This is the third part of our three-part series on the evolution of software. If you haven't read the first or second part, we recommend starting there.
Over the last 20 years BuildGroup’s team has developed a deep understanding of the infrastructure that shapes the way we interact with tech today. In this series of blog posts, we dig into the six key trends that we believe have defined the evolution of software and how AI fits within this context. This is part three of three and picks up right where part two leaves off. We recommend reading the series in order. Click here to start with part one or part two.
If you’d like to read an AI-generated summary of this post, scroll all the way to the bottom (thanks, Claude!).
A quick pause before we get into data science. The first four infrastructure waves really existed below the application layer. What does that mean? They enabled the creation of the SaaS applications that we know today, but they did not necessarily impact the functionality of the application that rested upon them. The fact that the SaaS application you use runs on Amazon and is based on open source makes it faster, more reliable, etc, but it doesn’t make the functionality of the application itself a step function better. When Microsoft began to offer PowerPoint on the cloud, the actual experience did not dramatically change compared to when it was downloaded to a local server.
That began to change with the next wave – data science. The primary use case for SaaS has been, and continues to be, workflow. Users depend on these platforms to do their jobs. Salesforce in sales, Marketo in marketing, and Workday in HR are all examples of workflow SaaS. Even machines depend on workflow solutions to do their job, often interacting with other “workflow” applications via an API to get their work done. These workflow SaaS applications proved to be very sticky with their users because of their centrality to their jobs.
As users began to adopt these workflow applications en masse, huge amounts of data began to be created from those interactions. These applications began to have the capacity to understand how their users worked and what they needed based on their interactions on the platform. The best SaaS companies recognized this and started setting up the systems to collect, analyze and present this data – what became known as “data science”. This led to a transition in SaaS, from workflow SaaS to data-enabled SaaS. Rather than work on the next workflow feature, data-enabled SaaS companies began using that data to surface insights back to their users.
That data could either be bespoke to the one customer based on their own data, aggregated with overall data from other users on the platform for benchmarking, or both. MailChimp, for example, analyzes email campaign performance, including: open rates, click-through rates, and conversion metrics. Mailchimp might suggest the best times to send emails or recommend changes to email content that could improve engagement, based on analysis of past campaigns and industry benchmarks. Suddenly, SaaS applications were suggesting where a user might want to focus their time to achieve a better outcome.
In addition, data-enabled SaaS companies now had the capability to become true platforms. A platform is nothing more than a SaaS solution that sits at the center of an ecosystem of other data-enabled SaaS platforms, where all parties benefit from the integration. Data is the fuel for those integrations. Without data, what is there to integrate? Take the Salesforce example again. Salesforce has built a tremendous platform by allowing applications to access the customer and sales data that exists in its platform. As a result, there are now literally thousands of applications built on Salesforce that companies depend on for sales and customer support. As a result, Salesforce is even more sticky than it would have been if it had simply remained a workflow app.
The “workflow SaaS to data-enabled SaaS” journey has been a critical focus for BuildGroup and our portfolio companies. Fiix is a good example of this transition within the BuildGroup portfolio. As I mentioned earlier, Fiix started as a workflow SaaS application for maintenance departments of manufacturers. Around 2017, the company recognized the opportunity to begin using the data within its systems to help their users make better decisions.
The first part of the journey was recognizing that they had not planned for the large scale collection and analysis of data. They needed to make an investment in getting their systems and data in a position where they could analyze them in real time to offer insights. Many workflow SaaS companies were in the same position and rushed to catch up. Some did not, and got left behind. Fiix made it, and in 2019 began rolling out its insight products.
One such product was a parts predictor that let maintenance departments know when they were likely to run out of supplies based on their own historical data and future projections about usage. For most of their users, this was a capability that they never had before. It led to a huge leap in product usage from customers using those insights, as well as interest from companies who might not have considered Fiix had it just been a workflow SaaS solution. And now with data available to share, Fiix began pursuing partnerships with other solutions serving manufacturers who wanted access to Fiix’s data.
Data science heavily impacted how applications are consumed. Not in the same sense as mobile, which literally expanded the locations where a user could interact with SaaS. Rather, it changed user expectations around usability – they now expected to be getting data to make better decisions, not to simply manage workflow. Users expected their workflow applications to be smart. If a SaaS solution had not adopted data science in its efforts and transitioned to data-enabled SaaS, they risked getting left behind. The waves continued to build on each other. But with the unexpectedly quick rise of artificial intelligence, in particular generative AI, those workflow SaaS companies who hadn’t data-enabled themselves really risked getting left behind.
With the launch of ChatGPT late in 2022, the world quickly became familiar with AI, or more specifically, generative AI. There is no need to share a lot of history here, as we are all living it in real-time. What has been impressive is how quickly the technology has integrated itself into applications in a way that substantially enhances their usability. It builds upon every wave that went before it, and in particular, the transition to data-enabled SaaS, to build a fundamentally better and more usable offering for customers.
For AI to be successful, it needs two things: data and expertise. Data is required for the models to consume and generate their output. For companies trying to use AI for competitive advantage in their offering, access to proprietary data is paramount. Otherwise anyone could take AI and apply it against publicly available data sets to generate a similar result. In addition, you need experts to train these models. Not experts in AI, but these are of course important. But experts in the use cases the models are being trained for: whether in the manufacturing, customer service, or legal spaces, for example.
Understanding the key questions that AI must answer for these users requires a deep knowledge of their needs. This is no different than the domain expertise required to build a successful workflow SaaS or data-enabled SaaS company. But it's even more important as “prompt engineering” – the process of interacting with a model to train it to deliver a better and more relevant result – is best done by experts in the field they are trying to serve.
Who is best prepared to adopt AI in this context and make it successful quickly? The very same data-enabled SaaS companies who rode the last wave. Data-enabled SaaS companies have tons of data, and because it is generated as a result of users operating on their platform, it is proprietary. And because these data-enabled SaaS companies have domain knowledge around the field and the customers they are serving, they are best positioned to train the models to deliver results that are relevant to their customers.
Incumbent data-enabled SaaS companies have a natural advantage over startups due to the need for large amounts of data and expertise, which new entrants will take time to build. Take Casetext, a previous BuildGroup portfolio company, for example. Casetext began as a workflow SaaS company enabling a better research and brief drafting solution for lawyers. The company collected public and private legal case law and legal research, and made its searchability much easier, simpler and relevant than incumbent solutions from Westlaw and LexisNexis. As their solution grew, they started adding legal drafting capabilities to their offering. The company understood that AI was coming, and very early on, Casetext made sure it was prepared by building the team and infrastructure to support the technology when it was ready for the market.
In the meantime, they became experts at how lawyers did their work. And that expertise paid dividends when GPT-4 was released. The company quickly integrated GPT-4 into their offering, and suddenly Casetext became a powerful tool that could replace much of the “rote” work of lawyers.
New legal AI startups did not have the data and expertise advantages that Casetext had earned by already being in the market for years, as well as experimenting early with AI in their offering. We expect to find many examples of this type of opportunity throughout the world of SaaS, and are equipped to help founders navigate this terrain because of our lived experience with companies such as Casetext.
A final word on the impact of AI on usability: I believe it will be profound. Think back to the early days of search. One day, we all went from this:
To this:
Yahoo expected users to do much of the navigation through their menus. Yes, there was search but many of us used the Yahoo “workflow” to check news, weather and email. In some ways, it was just an extension of the way we searched for information offline in a newspaper, magazine or library. Search was there, but it wasn’t necessarily the main focus.
Google, on the other hand, was 100% focused on search. There was no “workflow” on their home page. They did it behind the scenes for you, based upon your search. I remember being shocked the first time I saw it. How could something so simple be as good? But my best guess is that I stopped using Yahoo altogether (except for mail) within a short period of time. As did much of the world, as evidenced by Yahoo’s decline and Google’s rise.
We are going to see something as dramatic occur in the world of SaaS. We are going to go quickly from this:
To this:
Humans won’t need to manage the workflow, AI will do it for them. They won’t need to monitor either as AI will take notice when something needs to be addressed. Not sure where you need to focus? Just ask your AI co-pilot. Going offline for a bit and worried about managing your factory while you are gone? Have a conversation with your AI about what you need it to monitor and take action on while you aren’t available. It will be like having a super employee at our fingertips anytime we need them.
Within SaaS companies, the workflow to data to AI paradigm will become a self-reinforcing virtuous cycle:
Workflow will continue to generate valuable and proprietary data. That data will be used to surface insights and opportunities for action back to the user. That same data will attract partners who want to integrate into the platform and workflow, generating even deeper workflows and more extensive data. And AI will be layered on top to initially co-pilot with the user within the workflow – but eventually abstract them away from the workflow and the data, and transition the user relationship to the AI chatbot.
It's an exciting time to be investing in AI-enabled SaaS companies. So many early growth companies have built great data-enabled SaaS offerings that are ripe for AI enablement. While it's motivating to think about the opportunity to partner with these companies as an investor, it's even more exciting to think about the great offerings they will bring to market. Ultimately, technology is about changing the world with continuous and valuable innovations. And the next wave is just beginning.
The post discusses BuildGroup's investment thesis, which is centered around understanding the major infrastructure waves that have shaped the evolution of software and SaaS companies over the past 25 years. It identifies six major waves:
Each wave had a profound impact on how software is built, deployed, distributed, and consumed. The document provides details on how companies that were able to ride these waves early on gained significant advantages.The author believes the latest AI wave will have the greatest amplification effect. Data-enabled SaaS companies that have already amassed large proprietary datasets and domain expertise are best positioned to successfully integrate AI and become "AI-enabled SaaS" companies.
The document uses examples like Fiix, Casetext, and Anthropic to illustrate how companies transitioned through these waves. It argues that AI will dramatically improve software usability, abstracting users away from managing workflows to simply querying an "AI co-pilot." Overall, the thesis is that investing in early-growth, data-enabled SaaS companies ripe for AI integration represents an exciting opportunity to back the next transformative wave in software.
Enjoying this series? Subscribe to our newsletter for more Building Blocks from BuildGroup.
The information in this blog post is provided in good faith without any warranty. It does not constitute investment advice, recommendation, or an offer of any services or products of BuildGroup Management and it is not intended to provide a sufficient basis on which to make an investment decision. This document is provided for educational purposes only. Discussions of current or former BuildGroup portfolio companies are intended for educational and discussion purposes only. Any portfolio company so discussed has been selected based on objective, non-performance based criteria.
