From Hype to Reality: How AI is Transforming Industries

What I learned in a year of venture capital investing in AI startups

Introduction

I work on the investment team at the Venture Capital fund Frontier Ventures. For the last year, I have focused exclusively on AI startups. In this post, I will summarize my most important learnings.

The current AI boom was triggered by the release of ChatGPT 3 in November 2022. Initially, many professional investors were concerned that AI could be another hype like Crypto, a phenomenon with inflated valuations and little substance. Today, I no longer have this fear.

For clarity, when I talk about AI, I mainly mean Large Language Models (LLMs). More traditional AI, such as predictive Machine Learning models (ML), has already proven its usefulness beyond a doubt in areas like lending, X-ray analysis, and default detection in manufacturing. I will also exclude more hardware-focused use cases, such as autonomous driving, robot control, or self-landing rockets.

While Crypto has yet to find a good use case other than Bitcoin’s Store of Value, AI already has a real impact on many industries. The challenge, in my opinion, is twofold: first, the impact that AI has right now heavily depends on the industry, with some industries being disrupted already, others being next, and others far out. Second, AI is not a panacea. It handles certain use cases really well, but there are many others it doesn’t handle well. Hence, one needs to understand the strengths and weaknesses of LLMs to understand what AI can do:

Strengths and Weaknesses of LLMs

The word “Intelligence” in ‘Artificial Intelligence’ can be misleading. It suggests the algorithm is intelligent and can do anything. My argument is that LLMs are not intelligent at all. They are actually quite dumb but they appear smart. They often give answers that look correct on the surface. However, they reveal hallucinations (made-up information) or logical inconsistencies when examined closely. For example, LLMs also often get “stuck” and repeat the same incorrect answer even when explicitly asked not to. From a mathematical perspective, they are stuck in a local minima. The best way to understand the strengths and capabilities of LLMs is to imagine them as “transformers” of text. LLMs excel at taking in text in one format and outputting it in another format. This includes reading through thousands of pages in seconds and extracting exactly the information you want. LLMs can summarize, extract, lengthen, shorten, change the style of a text, change the language, respond in a certain way, and make analogies.

The Various Stages of AI Disruption by Industry

At a high level, LLMs are like conveyor belts for office work. They can automate boring and tedious tasks. Some industries will be disrupted before others. I classify the level of disruption into three buckets: a) already disrupted; b) in the process of disruption; and c) disruption in the future. I will first focus on B2B use cases and end with B2C use cases.

A) Industries Already Disrupted

Industries currently being disrupted include graphic design, marketing, coding, customer support, and education. AI tools in these industries have already seen significant user adoption. This widespread adoption demonstrates the substantial impact AI is already having in enhancing efficiency and driving innovation in selected industries.

In graphic design, almost all designers use AI in their everyday work. A Canva survey of over 4,000 marketing and creative leaders from nine countries reveals that 82% have used AI tools to generate unique images, and more than 500 million images have been created using Canva’s AI image products. The survey highlights that 97% of leaders are comfortable with generative AI, as it reduces repetitive tasks and allows them to focus on higher-level creative thinking and strategic decision-making. AI saves teams significant time, with 69% saving at least two to three hours per week and 36% saving more than four to five hours weekly.

In customer support, AI agents handle the majority of customer support questions. For example, Klarna revolutionized its customer service with a new AI assistant, which handled 2.3 million conversations in its first month, equivalent to the work of 700 agents. This AI improved accuracy, reduced repeat inquiries by 25%, and cut resolution times from 11 minutes to under 2 minutes. Projected to boost profits by $40 million in 2024, the AI allowed Klarna to reduce its customer service team from 3,000 to just over 2,000. (Klarna, February 2024. See here for an interview with Klarna’s CEO for more details.) LLMs are really good at automatically answering simple questions that are directly answered in the manual. It’s more convenient to ask the LLM directly the one question you want to know and get an immediate answer.

Today, AI generates 40% of all code, with GitHub Copilot being the world’s most widely adopted AI developer tool. This tool boosts developer productivity by 55% according to Microsoft (March 2023). Since its business release in December 2022, over 50,000 businesses and 1 in 3 Fortune 500 companies have adopted GitHub Copilot. Developers using Copilot are significantly more productive, with 55% expressing a preference for it, as per the Stack Overflow 2023 Survey.

AI is increasingly integrated into education, transforming how educators teach and students learn. According to recent surveys, 47% of education leaders use AI daily, while 68% of educators and 62% of students have used it at least once. AI tools like ChatGPT are frequently used for generating questions, creating quizzes, and writing lesson plans. The impact of AI extends to the stock market, with non-AI tutoring companies like Chegg struggling. Chegg’s stock dropped over 20% after reporting declining revenue and subscriber growth due to free AI tools like ChatGPT. The stock has fallen nearly 70% in the past year and over 95% from its 2021 peak. Its first-quarter revenue for 2024 fell 7% year-over-year, highlighting the challenges from free AI alternatives. (Yahoo Finance, 2024).

B) Industries in the Process of Disruption

Industries currently in the process of disruption include insurance and sales. Many tools are being rolled out at the moment and are at an intermediate level of adoption, unlike customer support, where selected large companies have already automated 70% of the work. This gradual integration is paving the way for broader AI adoption across these sectors, setting the stage for significant efficiency gains and innovation.

AI is automating the reading and writing of various insurance documents, including communication back and forth. The insurance industry is still incredibly paper-driven and has resisted digitization and automation more than other finance areas like banking. LLMs are the perfect tool to change that. Health insurance is perfect for this because it sits at the intersection of healthcare and insurance, both of which use specialized, hard-to-understand language that needs translation from one domain into the other. This is what AI excels at.

The 2023 State of AI in Sales Survey, which polled over 500 U.S.-based sales executives, found that 95% of organizations are using AI in sales. 84% have used generative AI in the past year, and 97% believe it is important to work with tech vendors that have an AI strategy. Generative AI is particularly useful for sales teams in repurposing messages to different audiences (32%), writing messages to prospects (21%), and getting ideas for outreach messages (20%). Additionally, 86% of sales professionals using generative AI to write messages find it effective. AI tools save the average sales professional over two hours a day by automating tasks like meeting scheduling and data entry, allowing them to focus more on human-centric tasks. AI isn’t replacing salespeople; it’s handling repetitive tasks and providing a starting point for sales content, which many sales pros then edit. AI tools have significantly impacted sales strategies, enhancing productivity and efficiency. (Salesloft, 2023 and HubSpot, 2023).

C) Near Future Candidates of Disruption

Future candidates of disruption include the legal industry, supply chain, and various B2C use cases.

Disruption in the legal industry will differ from what many expect. When people think of AI and legal, they imagine an AI that can answer complex legal questions. Currently, AI isn’t capable of this and likely won’t be for a long time. Instead, LLMs can automate repetitive writing processes that are standardized. These processes can involve either a few very long documents or many short documents that are largely standardized. In these cases, the actual writing of the documents becomes the bottleneck.

An interesting example is legal service providers for immigration law, where numerous long documents must be written as part of a large overall paper application. Two more areas ripe for disruption are Discovery and M&A. In Discovery, overworked junior associates need to review thousands of pages of evidence to find the metaphorical needle in the haystack. In M&A, junior attorneys need to comb through thousands of pages of contracts and other legal documents during the due diligence process when one company buys another, looking for hidden liabilities. Both areas seem perfect for LLMs.

Another area to mention is the supply chain. I haven’t delved deeply into this yet and am not an expert, so I will keep this brief. Supply chain processes are still surprisingly paper-based, with many emails being sent back and forth. Semi-automating the reading and writing of this communication seems another natural use for AI, and the potential seems clear. Logistics are largely paper-based and ripe for disruption. Gartner predicts that 50% of supply chain organizations will invest in AI and advanced analytics applications by 2024. Leading companies like Amazon, Walmart, FedEx, and UPS are already using AI to enhance supply chain productivity and resilience. AI automates manual processes, improves demand forecasting, and optimizes logistics. For example, AI-driven automation can significantly cut costs and improve efficiency in tasks like purchase order generation and shipment booking. (Kognitos, 2023)

The final area I will mention is B2C AI use cases. In the future, it is likely there will be a wide range of B2C AI services built. Obvious use cases include AI physicians, veterinarians, dietitians, personal finance advisors, lawyers, or tax advisors. These services will never be as good as experts but will answer 90% of everyday queries, providing enormous value for many people in standardized situations. Most importantly, they will help the poorest who cannot afford professional doctors, lawyers, financial advisors, veterinarians, agricultural advisors, etc.

The obvious challenges are twofold: first, preventing hallucinations and only giving advice that is correct to a very high level of certainty. This is a major difference between B2C and B2B use cases. In B2B use cases, users are experts and can critically judge the output of an LLM to see if it looks correct or not. In B2C use cases, users are not experts and are much more easily fooled by hallucinations. The second challenge is building a sustainable business that differentiates from the big model providers, such as ChatGPT. Startups need to add a layer of value beyond what the model providers offer. For example, startups could offer a simplified UX that makes it much faster for farmers to use the startup’s app than to manually type questions into ChatGPT’s chat interface.

Conclusion

Different industries are at various stages of disruption due to AI. The release of ChatGPT 3 has shown AI’s substantial impact, dispelling initial investor fears. While traditional AI applications have proven their value, LLMs excel in transforming text for tasks like summarization and extraction. However, they also have limitations like hallucinations and logical inconsistencies.

Currently, AI disrupts graphic design, marketing, coding, customer support, and education by streamlining processes and enhancing productivity. Industries like insurance and sales are integrating AI to automate repetitive tasks and improve efficiency. Future disruptions are likely in the legal sector, supply chain management, and various B2C services.

---