By Malik Nabulsi, MBA Summer Associate
“The artificial intelligence problem is taken to be that of making a machine behave in ways that would be called intelligent if a human were so behaving.”
— 1955 Dartmouth Research Project Proposal
Artificial Intelligence (AI), as a concept, is decades old. But, only now are we seeing real-life applications of this science fiction-like technology. Officially coined in 1955, AI has gone through several periods of hype and stagnation over the last half-century. This fluctuation is in part due to false starts in needed infrastructure pieces (data and processing chips) and machine learning (ML) methodologies (how computers learn from that data). But, last year’s record $6 billion in venture capital investment in AI/ML startups, as well as recent breakthroughs in deep learning, point to a single consensus: AI is here for good. How and when businesses adopt AI in their operations will be central to staying competitive in the “second machine age.”
During my MBA summer internship at Lerer Hippeau, I chose to dive deeper into the key enablers of AI adoption and to separate true venture-scale opportunity from the noise in the current funding environment. My perspectives on defensibility in the manufacturing and B2B sectors, informed by five years of advising and investing in these industries, were widened during my research and conversations with AI experts.
I set out to define AI and machine learning technology, explore the relationship between investment and adoption, and to provide investors a framework for achieving venture-scale outcomes in the space. Here are my takeaways.
Artificial Intelligence: Goals, Applications, Drivers
Artificial intelligence can be broken down into a set of machine tasks (or goals) and the techniques to achieve those goals.
AI aims to accomplish a variety of goals, ranging from identifying an object in an image, to powering autonomous vehicles and applied robotics. One of the most prevalent techniques to achieving these goals is machine learning (including deep learning), or a field of computer science that enables computers to learn without being explicitly programmed. Companies have been using AI since the early 2000’s to enhance workflow applications (e.g. InsideSales) or optimize product recommendations (e.g. Amazon), but recent breakthroughs in deep learning have ushered in a new wave of companies that are delivering AI as their core value proposition.
“AI-centric” and “AI-enabled” solutions are the most advanced applications of artificial intelligence technology today. AI-centric solutions solve lower-risk use cases and are built entirely around AI. In other words, if the underlying AI were to fail, these applications would be commoditized workflow solutions at best. X.ai, a Lerer Hippeau portfolio company that automates appointment scheduling, is a great example of an AI-centric company.
On the other hand, AI-enabled applications are only possible given recent breakthroughs in machine learning. These solutions use AI to help predict future events or outcomes. For example, portfolio company Augury leverages its proprietary database of machine vibrations and acoustic signatures to predict industrial equipment failures before they occur.
There are three key drivers behind the inflection point we’re seeing in AI’s evolution:
- Data Proliferation: Massive amounts of digital data are being generated with the rise of IoT and increasingly digital services.
- Faster Hardware: The processing chips that enable machine learning computing are becoming exponentially more powerful at much lower costs. A processing chip in today’s average gaming PC would have classified as the world’s most powerful supercomputer in 2002.
- Better Algorithms: Open-source frameworks such as Google’s TensorFlow have made available powerful machine learning algorithms to developers around the world, accelerating the development of more advanced use cases.
What these enablers have unlocked is a much wider scope for AI’s transformative potential. Businesses can project growth with increased accuracy given advancements in real-time forecasting. Manufacturers can produce output more efficiently given optimization of workflows and inputs. Marketers can promote campaigns with higher conversions using personalization and targeting. Brands can provide more enhanced user experiences through automation and convenience.
Investment vs. Adoption
In response, investment in the space has been on a record growth trend over the last ten years. VCs invested $6 billion across 643 U.S.-based AI/ML startups in 2017, 12 times the amount invested in 2008. Still, this amount is only a fraction of the $20–30 billion that McKinsey estimates was spent by Big Tech, including Google and Baidu, on internal R&D in 2016. However, a closer look at AI adoption rates across all industries reveals a surprising contradiction to the pace and magnitude of global investment. Based on BCG survey data, only 16% of companies in the largest global economies have implemented AI in one or more use cases.
Unpacking the reasons for this gap is the first step in understanding where venture-scale opportunities exist within AI/ML:
- The Productivity Paradox: The first reason we see a gap relates to the classification of AI as a “general purpose technology” (GPT), like electricity or the internet. Earlier this summer, Chad Syverson, an economics professor at the University of Chicago, presented on this topic to the Dallas Federal Reserve. Specifically, he explains that GPTs require complementary assets to be invented and installed before adoption and productivity can really take off. In sum, AI can take years, if not decades, to standardize because stakeholders must first understand and apply the technology to their specific business case.
- Horizontal (vs. Vertical) AI: The second reason relates to where investment capital has flowed to date. A majority of investment in AI has gone towards horizontal AI, opposed to vertical AI. The former seeks to democratize machine tasks across a variety of use cases. Examples include Google’s DeepMind or startups seeking to advance natural language processing. The challenge in this segment today is that startups increasingly face a price and data disadvantage relative to Big Tech’s alternatives. More importantly, horizontal AI doesn’t customize solutions and workflows around the customer’s core problem.
The combination of business adoption requiring use-case specific solutions and the fact that most investment to date has gone to horizontal AI creates an exciting window for investment in vertical AI solutions.
By identifying the sectors which have the most to gain from AI adoption, and partnering with founders that seek to solve priority industry problems with AI/ML, investors can much more effectively identify venture-scale opportunity in the space.
Most promising sectors
Investors should consider the status quo profitability (and productivity), total addressable market (TAM), and the quality of the industry’s digital infrastructure to best identify investment opportunities in AI. To visualize this across the S&P 500, I mapped each industry according to three variables in the graph below:
The key takeaway here is that energy and healthcare, both relatively large TAMs, are most likely to be disrupted by AI as the quality of their digital infrastructure (“data-spheres”) continues to improve.
Healthcare: U.S. health expenditures were $3.3 trillion in 2016, representing a massive opportunity for AI-enabled cost savings across the sector. Per McKinsey survey data, early adopters within the vertical expect AI to raise operating profit margins by five percentage points by 2020 through automating labor-intensive workflows.
- Recommendation: Significant runway exists within healthcare for early-stage companies that de-risk and drive efficiencies in the $65 billion drug discovery and clinical trials market. Accelerated drug pipelines are hyper critical for pharma companies, and investors should expect to see M&A activity intensify as large players seek to mitigate structural inefficiencies in scientific discovery, clinical trial enrollment and medical adherence.
Energy: U.S. energy expenditures were $1.0 trillion in 2016, representing another promising opportunity for AI disruption. Recent advancements in smart grid, IoT and energy storage technologies aim to further digitize an industry that has historically struggled with data availability and volatility.
- Recommendation: The domain with the largest potential impact from AI initiatives is the oil and gas sector, given the capital-intensive nature of the industry, as well as the ability to now combine geological data with historical production data. Goldman Sachs estimates that a 1% improvement in inventory and production costs driven by AI would result in $140 billion in savings over a ten-year period.
Vertical AI business models poised for success
Data aggregation and unification are no longer enough to build to a long-term competitive advantage in AI. The rise of AI-centric and AI-enabled applications calls for a paradigm shift in how investors assess product defensibility. Accordingly, investors should key in on the following drivers of value:
- Full-stack Product: Does the company offer a full-stack, fully-integrated solution to a headline industry problem? By owning the full data value chain from the user interface, to the underlying machine learning models and functionality, vertical AI startups can more efficiently improve their algorithms and command better pricing power.
- Domain + ML Expertise: Does the founding team include a combination of subject matter and data science experts? This partnership is especially critical in higher-risk use cases, such as healthcare and energy, where scale requires trust between customers and the management team.
- Proprietary Data: Has the company established a data asset that compounds in value and is difficult to replicate? This factor is the most critical for defensibility and includes data acquired to train the machine learning models, as well as ongoing data streams from the customer.
- AI-First: Does AI/ML drive the core value proposition? The underlying AI/ML technology must be central to the value created for the customer, versus serving merely as an optimization layer or feature.