Artificial intelligence is a broad and overused term. Any startup raising money today needs to have a data and machine learning story, and every enterprise is trying to understand how they can unlock the value of their customer and transaction data with AI.
We strongly believe that every application that is being built today is an intelligent application. Applications which don’t use data to build continuous learning systems that provide better value with more data are going to become obsolete in the market as we move forward.
But what do venture capitalists really mean when they say they want to invest in intelligent applications, AI, or machine learning? When we discuss intelligent applications, we are looking for applications that can use AI to improve outcomes by an order of magnitude, and there are three paths that these applications typically follow. Here are the three types of intelligent applications that we are most excited to invest in at Madrona: Automators, Augmenters, and the Avant Garde.
Automators: applications that discover, automate, and integrate workflows
The first category of intelligent applications focuses on identifying repetitive, time-consuming, or difficult processes and creating new ways to handle these workflows in a way that allows customers to focus more of their time on high-value synthesis and cognitive work. This is a cornerstone in the digital transformation that every enterprise around the work is going through currently.
The most well-known companies in this space today are the robotic process automation (RPA) vendors, including UiPath and Automation Anywhere. These companies build software that allows companies to automate individual steps of a workflow, such as opening up a PDF document, extracting key data, entering that data into another system, and combining these steps into an automated workflow.
The RPA vendors have built horizontal platforms, and they partner with systems integrators and consultants to deploy and customize their software at large enterprises. Nevertheless, despite the success of RPA, it is barely scratching the surface of what is possible with AI. With innovations in computer vision, natural language understanding, and other deep learning techniques, there remains a lot to be done with RPA.
In addition to traditional RPA projects, we are also seeing more and more companies build “RPA”-like automation into new products to create end-to-end workflows for specific use cases and industries such as legal services, healthcare, and real estate.
These workflows often combine a primary workflow that has a high degree of automation with “human in the loop” systems designed to deal with fallout and deviations from the automated workflow. Over time, the magic of these products comes from integrating human and machine into one seamless customer experience and using data and feedback loops to continually improve the experience.
Some of the most interesting companies in this space go beyond automating one workflow to automating multiple workflows and creating a new integrated workflow. For example, a company such as OpenDoor has combined the process of valuing a home with the process of closing a real estate transaction and created a new way for consumers to approach the real estate market.
Augmenters: applications that advise or assist humans to perform a job exponentially better
While many companies are focused on automating a set of primary workflows, until we have Jarvis or HAL-style AIs, it will be difficult to automate all of the corner cases and exceptions that can fall out of a process, and this makes it hard to set expectations around automation.
One way to illustrate this is thinking about an “AI” such as Alexa — Alexa is great for one or two key tasks (e.g., turn on a timer, what’s the weather?), but it is hard to use Alexa for 50 tasks because there are too many unknowns, and it’s hard to predict how Alexa will react to unknowns.
For that reason, outside of automating a selection of key tasks and processes, a better use for AI and machine learning is to coach people on how to get better at specific jobs or provide tools that can support someone in completing a job. Even though AI cannot automate many complex cognitive tasks, it can be very helpful in gathering data on inputs and outcomes to help humans improve their performance.
Some of the best examples for AI coaches today are for tasks that are dependent on text and structured data – like writing an email, job description, or product description – but do not require a lot of context or real-time updates to the system. Palantir is an example of a large company today that has taken the approach of combining human analysts with machine intelligence to produce a system that is more effective than either human or machine alone.
We are now also starting to see AI advisors popping up in areas where it has traditionally been difficult to analyze and collect data, such as complex email and chat correspondence or voice and video chats. For example, Gong and Chorus use data from Zoom calls to understand how sales reps could improve their sales meetings and improve sales performance over time.
Over time, these companies will be in the position to build high value datasets and integrate into multiple systems of record and systems of action. Given their understanding of performance and outcomes, they could also become important platform providers themselves.
Avant Garde: applications that create net-new products and experiences enabled by AI
The last category of applications are the ones that create completely new experiences and products by using machine learning. In other words, things that just weren’t possible before the combination of low-cost cloud computing, massive amounts of data, and new machine learning algorithms.
For example, self-driving automobiles create completely new vehicle form factors, business models, and services that would not be possible without ML breakthroughs, and Siri and Google Home’s voice assistants enable completely new interaction models that would not have been possible without advancements in natural language processing.
Many of the companies in this category are pioneers in bringing important new technologies such as computer vision, deep learning, robotics, and NLP to consumers, so it is a very dynamic space to watch because it sits at the intersection of massive markets, novel technologies, and novel business models.
This means the companies can be massive opportunities if they work, but they may also be harder to measure and monetize in the near term. For example, a company such as Adaptive Biotechnologies leverages decades of research on the immune system, next generation sequencing machines, and machine learning in order to detect changes in the immune system to diagnose disease, but it has had to spend time finding the right customers and business models to monetize their technology.
Large pools of capital certainly help in building these “net-new” experiences, and the experiments from big tech companies including Amazon’s Go stores or Waymo’s autonomous vehicles have required big investments. However, there are many examples of startups creating these “net-new” experiences as well, such as startups competing directly with Amazon on cashierless stores or Waymo on driverless vehicles. These companies all require deep tech and market expertise, and the winners will be the ones who can find the right ways to apply new technologies to customer problems.
Intelligent apps are applications that use “artificial intelligence” to create continuous learning systems that deliver rich, adaptive, and personalized experiences for users. While these intelligent apps have a wide range of different customers, partners, and builders, the most compelling companies are typically Automators, Augmenters, or Avant Grade that can demonstrate order of magnitude improvements in business metrics.