be a situation where 10% of the bots make 90% of all the money going to developers.
Multi-platform UI
Poe is available on all major platforms ( Web, iOS, Android, MacOS, and Windows), which not only increases the platform's attractiveness but also gives singularity to the user experience.
Model Independence
Developers building on, say, OpenAI’s API are constrained to the LLMs available on Open AI. However, on Poe, developers can build using any model or combination of models. There’s a clear incentive here for developers who want to experiment with multiple models.
Nonetheless, in a world where most models may end up having marginal performance differences, having optionality may not be very attractive. Also, model-native platforms will continue to have control over the evolution of their model, which is a sort of advantage that aggregation platforms like Poe do not have.
In Closing
Quora's recent $75 million funding round marks a significant milestone in its pursuit of mainstreaming AI through the Poe platform. This financing round also highlights the current complexities in fundraising, with rising interest rates affecting company valuations.
While Quora's growth trajectory may not have met initial expectations, the infusion of capital provides another opportunity to reassess strategies and align them with realistic growth projections. Quora's ambition is for Poe to become the go-to platform for consumers and developers in the AI space. This is comparable to the web browser's role in early internet development and indicates a broader vision of accelerating the adoption of AI.
As Quora navigates this path, addressing concerns, refining monetization strategies, and maintaining a competitive edge in the evolving AI landscape will be crucial for the success of Poe and, consequently, Quora's future in the AI-driven world.
About me: I’m a product leader at Stears and convener of Abuja Product Network. I have led teams to build payment, data platforms, cleantech, and AI products —an excellent thinker and doer. Subscribe to my newsletter on Substack. Follow me on LinkedIn, and X. For speaking engagements, please email: durotoluwaolumide@gmail.com