These AI Applications Are Highly Favored by VCs This Year
-
2023 is coming to an end with less than two months remaining.
In the current wave of AI, which AI applications are the most promising and worth investing in? On this point, the investment community appears to have reached a consensus.
Recently, an AI-focused news platform, AIbeat, compiled a list of the top 10 highest-valued AI startups globally.
If one filters the companies on the list, they will find that more than half of the products belong to the same category of AI applications.
In the list, whether it's ChatGPT, Claude, Inflection AI's Pi, Jasper (focused on writing functionality), or Cohere (focused on enterprise services), all can be classified as AI assistants.
However, in the current generative AI landscape, these applications, which primarily feature natural conversation capabilities, are showing an increasingly obvious trend of homogenization.
The same question can be answered by ChatGPT or assisted by Claude.
Moreover, apart from a few leading companies, most teams struggle to establish high technical barriers.
Given this, why are such applications still viewed favorably by the investment community?
1
The Significance of Personalization
Regarding why VCs are optimistic about AI assistants, we can find an explanation in the funding process of the startup Writer.
In early September, Writer, an AI startup providing full-stack content creation solutions for enterprises, announced the completion of a $100 million Series B funding round, with a post-money valuation exceeding $500 million.
Its main product is a B2B AI assistant called Writer.AI. This funding round was led by ICONIQ Growth, with participation from WndrCo, Balderton Capital, and Aspect Ventures.
Among them, ICONIQ Growth holds a highly respected position in the investment world, known for successfully investing in many well-known companies such as Tencent, Zoom, Send Bird, and Flip Kar, making its opinions highly representative.
When discussing its investment philosophy, ICONIQ Growth mentioned: Companies should initially focus on new customer growth. However, once the company's ARR (Annual Recurring Revenue) reaches a certain level, excessive focus on new customers can increase uncertainty and lead to churn. Therefore, after ARR grows to a certain amount, companies need to focus on maintaining existing customers.
In terms of customer retention, Writer has shown a commendable performance.
Over the past two years, Writer has achieved a 10x revenue growth and maintained a net revenue retention rate of over 150%, while serving hundreds of major enterprise clients including Spotify, L'Oréal, Uber, Handshake, Hubspot, and Deloitte.
The key to Writer's high user retention lies in its core product's customizable and personalized features. Specifically, Writer's knowledge graph seamlessly integrates with clients' critical information sources and documents. This means Writer not only accesses and consolidates key enterprise data but also delivers "tailor-made" insights when answering questions, analyzing data, researching business, or creating summaries.
This connectivity and integration capability ensures that the generated content aligns more closely with the business needs and regulations of enterprises.
Similarly, this "personalization" trend is evident in other AI assistant applications. For instance, Otte.AI, an AI tool specializing in voice transcription, provides targeted analysis and suggestions based on different voice conversations. Additionally, using speech recognition and diarization technology, Otte.AI can identify individuals by their voices. Once a speaker is detected, the system creates a voiceprint profile for them, enabling the identification of all their subsequent speech.
Beyond Otte.AI, RewindAI is another example of success through personalization. Its core feature is offering a "memory assistant" capability.
RewindAI automatically records all information on a user's phone or computer (with consent) and supports review, retrieval, and summarization. It captures content viewed in Safari browsers, imported screenshots, and more, providing browsing and search functionalities. Users can quickly navigate past content by scrolling through a timeline.
All these factors demonstrate the prevalence of personalization in current AI assistant applications. This 'tailor-made' approach for users undoubtedly enhances user engagement and extends usage time. Moreover, personalization implies 'a thousand faces for a thousand users,' significantly expanding the application's coverage and raising the ceiling for user scale.
2
Where Are the Barriers?
Beyond personalization, the most significant strategic advantage of AI assistants lies in the proprietary data barriers built through continuous interaction.
During the mobile internet era, whether it was Baidu, Tencent, Google, or Amazon, none placed as much emphasis on or utilized data as extensively as they do today.
At that time, the prevailing view among major players was that only users and traffic were the most critical factors. Whoever could spend more money to capture more users would achieve greater economies of scale.
Under the ruthless logic of 'traffic is king' and 'scale is king,' the entire market was a zero-sum game. An increase in users for Platform A often meant a decrease for Platform B.
In this battle for market share, many platforms often find themselves competing in overlapping niches, forcing them to fiercely compete in applications, features, and content. When competition in applications, features, and content reaches its peak, the entire logic of the internet struggles to progress further.
The emergence of AI assistants has brought a turning point. By continuously adjusting and optimizing user behavior data in real-time, AI assistants can accumulate unique proprietary data for each user. The diversity and individuality of human beings ensure that the niches built on this proprietary data will not be overcrowded, hyper-competitive, or zero-sum games.
This is because AI assistants target specific individuals or businesses composed of different individuals, rather than abstract labels or groups defined by algorithms. This difference is primarily determined by the distinct technical characteristics of AI compared to previous-generation algorithms.
Unlike AI assistants, previous-generation algorithm systems relied on limited datasets for training and often depended on manually designed features for recommendations. However, these features might not fully capture the complexity and diversity of user behavior.
For example, if a user is male, a video app might predominantly recommend military or political content, even if the user rarely engages with such videos, or it might repeatedly suggest videos with the same tags. In contrast, AI assistants excel by using deep learning and reinforcement learning technologies to collect and process user behavior data in real-time, continuously optimizing to capture nonlinear, more complex, and nuanced data features.
In this way, users are liberated from broad and general labels, becoming 'unique' individuals.
Even if some AI assistants have overlapping functionalities, the presence of proprietary data means that users will gradually adapt to and become accustomed to the AI assistants they have been using, making them less likely to switch to other applications.
Therefore, AI startups built on such proprietary data will have stronger vitality.
3
Possible Challenges
Given the immense popularity and bright prospects of AI assistants, what challenges or issues might they face?
On this point, the AI unicorn Jasper serves as a cautionary tale with its painful experience. As one of the first companies to enter the AIGC field, Jasper quickly gained favor from investors and achieved a very high valuation.
As early as 2022, Jasper had amassed 1 million users. In October of that year, the company secured $125 million in Series A funding led by Insight Partners, reaching a valuation of $1.5 billion and entering unicorn status.
However, the good times didn't last. In February of this year, Jasper projected annual revenue of $140 million, but by summer had revised this estimate downward by 30%, followed by layoffs in July.
More recently, Jasper has reduced its internal employee stock valuation by 20%.
Many analysts attribute the struggles of AI assistants like Jasper to their lack of proprietary models. To date, Jasper still relies on ChatGPT's API interface, making it essentially a "wrapper" application.
Without proprietary models, these platforms cannot directly access or analyze user data, making it difficult to implement targeted adjustments and optimizations.
This raises the question: Can AI assistants without proprietary models like Jasper still find their ecological niche?
As mentioned earlier, while one of the core advantages of AI assistants is their personalized customization capabilities, such customization can be divided into passive and active types.
Passive customization refers to AI assistants that can automatically collect, analyze data, and perform adaptive learning without requiring user configuration.
Examples in this regard include Pi developed by Inflection AI and the previously mentioned Rewind.AI. Active customization refers to AI assistants that require users to set up and customize according to their own needs and preferences.
For various open-source AI assistants or those using third-party models, the active customization approach is clearly a better alternative in the absence of proprietary data.
For instance, Polyglot is an open-source AI-powered language training platform client that helps users practice speaking skills in multiple languages. Using AI technology, Polyglot provides personalized speaking training suggestions and feedback.
Users can select the language and difficulty level based on their needs and proficiency. Polyglot then offers real-time feedback and suggestions on pronunciation, grammar, and vocabulary to help users improve their speaking skills.
Another more prominent example is the well-known Poe.AI.
In this application, which resembles an App Store for large language models, users can actively customize AI chatbots with different personalities, identities, and functions to make the AI more personalized and aligned with their expectations. Although all models in Poe.AI come from third-party APIs, users' personalized needs are still met through this 'user-created' functionality.
4
Summary
If we were to discuss the greatest commercial value of AI assistants, it would be their role in previewing the 'iPhone moment' for artificial intelligence.
Since the rise of the current wave of AI, countless businesses, venture capitalists, and investors have been contemplating and searching for AI's 'iPhone moment.' Looking back, we can see that AI assistants have already laid nearly all the groundwork required for such a moment.
In 2007, Apple introduced the revolutionary iPhone 1, which combined a touchscreen, camera, music player, web browser, and other functionalities into one device. It transformed how people communicate, entertain themselves, live, and work, ushering in a new era of mobile internet.
Similarly, as an integrated application, AI assistants can interact with users through voice, text, images, and other means, providing services like search, reservations, writing, and Q&A to meet various user needs and scenarios.
Going even further than the iPhone, today's AI assistants can continuously learn and evolve, improving their relationship with users. Some AI assistants also offer human-like companionship and emotional functions, transcending the role of a mere tool to become an inseparable 'intelligent companion.' Currently, many AI assistants excel in individual functionalities, but these features remain scattered and unintegrated.
With market feedback and development, if an ambitious company takes the crucial step forward, a 'masterpiece' that combines the strengths of all existing AI assistants will emerge.
At that point, the true 'iPhone moment' for artificial intelligence will arrive.