Unicorn Briefing #1: Perplexity AI
Jamie Woodbridge
Strategic Partnerships | Angel Investing | Startup Advisor & Mentor | x-Google | xTwilio
By Jamie Woodbridge
20 May 2024
Introduction
Perplexity AI [link] is taking on both Google and OpenAI (et al.) for the future of Search. It combines a traditional search index with large language models to provide an "answer engine" experience. This AI-native answer engine aims to deliver the most accurate, transparent, and instant answers to queries via a conversational chat interface. In doing so, it seeks to fundamentally reimagine the search experience for both consumer and enterprise users.?
Key Metrics
Check out the Perplexity.ai Blog for the latest publicly announced company metrics.
Mission
Reporting their Series A funding in March 2023, Perplexity described their mission as follows;
”...[our] long-term mission is to become the best platform for answers and information, serving as the go-to source for people seeking quick, accurate answers tailored to their asks.”
See also an early fundraising deck here.
Founding Team
The founding team has deep experience and expertise in Machine Learning at world-leading AI research centres such as Deepmind, Facebook (FAIR), and Google. Several have PhDs in CS/ML. This team excels at the sort of back-end infrastructure a product like Perplexity requires.?
Product
Perplexity launched its first product in December 2022 and has iterated rapidly since. Rather than making users sift through links to websites, Perplexity uses large language models to directly provide fully-formed answers to search queries via a natural language conversational interface. It combines traditional search pipelines with large language models like GPT-4 and Claude 3 while emphasising accuracy through rigorous citation of referenced sources for each answer.?
This instant answer approach is touted as more efficient than traditional search and more accurate than foundational LLMs prone to ‘hallucinate’.
The basic answer engine product is free to consumers who can upgrade to a Pro tier, priced at $20 per user per month. An Enterprise version—Enterprise Pro—was also recently launched. In a tweetstorm accompanying that launch, Perplexity CEO Aravind Sriniva positioned the B2B product squarely as a work productivity tool targeted at knowledge workers carrying out research tasks.
Srinivas is betting that a narrow and focused effort to create the best AI-native answer engine will significantly challenge some high-value parts of Google’s search business.?
This is a bet that the search behemoth’s size, bureaucratic inertia and shareholder addiction to continued paid search advertising revenue growth will combine to form a perfect ‘Innovator’s Dilemma’. Google won’t be able to build a comparable rival product without cannibalising the lucrative revenues generated by its main search business.??A business which depends on users clicking on links and visiting advertisers’ websites. Fundamentally, Perplexity seeks to provide answers on its own site in a conversational thread. Users can access sources and citations but shouldn’t need or even want to click away. This undermines the core auction mechanics of how paid search advertising works.
In Srinivas’s own words:??
“What makes me confident is the fact that, if they want to do it better than us, they would basically have to kill their own business model”
- Quoted in The New York Times, 1 Feb 2024
Distribution Strategy
Srinivas is on record stating that Perplexity’s biggest competition is not Google but rather a simple lack of awareness among consumers about Perplexity’s own solution. This feels right. The company is still very young and hasn’t yet achieved a mass consumer awareness breakthrough. My network of largely Big Tech employees and alumni software engineers has yet to use the product (caveat: unless they are based in Silicon Valley). If my experience is anything to go by, it takes a few sessions to get your head around Perplexity and why it excels for research-based search/answer and fact-finding tasks vis-a-vis Google Search, Gemini or ChatGPT.
To address this awareness gap, the company is taking a two-pronged approach to customer acquisition: top-down and bottom-up:
Funding & Revenue at-a-glance
They’ve raised new funding twice already in 2024 (see full section on funding history below for details) with?Techcrunch reporting a further $250 million funding round imminent at a valuation between $2.5 billion to $3 billion.?(not confirmed)
While currently unprofitable, Perplexity's disruptive AI technology, rapid growth, seasoned founders, and notable investor backing make it appear an exciting investment opportunity in the evolving AI search market.
“While there has been a lot of excitement around business-to-business and developer-related AI capabilities, AI will also transform how we access information… Perplexity is at the forefront of this shift as one of the few consumer AI products to reach this major milestone of 10 million MAUs. They will become the go-to place for trusted information, which is why I am so excited that Perplexity is Nvidia's first consumer investment from our corporate arm.”
- Jonathan Cohen, VP of Applied Research at Nvidia.
Traction & Growth
Perplexity has seen explosive growth to tens of millions of users and queries within its first year, raised significant funding at a unicorn valuation, landed major enterprise customers, and inked global telco partnerships to drive further consumer adoption worldwide in 2024.?
Selected References:
Hiring Progress
The team remains small and lean. The hiring goal for the end of 2024 is ~65 full-time employees, which should help with speed of execution.?
A recent interview with Lenny Ratchitsky also provides valuable insights into the way the founders think about hiring and building out the organisation.?Johnny Ho covered the company’s approach to hiring and building products. This stresses small teams with a very flat structure with a lean focus on speed and relentless shipping.
Key takeaways:
领英推荐
Product, Problem, Solution
Perplexity vs Foundational LLMs (Claude, Gemini, ChatGPT4)
Perplexity occupies a distinct category as an AI search engine providing sourced web answers.
In contrast, Claude, Gemini, and ChatGPT4 are multi-purpose foundational LLMs focused on open-ended dialogue generation based on their training data. They don’t have specialised search abilities or source transparency.
Product Capabilities
While lacking the versatility of multi-purpose LLMs, Perplexity's focused approach, web indexing, and citation capabilities are designed to make it superior for transparent, up-to-date search and Q&A use cases specifically. This focus comes with various trade-offs where Perplexity might be expected to fall short compared to chatbots powered by multi-purpose LLMs.
Areas Where Perplexity Potentially Excels
Areas Where Perplexity May Fall Short
Technology
Perplexity leverages large language models (LLMs) from multiple vendors like OpenAI (GPT-3, GPT-4), Anthropic (Claude), and their own experimental models. However, the exact details of their proprietary models are not disclosed. They have evidently built additional layers and capabilities on top of these LLM foundations to enable features like source citation, context awareness, and (some) multi-model access. Plus, of course, their web-indexing infrastructure.
Beyond that, there is limited publicly available technical information about Perplexity's underlying architecture, data pipelines, knowledge retrieval and ranking systems, or model fine-tuning approaches that power their differentiated search and answer experience. Nor could I find any in-depth coverage on the AI/ML technologies and algorithms involved.
Worth noting, at its recent Enterprise Pro product?launch Perplexity also signalled an intent to offer a wide range of underlying models, both proprietary and open source alike.?
Business Model
No Ads… for now
Perplexity was founded on the belief that search should be an objective experience free from the influence of advertising interests that can bias traditional search engines. Currently revenue comes entirely through user subscriptions:
It typically takes longer to reach a run rate of $100 million ARR through SaaS MRR subscriptions. Revenues at this level will be required to justify a multi-billion dollar valuation (more on this in the Valuation Analysis section below).?
Churn also has to be factored into any freemium B2C/B2B business model. Some proportion of the consumer and business customer base will unsubscribe each month, returning to the free version. There is no reliable information on churn rates at this time. We also lack good data on the proportion of users converting to paid each month.
In an interview with Kevin Roose of the NYT in Feb 2024 Srinivas stated that Perplexity did not have plans to run Ads. He also confirmed that fewer than 100,000 users were paying for the premium version. Whether this anti-ads stance is sustianable long-term is an open question.?
The ‘no Ads’ dictum seems to have been softened in a Wired article dated 21 March 2024.?
Key Risks
Competition / Defensibility
The company is navigating a super competitive search market dominated by tech giants like Google and OpenAI.?
While historically, nimble startups have leveraged speed and focus to disrupt incumbents, Perplexity faces a unique challenge as its competitors are also agile and backed by the resources of the world's largest tech companies. In this AI arms race, Google, despite initial missteps with Bard/Gemini, is determined to avoid falling victim to the "Innovator's Dilemma. See Google’s one ?Search Generative Experiment which has been expanded worldwide.?
It is not clear that Perplexity will be able to differentiate its offering in the medium term. Their main advantages are speed to market and velocity of product iteration. Google’s key advantage remains its unrivalled distribution (though this didn’t help it much with prior product launches such as Google+, Wave, Helpout, and countless others, which it attempted to bake into its core search experience).?
Revenue Model / TAM
The current revenue model doesn’t justify its billion-dollar valuation. This value is determined by investor positivity about potential future growth. It is not yet clear if there is a viable route to achieving the necessary revenue growth (see more detailed valuation analysis below).?
For now, continued investor funding is required to fund operations and capex/R&D. This is still a very nascent market with new business models and revenue models expected to emerge. It’s possible that momentum and speed of innovation will enable a revenue unlock in future quarters. However, It remains unclear what this might be.
Perplexity is attempting to define a new category - Answer engine as the next-generation search engine. There is no precedent by which to estimate a realistic TAM for this category. Especially if it turns out to have a broadly value-destructive effect on the multibillion-dollar advertising industry that underpins the global search market dominated by Google. It’s possible that the TAM for Anser engines is a small fraction of the traditional search market.?
Legal/Copyright
Perplexity uses foundational LLMs like OpenAI that are being sued by publishers (see here for coverage of The NYT legal action): This may also impact Perplexity as a downstream user of these models. The could also be a direct target of litigation given the product explicitly synthesises data from publisher websites it has indexed.?
Perplexity is not available in the EU. Its API Terms of Service call out the region's stringent regulatory requirements around GDPR, Data Protection. This could limit the breadth and velocity of global adoption of Perplexity services and products.
Infrastructure / Cost-to-serve
As stated above Perplexity leverages third-party foundational LLMs as well as building and using its own to power its answer engine. Compute costs remain stubbornly high. A useful comparison is OpenAI. Multiple reports throughout 2023 pointed to it being deeply unprofitable, with total annual losses ballooning to $540 million despite revenue topping a billion-dollar run rate. Reportedly, in early 2023, the company was spending around $700,000 per day to meet ChatGPT demand. Given increases in user numbers and model diversity and capabilities, it can reasonably be assumed that costs will continue to increase even as revenue surpasses the $2 billion ARR mark in 2024. Clearly maintaining pace with the broader pack of competitors will be hugely expensive for Perplexity, too.
Investors & Funding-to-date
Perplexity has raised over $165 million from prominent investors including NEA, IVP, Nvidia and Jeff Bezos. Nvidia's investment is particularly strategic given its dominance in AI hardware. The most recent fundraising, a Series B1 round in April 2024? valued the company at around 1 billion dollars.
IPO Horizon
Unknown, likely several years away as the company is only 18 months old.?
Fundraising To-Date
Other Notable Backersnbsp;
Stanley Druckenmiller, Garry Tan (CEO of Y Combinator), Dylan Field (CEO of Figma), Daniel Gross ( former head of AI at Y Combinator), Tobi Lutke (Shopify CEO), Elad Gil (Founder, Color Health), Nat Friedman (Former CEO of GitHub), Bob Muglia (Former President of Microsoft), Susan Wojcicki (Former CEO of Youtube), Paul Buchheit (Creator of Gmail), Soleio (Designer of Messenger, Dropbox), Yann LeCun (Chief Scientist, Meta), Andrej Karpathy (Founding Member, OpenAI), Ashish Vaswani (Lead Inventor of Transformer), Amjad Masad (CEO, Replit), and Clem Delangue (CEO, HuggingFace).??
Selected media coverage of fundraising:
IPO Horizon
Unknown. Likely several years away as the company is only 18 months old.
Valuation Analysis
... Continue the rest of this article: