Study: Constellation Software Acquisition Runway
Vertical Software Protection
One of the frequently cited issue with Constellation Software (CSU) strategy of vertical SaaS (vSaaS) acquisitions is: How long can this last?
Especially in this era when AI is lowering the cost of developing software. Will increased competition reduce opportunities?
While it's true that AI has the potential to disrupt software, we argue that it will hurt horiztonal SaaS (hSaaS) but not vSaaS.
Think about what tasks you're using AI for now:
Data discovery on large sets of documents and data.
Editing, writing, note taking, research.
Basic data analysis through horizontal software like Excel.
Call centers, customer support.
Coding, math, simulation etc.
There's a common theme: If the task of a product can be replicated by prompting AI, then it's in danger of extinction.
AI is absolutely encroaching horizontal use cases, but they’re terrible at vertical markets. And they always will be.
Why so?
Because vSaaS has some interesting protections.
Domain Complexity
Toast is a vSaaS for restaurants and it understands the domain deeply. The software is not just about taking orders and making payments.
It understands the UI for the kitchen is different from the waiters, labour scheduling for weekday lunch is different from Saturday nights, backend integration with accounting module…
Can AI help a restaurant operator? Yes, only with generic advice like making reservations, but not running the entire show.
Integration
The construction industry vSaaS is a good example of how important software integration is.
The dynamics of managing a construction project is a nightmare. In big projects, you can employ up to thousands of workers. Some of them have unions, some different insurance contracts, pay schedules, work locations… And work doesn't start until someone gets paid. Each day delayed on a construction yard is very costly.
So the vSaaS must integrate entirely into the whole database, including compliance and regulations. These are very industry specific, which is exactly where vSaaS creates value.
Again, AI can create code, but it can't navigate the complexities involved with integration.
Trust
This is the common reason why vSaaS is so sticky. When it works and employees are trained on it, the software gains trust.
Don't underestimate trust. Suppose you run a hospital, if patients data are not accurately retrieved, or medicine compliance is not up-to-date… the consequences are enormous.
It's not just financial loss, people can get hurt and reputation can be destroyed. If that happened, the entire business is finished.
ChatGPT Health
Let's examine a counter example: OpenAI launches ChatGPT Health letting users upload health records and provides advice.
But notice what they are not doing?
ChatGPT Health is not trying to run a medical enterprise. It did not build an insurance claims module, nurse scheduling system, lab tests directory…
What they went after is the consumer health information layer. This is a horizontal layer sliced across a deep vertical.
Incumbents vSaaS providers like Allscripts are protected because of:
Deep integration with existing legacy systems.
Clinical workflow.
Regulatory knowledge.
Trust and relationships with end-users.
Eventually It Happens
Let's push the argument further. Could AI eventually get good enough such that it invades the vertical and end-users program themselves?
First we need to think about the economics of AI development.
Currently, we don't even have a clear AI model winner. All the leading models are competing amongst each other, trying to develop a foundation model that is cheaper, faster and smarter.
What amount of resources do you think is given to AI research to solve a specific vertical niche market?
We guess it's almost zero. Because going deep into a vertical to compete is a massive distraction to winning the AI model race.
But what if a model eventually emerges as the winner and decides to go invade vSaaS? Yes, that's possible, however during that time elapsed vSaaS is not just watching the show. Every one of them, to remain relevant, would also implement AI into their own products.
Attacking hSaaS
In our opinion, the likely outcome is to direct resources to breaching hSaaS moats. Because they have a few weaknesses given the proliferation of AI:
Low switching cost: By definition hSaaS users are not tethered to specific workflows.
Can be replicated: In the past, software development skills make a huge difference between a good and bad product. Now with AI the skill advantage decreases.
Weak network effects: When AI improves, the benefits are uniformly distributed to everybody.
AI models are getting really good at API orchestration. Tools like LangChain and AutoGPT can already read API documentation, write integration code, test it, and handle basic error cases. If you’re a developer who needs to connect hSaaS products with API, you can probably prompt an AI model to do it for you.
When hSaaS license renewal comes due, the life time value of hSaaS will be questioned.
In the meantime, vSaaS players get even more entrenched with every new customer.
What Happens Next
If our thinking comes true, then the economics will reveal themselves to rational and intelligent entrepreneurs. At the core, software is an incredibly profitable business. For those entrepreneurs, they would rather be creating vSaaS, and we will see more vSaaS companies enabled by AI tech (the industry calls them vertical AI). This means an increased supply acquisition targets for CSU.
But don’t be surprised if CSU doesn’t acquire them, because these new shiny startups don’t sell cheap. It’s possible that AI implementation at CSU offsets the new competition, but acquisitions are tamed in a hot market. We play the long game and trust that CSU will see deals when valuations turn sour.
We can also speculate on the potential vSaaS market expansion with the help of AI.
Just take 2 examples of large spending in vertical markets on services that AI could capture over the next decade:
Insurance Third-Party Admin (TPA) spend: $400b annually.
Life Sciences BPO/CRO spend: Estimated from $100b to $400b+ annually.
What's particularly interesting is the divide between the services spend and the software spend. Veeva is the largest life sciences software vendor and does $3b revenues. Guidewire and CCC, two of the largest insurance software vendors, do $2.2b revenues combined.
These are great vSaaS companies, but are smaller by a factor of 100x than the outsourced labor opportunity.
Will AI capture a meaningful percentage of the outsourced services revenue? We don’t know, but if AI captures any meaningful piece, there will be huge value creation.
Overall, the end users benefit from these developments and the broader industry should grow as a result.
So back to the initial question: How long is CSU runway?
We think it's very long.