S32’s Wesley Tillu: Investing in the Age of AI Acceleration
Contents
ABSTRACT
🌐
AI adoption is accelerating within enterprises, creating a growing opportunity for application software companies. Wesley Tillu, Partner at S32, details how application AI startups are differentiating themselves from cloud and incumbent software providers by establishing ownership of the user experience and leveraging proprietary data to continually improve their models. Simultaneously, end users are becoming invaluable champions of adoption at the enterprise level.
KEY POINTS FROM WESLEY TILLU'S POV
What makes this acceleration phase of AI adoption such an important factor for AI applications going forward?
As enterprises transition from AI pilot projects to full-scale production use cases, application software companies will experience accelerated growth relative to legacy platforms. The transition of AI technologies from experimental pilots to full-scale enterprise deployment marks a significant evolution for AI in the enterprise, not only in terms of the maturation of AI capabilities but also in their readiness for widespread operational use and adoption. “Following innovation at the model and infrastructure layers over the past couple of years, application AI software will experience growth across industries, streamlining workflows for many enterprise users,” says Tillu. “Going forward, there will be significant opportunities for startups to capture revenue across the entire enterprise software stack.”
The consumerization of AI will drive enterprise adoption of applications that empower end users in their personal lives. “Many large tech companies began as consumer-focused businesses and grew because their eventual enterprise users were originally enthusiastic personal users,” adds Tillu. “For example, the widespread personal use of Gmail eventually led many enterprises to adopt Google’s full suite of products.” Much like Google and Apple, which started as consumer companies and became top of mind for enterprises, tomorrow's AI companies will need to have the individual user and consumer in mind. “There are many use cases of AI that will require the individual user to be the product champion within the organization, such as in code development, sales and go-to-market, and 3D design. The extent to which AI applications can focus on the end-user experience and make an impact will become a critical driver of their ability to succeed in enterprise sales.”
Moreover, focus on the user experience will ultimately enable AI application companies to build a competitive advantage and progressively improve performance. “The moat for AI applications is the user experience—the ability to create differentiated, engaging interactions for the user, even though the platforms may rely on the same foundational models as other applications, whether closed-source models from OpenAI, Anthropic, or Cohere, or one of the available and capable open source models.” Establishing these data moats enables AI applications to leverage data from user experiences to continually improve model performance through reinforcement learning from human feedback (RLHF). The more data a company has, the better it can refine their models.
What are some of the use cases or business models that may be associated with this category?
Startups developing AI applications are proving their ability to disrupt the end-to-end customer journey for numerous industries, spanning game development to energy optimization. Using data to create powerful user experiences requires not only controlling as much of the user interface as possible but also implementing effective mechanisms to consistently gather feedback:
An example in the gaming industry is Inworld AI, an S32 portfolio company that specializes in creating an AI engine for gaming, the metaverse, virtual reality (VR), and augmented reality (AR) environments. “Inworld uses AI to enhance both the game developer and player experiences,” says Tillu. “They have developed proprietary technology based on the latest AI models and offer a robust developer platform. As companies launch new games, Inworld transforms the consumer experience while capturing feedback from users to improve the quality of the characters' interactions.” Inworld's platform includes features that allow for the collection and analysis of user feedback on AI characters. This feedback mechanism can include rating character responses, providing input on dialogue realism, and other interactive elements that help developers refine and improve the AI characters' behavior and responses over time.
For Phaidra AI, an S32 portfolio company that develops AI technology for optimizing industrial systems in data centers, capturing end-to-end user feedback is integral to enhancing the performance and efficiency of their AI platform. Phaidra's approach involves deploying AI to autonomously control and optimize complex cooling systems, allowing for real-time adjustments that improve energy efficiency and operational stability. This setup allows Phaidra to continuously gather feedback from operational data, which directly informs their AI models to improve decision-making and predictive accuracy. In doing so, Phaidra not only optimizes the physical operations of these facilities but also refines its AI algorithms based on actual performance metrics, ensuring that the system evolves to meet the changing needs of the industry.
Today, many AI startups are selling to enterprises by showcasing value through enhanced efficiency and time savings for employees. In the long term, AI agents or agentic systems may replace entire workflows and even reduce headcount within organizations. Many AI applications are finding success currently by demonstrating return on investment (ROI) in numerous industries by improving performance and accelerating project velocity as an augmentation of human-in-the-loop workflows. However, as AI agent software demonstrates the ability to automate larger portions of enterprise workflows, the hurdle for winning enterprise budgets will increasingly emphasize replacement, rather than augmentation, of specific workflows. Two examples of this ‘arms race’ to prove ROI include knowledge worker co-pilots and cybersecurity tools:
Enterprise co-pilots will differentiate through their ability to replace the tasks they currently augment. “Successful agentic systems provide real and meaningful value in terms of the time savings and efficiency boosts to users today. This is the first phase of proving return on investment to enterprises,” says Tillu. “The second phase will ultimately be automating away tasks so that there’s no need for a human in the loop. Of course, certain tasks will be automatable far sooner than others.” Ema.co, a S32 portfolio company, provides a universal AI employee for different personas within an organization, such as data analysts, customer support analysts, and HR analysts.
In cybersecurity, applications must continue to automate away lowest-tier tasks both in the broader cybersecurity category and in security for AI. “In cybersecurity, one of the most immediate ways to achieve headcount reduction in the cybersecurity space will be by automating away some of the tasks of tier 1 analysts.” Innovative companies are enabling security analysts to focus on more complex tasks or even obviating the need for extensive tier 1 staffing, enhancing efficiency and redirecting human resources to higher-level security challenges.
What are some of the potential roadblocks?
Cloud and incumbent software providers are continually embedding AI into their product suite, increasing competition for startups at the application layer. While infrastructure services such as storage and compute remain growth drivers, incumbents like Microsoft are now integrating more sophisticated AI-driven tools and applications such as Copilot 365 and Dynamics 365 directly into their platforms. This progression allows them to not only retain customers by offering a more comprehensive suite of products but also to attract a broader client base looking for end-to-end solutions. Without differentiated ownership and access to user data, startups are at a tremendous disadvantage to cloud providers. “Startups must be very strategic to ensure their business models don't fall into what we internally refer to as the "zone of commoditization,” says Tillu. “They need to evaluate whether major cloud providers such as Amazon, Google, or Microsoft could offer a similar product for free to acquire users and upsell through their platforms. To avoid this zone, startups must differentiate by creating a unique user experience or owning a customer base that incumbents cannot easily access.”
IN THE INVESTOR’S OWN WORDS
Wesley Tillu
Many think AI has reached its peak, but with new model architectures, the continued explosion of data volumes, and accelerating enterprise adoption, we are still in the early stages.
Following significant research advancements in the last five years and the release of OpenAI’s ChatGPT in late 2022, many enterprises in 2023 began to carve out budgets for pilot projects to explore and assess the potential of AI. Now, the focus has shifted to proving the ROI of these initiatives. Production deployments of AI have steadily grown this year and we expect large scale enterprise AI deployments next year.
As active cyber investors, we are especially excited about the potential for AI to disrupt the security operation center (SOC) and enhance security for AI – a growing subcategory as AI models move into production.
Another area with opportunity for AI to demonstrate value and disrupt existing markets is in service-heavy industries, such as consulting, legal, finance, or support sectors. Through increasingly agentic systems, AI can demonstrate immense value and reshape these industries.
Both enterprise and consumer are ready to adopt deep tech for consumer technologies as interest and spending in the space has become greater than ever before...
Over the last decade, the first wave of vertical software adoption transformed mainstream industries, including restaurants, home services, and salons....
