Seeking a friend for the end of the world (of software as a viable business, apparently)
Quick thoughts on some potentially durable survivors of the recent "SaaSpocalypse" selloff
*This is a compilation and continuation of a series of substack notes I made starting on March 5. I’ve added some names and details since then to the point where it made more sense to just compile everything into a post. You can use the small nav menu which (at the time of this writing) should be visible at the center-left of this post (when viewed on the Substack site) to skip around directly to the various names covered.
Software has taken a beating YTD on so-called “SaaSpocalypse” fears of AI eventually being able to act as a cost-competitive, infinite, and on-demand bespoke app store for whatever need an enterprise may decide to spec for. Now, however, software has started showing strength (or at least a stop to the bleeding) relative to everything else during the volatility of continued US/Iran (and associated shipping & energy) concerns (as well as low correlation). I won’t try to “post-dict” exactly as to why, but it is happening (for now) —sans the occasional panic in the market each time a new LLM model is released or discussed by the big model labs. On that note, I thought it would be interesting to pick through the rubble of SaaSpocalypse survivors for those names that might be least affected by the concerns of that dip-inducing narrative and (ideally, maybe) even be beneficiaries.
BTW, when I say “AI”, what I mean (and I think what most others mean as well) is LLMs and the current wave of software tools/features that all use LLMs as the reasoning/interpretation engine behind their ability to act and plan/communicate with users, systems, or to themselves or each other for the purpose of completing generalized tasks. (Keep in mind that AI in the sense that was most commonly referred to right before the explosive popularity of ChatGPT and LLMs was really more accurately described as highly-specialized or task-specific implementations of “machine learning” techniques (think of statistics-based spam filters or recommendation/ranking systems for email or music streaming services as examples of this latter category)).
Commentary
The tech industry is going through a set of sudden step-function changes all at once. One is upwards in customer expectations & understanding of what is possible and a step-change downward in the speed and cost of code/data production/analysis/aggregation for all incumbents, outside challengers, and de novo developers. By leveraging AI agents, the total number of user seats that customers may need to pay for under traditional seat-based licensing is also declining, though the total amount of “activity” may be increasing due to using these agents. Meanwhile, AI is introducing whole new categories of black-box liabilities on all sides (eg. auditability/compliance challenges in environments that require determinism, causal reasoning, explainability, and accountability; LLM hallucinations; and new ways hackers can exploit these systems). What we want to do here, IMO, is determine what is becoming more scarce (or abundant) —in meaningful magnitude, direction, and de/acceleration— and determine who actually profits (or is diminished) by it.
Something I think is interesting is this 2022 graphic (below) of value capture along the aviation industry’s supply chain. We see that the vast majority of invested capital in the aviation industry —which comes from investors and lenders who, excluding the government, you would assume are trying to make a profit— is deployed into subsectors with average returns below their cost of capital. This apparent crowding into negative-value-creation subsectors included the most visible subsectors that come to mind when people typically think of the aviation industry and has persisted into (at least) 2024. I wonder what parallels to this might emerge in hindsight as the AI industry matures.
I think it makes sense to assume that the story of which parties in the AI ecosystem that ultimately emerge as the ones to capture the most value by the time we get to the “mature” era in the adoption S-curve will have many surprises. That is, the “winning” subsectors will not be predictable —they may not yet even all exist— and I’m not counting out (certain elements) of the software industry so soon.
There are many ways the future can play out, so it may be easier and less prone to salience- or availability-bias to simply catalog those sectors that presently have the most constraints imposed on them (or, invertedly, constraints keeping competition out) and have (de)accelerating “attack surface” (de)growth. The way I see it, going after the space(s) with the lowest/shrinking attack surface (but perhaps hardest/growing constraints) is a good way to increase the chances of picking winners occupying or soon-to-be occupying true bottlenecks (or at least to help avoid picking the biggest long-term losers).
Starting with the least attractive (and this will not be a comprehensive list), I put traditional horizontal SaaS businesses. Yes, there are switching costs but the situation is one of customers switching from a general tool to another general tool (apples-to-apples) while LLMs cut dev costs for comps (and cut (re)integration cost for clients thinking of switching), open-source LLMs kill any AI differentiation via the closing quality gap vs open-source, and the hyperscalers continue to bundle equivalent generalized functionality themselves (eg. Google Calendar + Gmail appointment scheduling integration for a Google Meet video call, perhaps with an attendee connecting from a Pixel phone running the Android OS, to discuss/collaborate on a budget saved to a Google Sheets doc) 1. A bit lower on the vulnerability list, I include the big standalone model labs. Often the most visible players in the ecosystem, they are continuously threatened by the aforementioned open-source quality catch-up and have little switching costs between them. (FD: I have a meaningfully-sized Alphabet position that was accumulated from before the SaaSpocalypse and through the US/Iran volatility).
Nearer to the very bottom of the competitive vulnerability ranking, I put “powered land” owners benefiting from the scarcity of appropriately zoned locations and physical scarcity of available land for such purposes that are also proximate to necessary energy/data generation/transmission infrastructure, yet also far from anti-data center NIMBYism. While not quite as protected along the dimension of hard constraints as the data center infrastructure & land owners, but in this same general safe area along the dimension of competition exposure —in contrast to the horizontal SaaS companies which I think score poorly on both constraints-hardness and competition— I also put niche vertical market software (VMS) in the tradition of Constellation Software and Roper Technologies (for reasons I’ll attempt to cover in this post).
.
This post is just a brain dump overview of a few of those beaten up software names that may have good chances to actually retain or strengthen their moats in the long-run as AI/LLM-use continues to develop. Given that it’s one of the most disfavored sectors of the market right now —and not for no reason, to be fair— I think it’s currently one of the more interesting areas to look at for individual oversold situations (keeping in mind that beaten-down does not necessarily mean oversold).
(FD: I own small amounts of most names here and may add or sell at any time. I don’t see any need to rush into any position in this space as I doubt there’s going to be any single major event that de-risks software all at once from the shadow of AI in peoples’ minds any time soon. In any case, many of these names are additionally rightly or wrongly discounted for particular reasons beyond AI —and generally the actual reason these names were even on my radar at all— that market consensus appears to believe have a reasonable chance of being resolved only after an unfavorable length of time; so again, averaging in at the rate of small nibbles is advisable here.)2
.
ROP (Roper Technologies)
Anyone unfamiliar with the Vertical Market Software can see the overview here (https://www.deepsailcapital.com/_files/ugd/50dba3_8f765d8366be4a28b9bc97092d44291c.pdf), here (https://www.colinkeeley.com/blog/vertical-software), and for a good overview on Roper specifically, the older podcast here (if you watch the video on YouTube, you can use their Gemini Ask feature right below the video player to interrogate the transcript to get an understanding of Roper as well):
Roper today owns high-margin software systems that cater to a wide range of niche, generally acyclical, industries in the US such as government contract billing/compliance mgmt, law firm billing/mgmt systems, healthcare IT, childcare & education administration software, North America’s largest freight/carrier matching network, a church engagement platform, and hardware + software for water utility metering and billing. The cashflows from these decentralized business operations are then used to acquire additional VMS systems for ROP’s portfolio either as tuck-ins to integrate/enhance existing systems or to branch into new niches. (I think of ROP as an experienced permanent capital VMS “fund” through which one can express the software-ain’t-dead-yet idea. Being part of the S&P500, most brokers also support buying fractional ROP shares, which even further lends to the “fund”-like character w/ which one could treat the stock).
ROP systems are highly integrated into customer workflows (in terms of both employee training and systems integration) and hold valuable historical customer data —data that can’t just be scrapped from the internet and fed to open LLMs— which makes the systems sticky (especially to the extent that ROP systems additionally function as cross-customer networks, as with the freight matching network of ROP-owned company DAT Freight & Analytics). 3
The fact that much of ROP’s existing segments and M&A targets are in industries with risky/complex regulatory/compliance validation hurdles means even more so that individual businesses will still likely be less inclined to try to bother building their own systems even as AI reduces internal dev costs (this is a rather overlooked element IMO). This switching cost compounds when coupled with ROP’s strategy of being systems-of-record, mission-critical, and low share-of-wallet for customers. (On the system-of-record point, mgmt. has explicitly stated that very little of their pricing is thus user/seat-based and so is less vulnerable to agentic AI productivity-per-employee boosts by customers). These systems also tend to need to be auditable and, most often, fully deterministic which is an inherent problem for LLMs whose outputs are probabilistic even at very low token-selection-variance “temperatures”.
One way we can validate this LLM-resistant stickiness is by observing that ROP’s weighted average gross retention ratio across their various business units (which measures the percentage of recurring revenue retained from their existing customer base between periods, excluding any gains from price increases, up-selling, or cross-selling to existing base) has remained steady around 95% gross retention in each year before and after the launch of ChatGPT in 2022 up to today. (Also keep in mind that the remaining 5% churn here does not necessarily imply that ROP is losing business to competitors as it can also be driven by customers going out of business or total usage of their ROP systems shrinking due to specific factors w/in their industry at any given time). Weighted average net retention (which includes cross/up-selling and expanded use by only existing customers) generally remains above 100%, meaning existing customers spend more YoY (though there are difference between the many products).
AI may also help ROP accelerate the problem-to-product feedback loops between their own decentralized business units and customers to further entrench ROP systems into customer processes and increase operational switching costs. Unlike with the horizontal SaaS players, ROP’s niche products all require a high degree of domain-specific knowledge (eg. to engineer for continued regulatory compliance, loopholes, and edge-cases), which is partially why ROP’s structure is so decentralized. This compounding bespokness leads to greater entrenchment; IMO the feedback loop is enhanced by agentic coding to the benefit of incumbents like Roper. Furthermore, as ROP targets small niche subsectors, the incentive for new competition to enter the space remains limited (unlike, again, for the horizontal SaaS players) 4. We can monitor how well ROP can actually monetize direct AI features in their existing software through 2026 vs any potential AI-native/enabled startups.
It may even be the case that AI and the lower cost of code could end up helping resolve concerns around the dwindling, high-quality, needle-moving M&A runway for VMS roll-up models such as ROP’s (to the extent that new entrant do (occasionally) break through). The company recently shifted from cheap cash-cow targets to growth-at-a-reasonable-price (GARP) M&A as was seen in the higher-than-normal multiples they paid for Procare (day-care mgmt. software) and Subsplash (a church engagement platform) in 2024 and 2025, respectively 5. IMO, these acquisitions were seen as manifestations/validation of a concern by the market that ROP was running out of good targets under the old strategy (irrespective of mgmt’s comments on the reasoning for the change) 6 and was one of several contributing factors to ROP stock’s price weakness even before the SaaSpocalypse in late 2025. Agentic coding + growth of the AI industry (and any resulting yet-to-be-formalized-or-invented subsectors/offshoots thereof) ==> more opportunities for VMS startups to cost-effectively address new niche markets ==> more M&A opportunities for ROP (while, again, I think ROP retains incumbency advantage in the spaces they already exist in). This benefits ROP in an asymmetrical way iff their scale, relationships, and reputation (as an acquiree “home of choice”) facilitates better M&A deal flow. This does seem to be the case as mgmt mentioned in the 4Q2025 earnings call that 60% of their recent tuck-in acquisition pipeline was “either proprietary or founder-driven”, meaning based on direct private outreach rather then from, say, an open competitive auction.
Note that ROP is currently trading at its lowest EV/EBITDA multiple since 2015 (when it initially completed its transition from industrial equipment to software/VMS and changed its name from Roper Industries to Roper Technologies) while YoY EBITDA growth has remained in the 9-12% range in each quarter since 2023, so you’re being paid to wait as price has stabilized in the $340-$350 range. Meanwhile, ROP initiated it’s first ever buyback program in late 2025 and has brought shares outstanding down 5%, back to around 2017 levels. ROP’s weighted average buyback cost basis so far is around $366.67/sh.
The market appears to be reacting to the AI concerns by completely resetting it’s terminal value assumptions on ROP. The compressed multiple on the higher EBITDA base implies the market is discounting the perpetuity growth rate of those earnings back to what it assumed at the start of ROP’s transition when they had no track record (unlike now); I suppose this is to reflect the extreme uncertainty around software businesses in the era of AI. IMO, for the reasons covered above, I think ROP will ultimately be beneficiaries —and we don’t need to be a hero in terms of sizing here to have a position make a difference in the long-run nor do we need to rush to allocate a full position all at once.
Aside from the roll-up considerations, these general ideas carry into most of the other software names I cover here…
CCLD (CareCloud)
Niche private practice outpatient EHR (Health Record System, eg. that often-clunky-looking UI you see nurses and doctors writing their notes into when you get a checkup) and integrated RCM (revenue cycle mgmt) clearinghouse biz (ie. clinical billing/reimbursement mgmt services). Note that EHRs must be CEHRT certified under the US HHS to be used by clinics for reimbursement from Federal programs; this is a very strict and complex certification process for any new (eg. AI/vibe-coded) EHR startup and hospitals are highly unlikely to take chances swapping this existing system-of-record for another, much less for a new entrant. The EHR system is priced on a per doctor basis, while the RCM business charges a percentage of the revenues collected by the RCM system.
After simplifying their capital structure (converting 80% the series A preferred shares and lowering the dividend payments for the remaining that could not mandated to convert), paying down their debt, bringing previously-outsourced R&D expenses in-house, settling several long-standing lawsuits, and shifting roles at the exec team, the company is now using their growing internal FCF (w/out any debt or dilution) for M&A. The company is rolling upstream and rationalizing similar/smaller EHR/RCM companies that service other private practice and rural hospital systems that are too small for the bigger EHRs and hospital rollups (which acquire private practices and then move them onto their larger standardized EHRs) to bother directly targeting. Particularly, CCLD’s 2H2025 acquisition of Medsphere now brings CCLD’s EHR capabilities to full-journey inpatient healthcare (vs previously just being limited to outpatient focus); that is, CCLD can now support the full patient journey from outpatient clinics to emergency departments & inpatient hospital beds. CCLD is maintaining the existing EHR systems for customers as they work to integrate them all together under the original CareCloud EHR + centralizing or downsizing acquired RMC operations and using their offshored workforce. CCLD appears to be cleaning itself up for, IMO, eventual sale at a higher multiple after receiving an unsolicited bid from a private equity firm in mid 2024 that seems to have lit a fire under the butts of mgmt.
I’d note that, as opposed to all other names covered here, CCLD may be particularly affected by ongoing US/Iran conflict since most of their medical bill coding workforce is outsourced to Pakistan, which relies on the —currently still closed/blockaded— Strait of Hormuz and Persian Gulf for the vast majority of it oil and gas supplies. On top of that, medical coding (the RCM task of translating a doctor’s notes on diagnoses, procedures, and supplies from a patient encounter into nationally standardized codes for billing and analytics purposes) is particularly vulnerable to ever-improving LLMs; whether this means that clinics opt to, or find viable vendors to support, doing this job in-house at more attractive rates vs using CCLD’s RCM services, which I imagine will be equally AI-assisted, is yet to be seen.
CCLD received an unsolicited takeout offer in May 2024 which they rejected, claiming that the takeout offer did not include any immediate buyout of the Series A preferred shares (these shares did not have the same explicit change-of-control protections as the series B preferreds at the time and would have presumably become a liability for the acquirer to be handled at a later time); mgmt reject the offer based on this disparity in treatment. At the time, the takeout offer was $5/sh for the 16.1MM outstanding common shares and a $25 redemption price of the 1,468,792 series B preferred shares ==> $117.22MM in cash for CCLD (w/ the series A preferreds still attached) ==> ~14x the avg. 2023 & 2024 FCF of $8.5MM (the prior year’s FCF was lower at $3.8MM, which would imply an even higher multiple, but let’s assume the buyer was accurately incorporating forward projections as well). While the company now is in better shape than it was in 2024, applying just that same multiple to CCLD’s higher 2025 FCF, we have a takeout price of ($20.5MM * 14x =) $287MM, which leaves $5.28 per common share after covering the $25/sh and (updated) $25.25/sh redemption prices for the remaining unconverted 984,530 series A and 1,511,372 series B preferreds. (From this, one can also do the math to play around with a lower or higher range limit by using, say, just the 2023 FCF number for a higher multiple (~30x) or just the 2024 FCF for a lower multiple (~9x); taking the mid between the 2023-based low and the avg-based multiple for a 12x FCF applied to 2025’s FCF, we get $4.31/sh for the common).
One last small gripe I have on this idea vs the others here is that it’s a one-and-done situation; the company cleans up and does a sale. I’d more ideally like something where I can put money in and just have it compound there over a long time without having to guarantee (even in success mode) that I’ll have find somewhere else to put it down the road (a nice problem to have, but a problem nonetheless).
AGYS (Agilysys)
Resort and casino management VMS (targeting middle- to upper-end resort operators) that is currently trading at the same valuation as when their near decade-long transformation (starting in 2017 under the current CEO) from computer distributor to pureplay VMS first went EBITDA-positive around 2022 (slightly lower now, even) after SaaS subscription profits started to offset the modernization R&D spend. It would seem their transformation from siloed on-prem software modules to profitable unified hospitality-focused cloud-based SaaS, on a relative time scale, arrived just in time for both a global jet-fuel crisis (via US/Iran/Hormuz conflict) and the LLM-driven SaaSpocalypse.
AGYS’s rGuest VMS modules handle and coordinate property/room mgmt/bookings (PMS), point-of-sales (POS), casino food & beverage comp and POS integration (highly regulated due to anti-money laundering laws), inventory mgmt, payment processing (fees here are volume based), and associated spa/wellness/amenities bookings. Note that (like many of ROP’s system-of-record products) these modules are not priced based on number of user seats, but based on volume of inventory managed (eg. number of rooms at a property or number of POS terminals & transactions). (As far as I know, hospitality PMSs like AGYS’s rGuest et al are not legally required to have any kind of data portability in the same way that exists for, say, EHR systems under the 21st Century Cures Act, so that’s a nice data lock-in feature vs CCLD).
As a testament to the stickiness of AGYS products, their Q3FY2026 earnings call 7 noted that several major customers switched some of their operations to different franchise brand flags, yet refused to switch to those brands’ mandated PMS systems in order to continue using their AGYS products (which are now being strengthened by integrating AI analytics into the existing data systems, eg. for better room/booking/upselling/upgrades mgmt).
As of calendar 1Q2026, AGYS has completed their pilot roll-out of AGYS installation/integration to replace legacy PMS systems for Marriott International properties across the US and Canada, a major project sales win for AGYS in 2022 that began development in 2023. (Is it crazy to think that luxury/premium hotels in the US and Canada should experience increased demand from travelers seeking safer alternatives to Persian Gulf luxury stays so long as Iran —now arguably more feather-ruffled than ever— keeps instability in that entire region elevated)? The project is now entering the full-scale implementation phase and is expected to 3x AGYS’s installed PMS base over the next several years (all of which is expected to require little/no incremental opex as most of the dev resources of the entire project are already in place from the piloting period) w/ significant financial contribution expected to start flowing to the bottom line by 2027. From this POV, one could maybe argue that AGYS is trading at (50/3=) 15x fwd (post-Marriott-implementation) EBITDA 8, not accounting for any properties’ additional use of the more-recently developed or acquired amenities-booking modules like AgilysysReserve or Book4Time that were not even available during the pilot leg of the rollout.
TBTC (TableTrac)
*Honorable mention
On the subject of casino mgmt VMS systems, TBTC sell and maintains the CasinoTrac casino mgmt system (CMS) and ancillary products that include modules for guest rewards & loyalty programs, marketing analysis, guest services, promotions, administration, vault and cage management, and auditing/accounting tasks. Again, US casinos and thus CMS systems are subject to strict state-specific regulations. Additionally, TBTC has exposure to casino operators on native american tribal jurisdictions (the company explicitly mentions the Indian Gaming Association Trade Show and the Oklahoma Indian Gaming Association Trade Show as key industry trade shows they prioritize annually to drive sales), which comes with additional and bespoke barriers to entry as well as risks.
Up until 2020, TBTC was focused on new system sales. Since that time, TBTC’s lumpy new-system sales appear to have come to a trickle (8 in 2025 vs 29 in 2019) and the company (quite unlike the growth focus at AGYS) now looks to be focusing on extracting more recurring value from it’s existing installed base of around 115 casino operators and 300 locations via higher-margin and lower-labor upgrade & maintenance projects. This idea is further validated by the recent 2026 transition of roles at the company where founder Chad Hoehne (owning 25.5% of the company) moved from CEO to CTO and the company’s CFO/COO moved into the CEO role, which to me suggests a re-focusing of the company on internal efficiencies and shareholder returns over growth. (Taking a page from Roper, we could consider TBTC, w/ their 10x EBITDA and EPS multiple, to be more like ROP’s prior cash-cow type of VMS, whereas AGYS is closer to their newer GARP focus).
The stock is extremely illiquid, so don’t expect any reasonably-priced orders to fill anytime soon. Recent insider buying —for the first time in 5yrs— has been around the $3.80-$4/sh range (w/ an absolute floor for the stock being around the $3.70 level).
CLBT (Cellebrite)
Israeli mobile device penetration & data extraction services/library (Cellebrite Unlock and Inseyets) + integrated digital evidence mgmt, (Pathfinder) ML analytics, and custody/case mgmt VMS (Guardian Forensics cloud SaaS w/ data volume-based fees).
After multiple quarters of disappointing Federal revenue growth and CIFUS-stalled Corellium acquisition (pentesting R&D platform), the company is now showing early inflection signs and is in the process of integrating Corellium into operations, launching the Guardian Investigate AI/ML module (similar to Palintir’s Gotham for storing, analyzing, and synthesizing all forms of digital evidence into court-admissible content), and approaching the tail end of FedRAMP High-level authorization to operate (ATO) certification which will open up greater Federal-level use/marketing TAM for CLBT’s Guardian suite 9. This certification was previously delayed —and the cause of some pre-SaaSpocalypse price volatility— in early 2025 due to DOGE shakeups in the Federal government that delayed the processing of the ATO as well delayed the required agency sponsorship (which eventually came from the DOJ) needed to start moving forward for Cellebrite Government Cloud. CLBT mgmt has already signaled that a large Federal deal is awaiting once the FedRAMP ATO is approved. 10
AI may get better and better at cracking mobile phones (though it also means the phone makers will get better at hardening them), but it won’t —by the fact of the probabilistic/stochastic way it is trained / generates outputs— be able to supply the court-admissible deterministic chain-of-custody audit and analytics data that CLBT systems specialize in providing nor can AI or synthetic data generation techniques replace the valuable corpus of real case evidence data that CLBT continues to collect (and charge recurring data storage fees on). CBLT is currently trialing AI-powered analytics to automatically comb through existing and incoming data to assist with, for example, finding new connections and leads related to existing cold cases.
Gross revenue retention for 2024 and 2025 (which, again, does not factor in any up-selling, price increases, or use expansion nor does it do well in capturing any customer-wide headwinds agnostic of CBLT’s competitiveness vs other substitutes) was around 92%, with net retention remaining above 100% through all years.
CASS (Cass Information Systems)
Another interesting VMS-like business I’ve been following (that, unlike the rest of the software world, has been slowly grinding upwards and approaching a technical breakout) is Cass Info. Systems.
CASS is a smallcap US-focused freight & utilities (energy, waste mgmt, telecom) invoice auditor (fixed price per transaction) and payment processing systems (dollar-volume/value priced, read as: inflation- or tariff-hedged) attached to a 120yr-old niche bank (interest income or think of as interest-rate-hedged) that is also used to facilitate said invoice payments. Around 40% of the bank interest income is from the funds held for the invoice/infosys customers during the time between invoice processing and payment (supplementing the profits they earn from the audit-to-pay cycle), while the remaining 60% are from loans to private businesses in the company’s home city of St. Louis, major restaurant franchises, and faith-based ministry mortgages (the latter being a sector that CASS also has exposure to via their TouchPoint church mgmt software). (CASS has also seen tailwinds from AI infrastructure buildout in the form of increased electricity usage and pricing flowing through to its utility invoicing business).
The company started in St. Louis, Missouri, in 1906 and began specializing in freight invoice auditing and payments as the city became a major trucking hub in the 1950s when the city was identified as a critical node in the Federal planning and construction of the interstate highway system —before which St.Louis was already a major railroad logistics hub.
Dividend investors may be interested to hear that Cass Bank began paying out a dividend during the middle of the Great Depression in 1934 and has since never failed to deliver a regular dividend —and actually raised their dividend through the years leading up to, during, and after the 2009 GFC (while their EPS remained generally unaffected as well through this period).
In handling the invoice auditing and payments for clients, CASS systems gathers data from a diverse & heterogeneous set of sources from various bespoke vendor invoices and customer procurement & sales systems (read as: inconsistent hodgepodge of non-standardized nonsense which is not inherently conducive to LLM processing). This is compiled to audit and assist with decision making against complex shipping contracts and parameters —that can differ greatly between customers— and spotting over-charges arising therefrom (https://www.freightwaves.com/news/freight-invoice-audits-and-why-they-matter). Combined with the banking subsidiary, CASS is able to act as a turnkey system for auditing invoices as well as paying/settling them, helping to entrench their products into client workflows as well as creating a regulatory barrier that AI-startups will have difficulty overcoming; there are strict regulations for non-traditional buyer companies attempting to acquire a bank and bank-holding companies are subject to compliance rules from Federal Reserve, FDIC, et al.
From the 10K: “The specific payment and information processing services provided to each customer are developed individually to meet each customer’s requirements, which can vary greatly.” That is, Cass systems are embedded in customer workflows and accounting systems in a highly bespoke manner that, again, makes it disruptive for customers to switch to another invoice auditor (requiring employee retraining and systems re-engineering, eg. to account for any differences in reporting style, format, or resolution from substitutes via what is returned from CASS). CASS pulls/pushes directly to/from customers’ various ERP and RCM systems to get/return invoice/audit/settlement data. Same goes for the utility/facility segment: “Many of Cass’ services are customized for the ESPs [energy service providers], providing a full-featured solution without any development costs to the ESP.”
CASS also acquired a small subscription-based (rather than fixed transaction-based) SaaS invoice auditing company —that they can also now supplement with their own data corpus and directly handle payments for— in late 2024 that specializes in ocean and international air fright auditing (vs their domestic US business). This small contributor could become a valuable segment in due time as/if deglobalization trends continue and supply chain auditing becomes ever more onerous.
Note that the bank funding from the non-bank side of the business is NOT held there because, say, there is some lag time between the moment where funds land at Cass Bank and when the goods arrive at the buyer destination like some kind of escrow arrangement. Rather, the lag time is mostly a contracted form of “payment” where clients allow CASS to make interest income on their funds before settling the transaction as the bank funds settle at Cass and the audit process takes place. As CASS’s audit process becomes more efficient with AI and more businesses take greater advantage of instant settlement methods like FedWire/FedNow or perhaps via tokenization, it would seem the actual necessary lag time should decrease and becomes more about clients simply “paying” with their own revenue cycle days. This should not be a problem (actual all-in “costs” in their various forms will work themselves out) so long as CASS can retain their value proposition in their auditing software (and customers still need to go thru a bank even if they want to use FedNow payments). That is, so long as they can leverage their data corpus of invoice transactions & audits and AI integration to defend against disintermediation from less-niche invoice auditing competitors, clients have less incentive to disintermediate from CASS on the banking side by settling at their own FedNow-supporting banks. (CASS also has developed CassPay to act as an integrated platform for general B2B integrated payments).
From an portfolio exposure POV, I personally —though this is not totally apples-to-apples in terms of customer type— compare CASS against the fright-matching network of Landstar Systems (LSTR); both generally being driven by US trucking transaction volumes. While avoiding making this any longer than it already is, I’d just say that LSTR’s business creates more variable layers of —largely out-of-their-control— expenses between their revenues and operating income lines vs CASS.
For personal sizing purposes, useful to note that the apparent price floor of $40-$45 has also shown validation via consistent insider buying at those levels.
.
I include data infrastructure systems like Snowflake et al in this horizontal category as well and can see a future where Google uses agentic coding to (more quickly than has already been happening and w/ less dev-time and opportunity cost to themselves) come up with better and more integrated intermediary tools/interfaces for people/systems to interact with data that is ultimately just stored in GCP cloud storage, BigQuery, CloudSQL et al. If software companies are (still) going to eat the world, IMO GOOG is well in the running to be the one that eats the software companies eating the world (w/ a call option on also powering a good deal of the AI companies via their TPU chips (only available on GCP)).
Expect volatility and don’t expect to be able to bottom-tick any given name with a big allocation in a single day. Personally, from a portfolio composition POV, I group all of these names into a single value-at-risk bucket/subbucket of “Software/VMS” when thinking about total portfolio allocation ratios and have been incrementally, though unevenly, adding to all names with that in mind.
Note this particular data moat is highly dependent on how exportable/portable customer data in ROP systems really is/remains, though this should at least be relatively less portable than general data from, say, enterprise ticketing systems et al given the bespoke nature of the businesses and processes that ROP products cater to.
Even for ROP’s recent GARP-focused M&A, their industry TAM estimates remain in the $1bn-$3bn range. For context, this is on the lower-end marketcap size for individual companies included in the S&P 600 smallcap index.
Historically, Roper acquired businesses at multiples of 7x to 8x EBITDA. These more recent acquisitions have were done at multiples in the high teens to low 20s.
In the Q3 2024 earnings call, mgmt characterized the change as a “modest improvement” designed to capture more value from their capital deployment by acquiring “maturing leaders” that are “a bit faster-growing”. So yes, more GARP-like, but whether this change was truly due to a recession in the pipeline of traditional ROP targets or the result of a deliberate and un-forced re-evaluation of their M&A strategy is still uncertain.
This covers calendar Oct 1 – Dec 31 of 2025, which seems to be the norm for hospitality companies so that their full fiscal year can cover the entirety of the bookings-heavy winter months
3x existing rooms under the PMS segment (as opposed to legacy on-prem) ==> 200% additional total room count (note that AGYS does not break down op-prem- vs PMS-managed rooms). PMS revenues likely grow way more than just a pro rata 2x compared to prior per-room revenues since PMS installation sales have tended to be with much greater module/add-on density than the legacy on-prem sales (2 add-ons avg per on-prem install vs 14 avg per PMS install). Even if we assume that the bulk of existing rooms are on-prem managed, all the additional 200% additional rooms are PMS-based, so I make assumption that the additional revenues will be not just 200%, but 300% of existing revenues. I also assume that the on-prem-managed rooms remain stable. Gross margin assumed to remain around the FY2025’s 78% and also assuming EBITDA margins actually expand from 20% to 25% starting in FY2027. In any case, so long as margins just remain static, that 3x revenue increase falls down to the EBITDA line in pro-rata fashion as mgmt has already mentioned that most opex for continued rollout for the Marriott project is already in place.
A FedRAMP Authority to Operate (ATO) certification formally validates that a cloud service provider’s product meets federal security standards, allowing it to store, process, and manage government data. See https://www.nccgroup.com/what-is-a-fedramp-ato/
“Yes, look, I mean, when you say backlog, I’ve got to be careful because the backlog would suggest that we have a purchase order. What we do have is we’ve seen our largest bid go out. We have scoped an initial contract for a large federal agency in their storage requirements, which exceeds 9 petabytes. And that would equate to a very large deal that would be certainly the single biggest transaction we would have ever done.” ~~~ CLBT 4Q2025 earnings call
This was very interesting, thanks for sharing.