Constellation Software has been on my watchlist for years now. But I couldn’t motivate myself to look into it while it trades at 50x cash flows.

Even Mark Leonard, founder and former CEO, said that in order to make 20-25% returns, you would need to buy Constellation after a major drawdown.

There are two truths in investing, and balancing them can be hard. One of them is that it can be the right decision to pay up for quality. You could’ve and would’ve paid a premium multiple for CSI for two decades, and you would’ve crushed the market.

However, one day, the second truth can hit your portfolio: every company will eventually face challenges. And when a stock trades at 50x cash flows, even the possibility of trouble can be enough to compress the multiple dramatically.

Very few companies have managed to postpone that moment for as long as Constellation has. But here we are. AI concerns have triggered Constellation’s largest drawdown to date.

For someone who hasn’t owned shares yet, it’s the perfect time to dive into Constellation. And if you already own shares and are uncertain about what to do, this might be even more important.

Let’s dive in!

— Daniel

Constellation – Decades of Outperformance Evaporated

Constellation has been one of the greatest compounders over the last two decades. A stock for which you needed to pay a premium if you wanted to own it. And yet, here we are. The stock is down over 50% from its highs, and CSI (short for Constellation Software Inc.) is now perceived to be facing one of its biggest threats ever.

At least, that’s what the market thinks. Constellation doesn’t own high-tech software companies. Most of its products are simple, niche, and highly specialized. On the surface, that might look like something AI could replicate in a matter of hours — if even. I’m not so sure about that.

But before getting into that debate, it makes sense to first get everyone on the same page.

Mark Leonard – The Genius who built Constellation

Mark Leonard is one of the most iconic CEOs out there. He not only looks a bit like Gandalf the Wise, but in many ways also embodies that same sense of calm and long-term thinking. I apologize, but it’s simply impossible for me to skip the Gandalf comparison when talking about Mark Leonard.

But jokes aside, he is a genuinely fascinating character. If you want to break into finance today, the typical path involves studying at a top university, getting strong grades, completing multiple internships, and navigating a lengthy application and interview process. Mark Leonard took a very different route before eventually working in venture capital. At various points, he worked as a dog handler, a bouncer, and even dug graves — just to name a few of his jobs.

It’s fair to say he built a broad base of real-world experience. Eventually, he found his way into venture capital and realized that investing was much closer to what he wanted to do long term.

However, there was one aspect of this part of finance he didn’t appreciate: the short-termism. Leonard wanted to invest in businesses “forever” and observe what happens when companies are allowed to compound over long periods of time.

Warren Buffett has clearly been a major influence on his thinking. It’s quite remarkable to consider Buffett’s longevity as an investor. Shawn and I have been inspired by him in our mid-20s, and Leonard, now approaching 70, has been influenced by him as well. That says a lot about the timeless nature of Buffett’s philosophy.

Leonard’s original idea was not to build a VMS acquirer specifically. Rather, his plan was to identify an attractive industry or business model and then construct a long-term holding company around it. He is, at his core, a value investor. Vertical market software simply turned out to be the most compelling opportunity set he could find.

Business Model – Why Vertical Market Software?

So what actually is VMS? VMS stands for Vertical Market Software, meaning software designed for a very specific industry or niche. A common example would be cemetery management software. These systems handle everything from mapping gravesites to scheduling burials and maintaining genealogy records. If you run a cemetery, this software is essential. Outside of that niche, it’s practically useless.

That niche focus is precisely what makes VMS such an attractive space. Historically, it has come with strong competitive advantages and structurally high margins.

The first moat is high switching costs. Constellation owns software businesses that, for example, run mission-critical systems like local bus or subway networks. As one can imagine, there is virtually no room for error. The system has to operate reliably 24/7, year-round. There is little appetite for experimentation or for replacing a stable system with a vibecoded AI alternative.

In all seriousness, though, this mission-critical nature is central to Constellation’s strategy. The company actively looks for industries that do not prioritize cutting-edge features. Instead, customers value reliability, continuity, and low risk above all else. Software that simply works is far more important than software that is new.

This naturally leads to the second moat: sticky recurring revenue. Roughly three-quarters of Constellation’s revenue comes from maintenance and recurring services. Returning to the public transport example, a city does not just buy the software once. It pays ongoing fees for support, bug fixes, regulatory updates, security patches, and access to new versions.

Given the high-margin profile of VMS businesses, some investors assume customers will eventually seek cheaper software alternatives. That is also one of the main reasons why AI is frequently described as an immediate threat.

In CSI’s case, however, software costs are often less than 1% of a client’s revenue. That alone makes me think the price argument is somewhat overstated. Would a customer really risk disrupting a mission-critical workflow just to save on software that represents roughly 1% of revenue?

To be fair, 1% of revenue can still be meaningful, especially for low-margin businesses. But if a system change leads to operational issues and even a temporary loss of, say, 10–20% of revenue, that trade-off quickly looks far less attractive.

Of course, one could argue that new AI-driven tools might work flawlessly from day one, with little to no disruption. That is certainly possible. But from a value investing perspective, the key question is the margin of safety. When a system is deeply embedded in daily operations and reliability is paramount, even small implementation risks can outweigh potential cost savings. I recently discussed this with a friend, Varun, who gave a helpful example:

Imagine you own a gym. It’s not exactly an industry where 24/7 uptime determines life or death, or whether an entire city comes to a standstill.

And yet, system integration has to work flawlessly. If something goes wrong — say, payment data doesn’t transfer properly into a new system, and you suddenly have to ask customers to re-enter their credit card information. Given that gyms earn a large portion of their profits from members who pay regularly but rarely show up, this can be a huge problem.

Those members typically don’t cancel proactively. Part of it is convenience, and part of it is psychological; they like the idea of being gym members. But the moment the gym reaches out to ask them to re-enter their payment details, that can trigger a cancellation. What was once passive inertia suddenly becomes an active decision point.

You can apply similar logic to many of CSI’s customers. Take cemetery management software. In theory, switching providers isn’t about life or death (pun intended). In practice, though, no one wants to explain to a grieving spouse that the system migration caused data issues and the exact location of a grave is temporarily unavailable — hopefully to be resolved next week.

Switching software providers sounds simple in theory, but I’ve spoken with half a dozen people who worked in the industry, either for CSI or for a subsidiary, or for companies that fit CSI's target profile. Everybody says the same thing. You don’t switch deeply integrated software if you don’t need to. It’s always a pain.

Some things work perfectly in theory, but are incredibly hard to implement in practice. SaaS is not just code. As the name suggests, it’s “Software as a Service.”

And I believe many people underestimate the difficulties and costs involved in implementing and maintaining software solutions. Let me explain this further by looking at the AI risk in detail.

How AI Threatens Software Businesses

Shawn and I have now analyzed roughly a dozen businesses that are supposedly “under threat” from AI. And, as is often the case, the answer to whether AI will disrupt software businesses is: it depends.

Some competitive advantages have clearly weakened in importance, while others may have become even more valuable.

Take S&P Global as an example. When I covered the company, I mentioned that around 95% of its revenue comes from proprietary data. That’s an incredibly powerful moat in an AI-driven world. AI models are excellent at scraping and processing publicly available information. But if the data only exists in your private database, large language models can’t access it — unless you choose to license it.

Regulation is another moat. Using S&P Global again: corporate debt needs to be rated. To issue rated debt, you need agencies like S&P Global, Moody’s, or Fitch. There is no shortcut. You don’t get an official credit rating from ChatGPT. AI might help reduce the cost of analysis, but it doesn’t eliminate the institutional role of credit rating agencies.

CSI is a more mixed picture. The company owns more than 1,000 businesses. It’s obviously impossible to analyze each one individually. I looked into roughly 30 subsidiaries. So instead of examining every asset, you have to assess the framework: How does CSI select businesses? What are retention rates like? And what proportion of its portfolio is realistically exposed to disruption?

As discussed above, CSI looks for mission-critical companies with deep customer integration and high barriers to entry in all sorts of verticals.

Think hospital systems, utilities, transportation networks, and court administration. These types of businesses represent roughly half of CSI’s portfolio. And even if customers wanted to switch — which they typically don’t, as reflected in CSI’s ~98% retention rate — it’s rarely as simple as picking a new vendor.

Court records have a legal chain of custody. They handle evidence and records that decide over convictions, fines, custody rulings, you name it. “Chain of custody” means there must be a clear, documented history of who created a record, who accesses it, who changed it, when it changed, and why. Choosing new software means that the court must prove that nothing was altered during the move, timestamps and access logs remain intact, permissions stayed correct, and records are still admissible and trustworthy in legal proceedings.

The amount of extra work required makes the change almost impossible. Add to that the data's high sensitivity. Every LLM or AI would’ve been tested and declassified to handle this sort of data.

You run into similar difficulties with utility billing, which has compliance audit trails, and with government permits, which have public records obligations.

Given the specific tasks most CSI companies perform, I think the bigger AI threat is coming from “below,” so instead of ChatGPT or Gemini (competition coming from above), there might be AI-native startups targeting certain verticals with small teams and at much lower prices.

But at the risk of repeating myself, I still think this risk is somewhat overblown, given the level of customer lock-in and the relatively low software costs for most of CSI’s clients.

Take a hospital, for example, where someone is responsible for core software systems. Would they really take the risk of replacing a system that has worked reliably for two decades with a new solution from an AI-native startup with a small support team just to save a few basis points in costs?

I can’t imagine anyone would want to take that bet. If it doesn’t work out and the workflow is messed up, you get fired immediately, and the hospital likely has a dozen lawsuits hanging over its head, and if it works out, nobody will applaud either. It’s expected that things run, and you “only” saved a couple of percentage points on the software costs.

Mohnish Pabrai used to say: “Heads, I win. Tails, I don’t lose much.” This situation can be described as: Heads, I win a little. Tails, I lose everything.

But what if companies start building their own software? In the end, who knows their business better than the company itself? Even Mark Leonard said that: “AI makes it potentially way more exciting for us to provide customization, but it also makes it much more likely that the client will do it themselves.”

Personally, I believe the impact of this will be on the margin. When you speak with small business owners, the constraints they consistently point to are time and capital.

Given that reality, how likely is it that they will divert scarce time away from revenue-generating activities and toward building their own internal software just to achieve marginal cost savings? In most cases, that seems like a low priority. From that perspective, it’s hard to argue that this risk alone justifies tens of billions in market cap erosion.

That said, I’m not dismissive of AI-related risks altogether. One scenario that could be genuinely disruptive is not necessarily better standalone software, but AI agents becoming the primary interface layer. In other words, the intelligence layer that customers interact with — automating workflows, generating reports, and assisting with decision-making — could sit atop existing systems.

So instead of having to maneuver CSI’s court software themselves, employees would simply instruct an AI agent: “Prepare tomorrow’s court docket, flag conflicts, generate notices.”

At that point, the agent vendor would take ownership of the customer relationship, and CSI’s software would become more of a back-end system of records. That would make it more “utility-like.” It wouldn’t be replaced right away, but the front-end owner, aka the agent, would capture most of the pricing power.

The good news is that CSI has the first shot at building this layer. CSI has the data, the domain expertise, and customer trust, and it knows exactly how customers use their software and what they want. What might become a problem is CSI's decentralized nature. This has been a huge advantage over the past decades, but it might turn out to be a problem in this case.

With that in mind, it makes sense to briefly walk through how CSI is actually structured. Understanding its organizational setup helps explain why rolling out in-house agents across such a large and decentralized ecosystem may be more difficult than it initially appears.

Constellation Software – Structure and Incentive Systems

One of the most important factors when analyzing programmatic acquirers is their corporate structure and the incentive system behind it. It’s probably not an exaggeration to say that Mark Leonard is one of the key pioneers of the programmatic acquirer model. Nicholas Howley, the founder of TransDigm, would likely belong on that same Mount Rushmore as well.

While their systems share similarities, Constellation has arguably taken the model further than anyone else. Despite being an exceptional capital allocator himself, Mark Leonard deliberately designed Constellation to be highly decentralized from the outset. Rather than centralizing decisions at headquarters, he pushed both operational responsibility and capital allocation down to the operating groups.

When you think back to the web of companies that CSI has acquired by now, you can imagine why Leonard wanted to run CSI as decentralized as possible. It would’ve been impossible for a single person to make all these acquisition decisions. For context, at its current scale, CSI completes up to around 100 acquisitions per year.

Beyond the sheer time constraint, there is also the issue of expertise. No single person can realistically have deep knowledge across dozens of niche verticals. By structuring the company around operating groups that focus on specific industries, CSI ensures that decision-makers with relevant domain expertise remain close to the businesses they evaluate and operate.

Most deals are in the $5 million range. Historically, when deals became larger—tens or even hundreds of millions—Mark Leonard would step in more directly. But the bulk of capital deployment has always been handled at the operating group level.

Another reason decentralization allows for much greater flexibility in acquiring companies is the “physics of human scale.”

Smaller teams tend to function more effectively, typically in the range of roughly 5 to 40 people. At that size, trust is higher, communication is clearer, and individuals feel genuine ownership over outcomes. As a result, when a business unit grows too large, the preferred approach is often to split it into smaller, more focused units to preserve agility and accountability.

When you decentralize your operations, you must have a perfectly aligned incentive system, though. As Charlie Munger used to say: “Show me the incentive, and I'll show you the outcome.”

The key metric for acquisitions has traditionally been IRR. Managers are generally expected to clear a ~20% hurdle rate before deploying capital. As CSI grew, Leonard became more open to larger acquisitions. Since competition for those larger deals is more intense, they often come with slightly lower expected returns. A ~15% hurdle has been considered.

However, Leonard was very reluctant to broadly lower the hurdle rate across the organization. One reason he cited can be described as the “magnetism effect.”

The idea is simple: you tend to get what you aim for. If managers know they are allowed to pursue deals at 15% IRR, the average acquisition IRR will naturally gravitate toward that level over time.

Why? Well, because managers are incentivized to deploy capital. So they prefer investing capital at 15% IRR to investing no capital at all. With that in mind, they might decide not to negotiate as much as they used to because they would rather not lose the deal entirely. When the hurdle rate was 20%, they negotiated because they had nothing to lose. They couldn’t make the deal at 15%.

You might ask where those high IRRs are even coming from. It’s not like more or less mature cemetery software has huge organic growth rates. Instead, the big drivers of return tend to be what multiple you pay, no surprise for us value folks, and how much margin improvement is realistically available after acquisition.

TransDigm is known for aggressively raising prices after acquiring a business, charging as much as customers will tolerate, supported by strong competitive positioning and limited alternatives. CSI also raises prices, but it doesn’t have as much pricing power as TransDigm. Many of its customers spend roughly 1% of revenue on software, and Constellation tends to respect that threshold.

Since they often acquire “old” software businesses, there are many levers to pull to save costs and improve revenue quality, too. A common one is to push the revenue mix toward higher-margin recurring maintenance, reduce low-return custom work in professional services, or at least price it properly, and cut costs on bloated overhead that accumulated under prior ownership.

On the operating side, Constellation has historically measured performance by looking at ROIC plus organic net revenue growth. The idea is to capture two value drivers within the metric. First, the returns on the existing capital base. And second, the ability to grow revenues in capital-light software businesses without needing to proportionally reinvest in factories, inventory, or working capital.

As Constellation scaled, free cash flow became a more important performance metric and gradually replaced the earlier focus on ROIC plus organic revenue growth. One reason is that ROIC can become increasingly misleading over time, especially once the original purchase price of an acquisition has been effectively recovered. In those cases, ROIC can look artificially high, reducing its usefulness as a decision-making metric and becoming even more problematic when tied to incentives.

Let’s go through a quick example: Imagine acquiring a VMS business that generates $2 million in EBIT for $10 million. Assuming a 25% tax rate, NOPAT (net operating profit after tax) would be $1.5 million. To calculate the one-year ROIC, you divide NOPAT by invested capital, which in this case is the $10 million purchase price. That results in a reasonable ROIC of 15%.

Now fast forward seven years. Over that period, the business would have generated a cumulative NOPAT of $10.5 million, exceeding the original purchase price. Because these businesses are typically asset-light, the tangible capital required to keep them running can be quite small.

If, for example, the ongoing invested capital in year eight is only $200,000, the ROIC calculation becomes distorted. Dividing $1.5 million of NOPAT by $200,000 of invested capital would imply a 750% ROIC.

At that point, the metric stops reflecting economic reality. Leonard has explicitly warned that this dynamic can distort incentives, as managers might appear to be generating extraordinary returns simply because the historical acquisition cost has already been recovered, not because new capital is being deployed exceptionally well.

Long story short, Free Cash Flow to Shareholders (FCF2S) has become the key metric to focus on when analyzing CSI. With that in mind, it would be natural to move straight to valuation and the investment decision. However, there is one more important development to address first.

Following CSI’s AI-focused call in late 2025, Mark Leonard announced that he would step down as CEO due to health reasons. The timing was, admittedly, unfortunate, but there is little reason to doubt that health was the genuine cause.

Leonard has consistently been one of the most candid and transparent CEOs in the public markets. It seems unlikely that he would step aside because of a new strategic challenge and frame it differently. Especially given that he has not appeared particularly concerned about AI risks, arguably with good reason, as discussed earlier.

Mark Miller will take over as CEO. You could call him something like an unofficial co-founder. He has been with the company for more than three decades. He founded the company that later became Constellation’s first acquisition in 1995. So he has been there right from the beginning. As a result, I do not expect any major strategic changes. The only notable difference so far is that we are now getting quarterly calls again, a tradition Leonard moved away from many years ago.

Valuation and Investment Decision

Alright, it’s time to talk valuation. To be blunt, my model essentially reflects a “business as usual” view of Constellation. If you believe AI will materially disrupt a large portion of CSI’s subsidiaries, then this likely isn’t the right investment—regardless of the price. The simple issue is that there is too little visibility into how many subsidiaries are truly exposed to AI risk and to what extent.

In other words, investing in CSI requires the belief that the moats discussed throughout this article — mission-critical software, high switching costs, and deep customer integration — remain largely intact in an AI-driven world. Personally, I view that as the more probable outcome.

With that backdrop, I model revenue growing at a CAGR of roughly 12%, which implies a gradual deceleration compared to earlier years, mainly due to CSI’s increasing scale. I’m also not assuming any meaningful margin expansion. At most, I expect slight improvements driven by a growing installed base, a higher share of recurring maintenance revenue, and potentially some efficiency gains from integrating AI into internal workflows.

If you want to paint a more bullish picture, you can assume some additional margin expansion.

Applying a 25x multiple to the projected cash flows and discounting them at 8%, I arrive at a fair value of close to $3,000 (in USD, not CAD). From there, I apply our standard 20% margin of safety, which results in an intrinsic value range of roughly $2,300 to $2,400 and implies an expected IRR of about 19%.

This comfortably clears our hurdle rate. That said, a model can only ever reflect the assumptions that go into it. Again, the model largely assumes a “business as usual” scenario, and there is a reasonable argument that the future may not unfold that smoothly.

I’m willing to take the other side of that uncertainty. Even so, we are not adding CSI to the portfolio today. Not because of concerns about CSI itself, but rather by the opportunity set that emerged during my research process.

In particular, I came across several spinoffs and companies within the broader CSI ecosystem that could benefit even more from a recovery and potentially offer higher upside over the next decade, given their smaller size and longer growth runway.

I’m currently working on an episode covering the most compelling of these opportunities. Since our portfolio is cash-constrained and every addition now comes down to opportunity cost, it makes sense to evaluate the smaller players first. If none of them ultimately present a more attractive risk-reward than CSI, we would be comfortable allocating to CSI instead.

For more on Constellation Software, you can listen to our podcast here.

More updates on our Intrinsic Value Portfolio below 👇

Weekly Update: The Intrinsic Value Portfolio

Notes

• Copart released earnings this week, and they haven’t been good. However, that wasn’t a huge surprise either. The market initially reacted by sending the stock down about 11%, but most of those losses were recovered the following trading day.

  • Net income declined 9.5% year over year, and revenue was down 3.6%. Free cash flow, however, increased 58%, driven by lower capex spending (fewer investments in new yards) and higher average selling prices (ASPs).

  • Importantly, Copart also began repurchasing shares again, buying back roughly $218 million worth. Management has historically been very disciplined about buybacks, typically acting when it views the stock as materially undervalued. In that sense, the repurchase activity can be interpreted as a positive sign.

  • As for the revenue and earnings miss, most of the headwinds appear cyclical to me. Lower vehicle unit volumes are likely tied to a consumer pullback in insurance coverage, and comparisons are tough given last year’s catastrophe-related volume boost. That said, IAA, Copart’s main competitor, seems to be executing well operationally and may be regaining some lost ground. Still, Copart continues to report “account wins,” suggesting that market-share loss is not the primary driver of the rather disappointing quarter.

Quote of the Day

"Over the long term, stock returns will be determined largely by which capital allocation decisions the CEO makes. Two companies with identical operating results and different approaches to allocating capital will derive two very different long-term outcomes for shareholders.”

— Mark Leonard

What Else We’re Into

📺 WATCH: Aswath Damodaran on the Software Selloff and the Macro Impact on Markets

🎧 LISTEN: Clay diving into Daniel Kahneman’s book Thinking, Fast & Slow (one of my all-time favorites)

📖 READ: A viral Article on X on the impact of AI on Saas (even better than the article itself are some counterpieces)

You can also read our archive of past Intrinsic Value breakdowns, in case you’ve missed any, here — we’ve covered companies ranging from Alphabet to Airbnb, AutoZone, Nintendo, John Deere, Coupang, and more!

See you next time!

Enjoy reading this newsletter? Forward it to a friend.

Was this newsletter forwarded to you? Sign up here.

Join the waitlist for our Intrinsic Value Community of investors

Shawn & Daniel use Fiscal.ai for every company they research — use their referral link to get started with a 15% discount!

Use the promo code STOCKS15 at checkout for 15% off our popular course “How To Get Started With Stocks.”

Follow us on Twitter.

Read our full archive of Intrinsic Value Breakdowns here

Keep an eye on your inbox for our newsletters on Sundays. If you have any feedback for us, simply respond to this email or message [email protected].