Market psychology is fascinating. I would venture to guess if you tracked the number of posts per unit time on this board, it significantly increases in times of portfolio highs and dwindles in times like now when many of our core holdings are down (and several even oversold).
It is precisely these times, however, that I believe the Saul style of stock picking shows its worth. These types of choppy, volatile markets induce a lot of fear and anxiety and cause irrational behavior. Many people (maybe the majority) are so worried about tomorrow that they forget the five-year plan. In losing foresight, they lose their edge and sell into weakness.
As Saulites, we don’t have that problem. We know our companies. We know their worth. When the market goes down, instead of a dwindling fortune we see a fire sale. We hold our noses and dip our toes in just a little bit more, buying when the masses are selling.
And then, months (or years) later, we reap the benefits.
Everywhere in my social media I see people talking about the huge software and AI selloff. Even Jim Cramer jumped in today: “Software no bottom yet…” When all the people are thinking one thing, who is left to join them? I’m not calling a bottom here, but it certainly is getting to a bit of a ridiculous level:
$APP down almost 38% from the highs
$CRDO down ~49%
$ALAB down ~40%
These are phenomenal companies that haven’t shown us any reason to panic. Earnings season is here, and could be the catalyst to turn us around.
If APP reclaims ATHs, that’s a 59% gain! If ALAB reclaims ATHs, it’s 65%! And if CRDO does, that’s 95%!!! It’s easy to feel despondent now when many are down YTD, but this is a game for those with an eye for the future.
If making money in the stock market were easy, everyone would be rich. But they’re not. Yet, look at the annual returns of our board members. It’s precisely when things seem the most bleak that our style shines the brightest. I’m beyond grateful for this board, for Saul, and for all of you that add value to our amazing community.
@DoctorRob, I agree completely. I often wish I had popcorn, as I watch the antics of market psychology these past few months! I especially love times when I see the same bit of news used as a reason for extreme greed and extreme fear by different groups. All stories. But if Saul taught me one invaluable lesson, it is to invest in the numbers not in the stories.
I found it an excellent exercise to do a survey of the old board favorites which are down 30% - 60% from recent highs. Many of these companies have new product offerings utilizing AI/LLM capabilities, often with very apparent benefit to both their customers and their own profits. Regardless of whatever else is happening in the world, the compressed valuations of these companies makes them look like an ever more attractive investment opportunity. This upcoming earnings season should provide some very interesting data…
When the market is in euphoric phase, companies can report poor or mediocre results, but the analysts and investors will find some positive slant to bid the stock up. This will last until it reverses, and then no news is good enough. Whatever it is, the analysts and investors will find some negative way of looking at it, and the stock will go down. So don’t panic at irrational sell-offs. If it’s one stock, then you have to make sure there’s not something wrong with that particular company. When it’s all your stocks and the pundits are all saying the big crash is coming, relax. It’s just something that they say now and then.
– Saul, Kowledgebase 2019 Part 2
Whilst I hold a lot of SaaS/Software exposure, I would expect that software attached to a platform should stand a chance of survival more than single point solution software only plays.
In the same way as Microsoft software was invincible back in the day because of the “attach” to the Microsoft operating system and the office suite which whilst still available as a bundle was a lock broken by anti trust regulators; I would think the modern equivalent would be the “platform” play.
Where software modules are attached to a platform or cloud equivalent of an operating system or a fundamental system of record as such or even non commodity hardware, then this should offer more value, less substitution prospects and provide more security and survivability. Think cybersecurity platform players like Crowdstrike and Palo Alto, think SAP, think Datadog, think the PaaS or SaaS players or think Shopify or even Snowflake and Databricks. For software with a hardware attach - think Square or Toast payment/hospitality physical terminals, Axon with body cams and Samsara with IOT sensor devices.
Of course all these software platform plays are also getting smashed, (I don’t want to talk about Shopify right now); but I would think their chances of survival vs the point solution pure software providers, will still thrive and will get separated from the rest of the pack once the market uncertainty around AI is removed and the investor industry figures things out.
Software companies, especially SaaS companies, are in real trouble from AI. They are getting attacked via multiple vectors, and only those that re-invent their products and their business model will survive.
Let’s start with the obvious attack vector, the one that many are talking about: Is your company using AI internally?
“Big Four” auditing firm KPMG hires, like all public companies do, an independent auditor (it can’t audit itself): specifically for them, Grant Thornton UK. What happened here is that, knowing the auditing business well, KPMG told Grant Thronton UK that they should be “using AI and other new technology to improve efficiency.”
OK, fair enough, but then the other shoes drops: KPMG then insisted that the cost savings from AI be passed on to them, and KPMG threatened to find a new accountant if Grant Thorton UK did not agree to a significant fee reduction.
They got a 14% discount from the previous year.
Any business that isn’t using AI risks being out priced in the market by competitors using AI to reduce their costs.
But, it gets worse:
Many of you will remember an old Board favorite - Monday .com. Well, this story came out recently:
Two CNBC reporters with no coding experience created a working Monday.com clone in under 60 minutes using Anthropic’s Claude Code AI tool, spending just $5-15 in compute credits.
One of the reporters, Jasmine Wu, writes:
We didn’t expect to get anywhere — we’re not developers and we don’t have any coding experience. …We started out simple, telling Claude [CoWork] to build a project management dashboard similar to Monday, with features like multiple project boards, assigned team members and a status dropdown. It spit out a working prototype in minutes. We then asked Claude to research Monday on its own, identify main features and recreate them. It added a number of other features, including a calendar.
The real magic happened when we connected the clone to an email account, essentially spinning up a customized project manager for our personal lives. The AI found Dee’s forgotten calendar invite for a kid’s birthday party (which she definitely didn’t have a gift for yet) and it added reminders to book tickets for an upcoming trip and sign a waiver for another kid’s birthday party.
Now, let’s be clear - this is not a replacement for Monday.com. This is a personalized application, it’s not deployed, doesn’t scale to lots of users, doesn’t have support, SLAs, etc. But, this shows that AI can build useful tools, and so the market for general-purpose, large-scale tools will shrink.
Nevertheless, MNDY stock crashed 20% on the news (combined with the company giving bad guidance, which isn’t surprising).
Monday.com, like most SaaS companies, needed to have products that support a variety of users at a variety of companies doing a variety of related things. When software was time-consuming and expensive to develop, the only way it made sense was to develop something general purpose enough to attract a large number of users. But, now with AI any user can create just the portions of many SaaS products they actually need both quickly and cheaply.
And yes, these companies have a “data edge,” but while some/many of these companies have your historical data, if they don’t provide services that make optimal use of that data for their customers, customers will leave.
And sure commercial software as an “accountability edge”: there are still things like support, maintenance, etc. to deal with, but already casual users have alternatives, and the products that can be built with AI will only get more sophisticated. My disruption hat says we should be seeing new companies crop up that fill the gaps, and/or that larger companies will build up small internal development support teams.
Think that’s bad? I’m only half-way done here.
Next up is the business model that SaaS companies almost always use - the per seat charge. Whether it’s the product itself that does a faster and better job thanks to AI, or customers that themselves employ AI, the trend is that there will be fewer seats of software needed. If instead of employing 5 graphic designers running Adobe Creative Suite you only need 1, you’re giving 80% less money to Adobe. The same holds true for Salesforce, Service Now, etc.
The SaaS companies are going to need to figure out better ways of charging for their services. The growth in new business they do acquire is going to be offset to a large extent by decreases in the amounts existing customers pay due to seat use reductions.
Salesforce recognizes this, and has already started experimenting:
And finally, we should recognize that we’re early days in AI adoption. Right now, AI mostly used as an add-on tool, placed within existing workflows. However, larger improvements are going to come as businesses rethink their workflows to be truly AI native. Whether this change is driven first by the SaaS providers or first by the users of SaaS remains to be seen. Bolting a chatbox onto your product is not a viable long term strategy.
This is real, folks. This is not a bubble. Sure, some AI-related companies will fail, especially those trying to push all new technologies and/or taking on lots of debt to finance their vision with no guarantee that some other company, legacy or not, will not do it sooner or better. The internet bubble wasn’t the death of the internet. And while many over-built internet infrastructure (eg fiber) too soon, I don’t see that parallels here.
It’s not like we’re going to hit a limit on the demand for intelligence.
Well summarized, @Smorgasbord1, thank you! Here are a few counterpoints worth considering:
don’t confuse “demoware” with real products: customers buy products, not demoware. Especially enterprise customers, who have the resources to fully understand their choices. A real product comes with ongoing work to advance the features and functions; it comes with support and training; it comes, typically for enterprise software, with a slew of integrations so that it can meet the complex requirements of sophisticated organizations. Most important, it comes with reliability. Who wants to be the CEO who has to explain on the earnings call that they missed their numbers for the quarter because they switched their CRM to something a couple people vibe coded in the back room, and it didn’t go well?
risk vs reward: so the risk can be high in switching to internally built software, instead of buying from a company dedicated to that software function. In return for that high risk, what are they hoping to gain? Saving on some license fees that amount to less than 1% of their OpEx? This doesn’t make sense.
take a look at the recent DDOG earnings report. Many AI companies are choosing to become customers of DDOG, spending $1m/yr or more with them, rather than build their own in-house solutions or build on top of the open source solutions DDOG competes with. Why? Because the engineers they have on their internal teams who would know how to build that kind of software are far more valuable to them building the next great AI product which they can sell to their customers, generating lots of revenue, rather than put to work building internal tools that would save them relatively little. If the companies with the most capable AI engineers aren’t wasting their time building software in-house that they can readily buy from someone else, why would it make sense for, say, a JPM or a UAL to build their own payroll system, for example? Why wouldn’t they instead put those people to work solving real problems in their core business?
I’m sure there are point solutions out there that are at risk, if for no other reason than the increased pressure on price negotiation. But that’s not the case for many of the software companies whose stocks are being cut in half right now. And I, for one, am backing up the truck in response. “Load ‘er up, it’s going cheap!”
I don’t disagree with your thoughts on support - indeed I called that out specifically as an “accountability edge” in my post. Many companies will indeed want to stick with a proven, commercial solution. However, the door is open for some companies that maybe don’t rely heavily on a product, or find that the commerical products don’t meet their specific needs, to use AI to develop their own solution. Remember, doing so doesn’t (or soon won’t) necessarily need any/many actual software engineers.
Developing software has gone from hard, expensive, and time-consuming to, well, not that hard, not expensive, and pretty darn quick - and getting more so every week.
Project management software is, I think, particularly ripe for disruption. Different companies have different project workflows, and one big task is trying to adapt existing project management software to be compatible with your company’s workflow. There’s often a lot of scripting customization going on. Even today, some companies write their own - that’s going to become even more widespread, I predict.
Even so, rolling your own software is just 1 of the 4 attack vectors I brought up.
The other 3 are:
• The customer doesn’t need as many seats of that SaaS, so, for companies that charge by the seat, like Salesforce and Monday, they’ll likely see their revenue go down. You brought up DataDog, which is interesting because DataDog doesn’t charge by the seat, they charge by usage - things like per-host, per-container, or per data ingested. That said, there is perhaps a risk that customers using DataDog (who tend to be very sophisticated themselves) can optimize their usage by internal use of AI. Customers spend $1m or more are obviously inventized to reduce those costs.
• SaaS providers will be forced to reduce their prices as new, AI-native competition comes to market. Being developed faster and with less overhead, these new companies can undercut on pricing. Yeah, it’s always hard to steal customers even under the best of circumstances, but if we’re talking growth, then it’s new business that has optionality.
• SaaS and software in general providers will eventually need to adapt to support their clients’ Ai-updated workflows. Sure, right now companies are spot-adding AI tools to their existing workflows, but the future will be redesigned workflows optimized to take advantage of what AI has to offer. This will put additional pressure on legacy SaaS providers to develop new products that fit better in those workflows.
Think about companies developing software projects. Now, non-technical product managers that can’t write a line of C++ code can create prototypes of the products they want to create. A PRD (Product Requirements Document) can be fed into AI, which will spit out a functional prototypes suitable for demonstrations, maybe even customer Alpha testing. This kind of thing used to take hords of engineers months to develop, so the barriers were higher.
As for backing up trucks, you’ll have to be pretty adept to spot which companies are at risk, which are doing the necessary new development, which are the right business models for the future, etc. Is is really a co-incidence that Monday got partially replicated by a couple of reporters in a hour using AI AND that the company gave soft guidance for the future?
With AI-developed products that can analyze not just your code but your running infrastructure coming, can DataDog protect its margins? For instance, is this really not indication of threats to DataDog’s raison d’etre? :
Similarly, Google’s BigSleep found 20+ real-world bugs in well-used projects like FFmpeg and ImageMagick.
So, tell me, who’s going to spend big bucks on legacy bug-finding software packages any more? And this is just the camel’s nose under the tent - it’s going to spread across many industries.
I appreciate your response, @Smorgasbord1, and here’s some more back at you:
Your strongest point, I think, is on pricing. Customers will still pay for value received but the way they pay may need to change. The process of changing price structures may be disruptive for some companies and that could certainly show up in their stock for several quarters. For many people on this board, that would be a problem. If you believe that AI is going to lead to broad employment cuts, that might give you pause.
To be clear I see AI as bringing disruption to some parts of the job market (entry level jobs in knowledge work in particular) and boosting productivity in other parts. I don’t subscribe to the “AI is going to put tens of millions of people out of work” thesis. That’s not how tech-fueled organizational change happens, in my experience.
Will we see a bunch of new entrants attacking established software companies? Sure, but in which markets are they likely to succeed?
Here’s the set of criteria I laid out here and elsewhere several years ago for how to find software winners in the AI wave:
“1) existing set of customer relationships: distribution is usually the hardest part of the journey so let’s identify companies that have that box checked;
access to tons of proprietary data that can be transformed into valuable information by AI: this is the heart of the transaction: I am looking for companies that can buy AI technology and use it in a transformational way, to spin straw into gold; [why “proprietary data”? Because if it’s not proprietary, then anyone can do it, so there’s no moat]
a track record of internal innovation: Building new products and new product lines is hard, and most large companies have forgotten how to do it. They can only do bolt-on acquisitions, which will really limit their ability to respond to this huge opportunity.”
Companies I’ve bought with this thesis in mind:
INTU: dominant market positioning, proprietary data, great internal culture of innovation.
Companies I’ve been buying lately because I think the selloff is overdone:
NOW: Strong enterprise distribution. Tight integration into customer business processes makes them hard to dislodge. Innovating quickly to stay ahead.
CSU, TOI, CHG: these are serial acquirers of small software businesses in niche industries. The companies they buy have built dominant market positions in markets nobody else ever cared about. Yes, somebody could probably use AI to code up a bus dispatch package for a municipal bus system in an afternoon - but why would anyone buy it over the established player that knows everything that is to be known about how to make their customers happy? To save a little money? That’s not enough motivation for the risk. Also these companies create most of the value via M&A. Prices are way down (they use comparables in the public market, for example) and buyers will be fewer (a good way for a PE fund to lose their investors is to try to convince them that their strategy of buying established software businesses is a winner).
DUOL: this has never been simply a business about teaching people languages. For half their paying customers, learning English is the motivation - but they are serious motivated learners and they are willing to pay for use of a learning environment which will help them actually achieve their learning goals. Online learning has been available for many years but course completion rates are around 1%. So yes, you could learn English for free somewhere, but will you? Probably not.
The other half of their customers are people like me who want a productive way to “waste time” and will pay for a good experience doing that. This is less about learning a language and more like playing a social game.
Advantages DUOL has for both audiences are 1) distribution and brand; 2) proprietary data - the largest repository of data on user behavior in the context of social online learning; and 3) a strong culture of innovation.
The market has knocked 75% off the share price (could be out of date) and that’s way overdone in my mind.
I see a huge market overreaction in software stocks to insights that were readily available as soon as AI emerged as a major driver of change in the software development world. That was many moons ago.
I am in disagreement with the conclusion there is going to be only a minor impact on our SaaS software companies. Years ago, programmers were required for developing my Excel Investing Database with 100 columns, 250 rows, 1000’s of data cells and 2000 cell’s with analytical conclusions, I did this with no formal training in Excel and no knowledge of programming.
AI will deliver an enormous change to traditional SaaS space, many companies falling by the wayside, others reinventing themselves, but any void will be filled in with yet undefined opportunities. I will attempt to reflect this condition in the marke up of my portfolio.
I admit that I haven’t specifically researched current AI code writing, although 10 years ago I was involved with one of its precursors … but there is an old observation that typically software follows an 80-20 or even a 90-10 rule, i.e., that one can write 90% of the software with 10% of the effort and the other 10% takes 90% of the work. I suspect that AI is writing the easy part, which looks like a nearly done system, but which turns out to have some really hard bits that are not there yet. Could be there are offerings of vertical-specific hard bits.
That’s a fair summary of how Duolingo used to lead the pack, but as a long-term user, I see things very differently now. I’ve started using custom Gemini Gems for my language practice and the experience is just on another level.
While you could argue that Duolingo is in a great position to use AI, they haven’t done a very good job with it so far. Plus, I’m already paying for Gemini. I don’t see the point in paying twice when I can customize my own workflow and get better results without the extra subscription.
I won’t argue about the valuation since the selloff was so big, but I think the AI disruption is real. When an individual can build a better learning setup, the outlook changes. I definitely wouldn’t put my money there.
What you’re saying, @SailingDev, is not a surprise: for people who want to learn a language and who can afford a high end personalized AI experience, there are going to be better solutions than DUOL. Despite public perception, that’s not actually one of their core markets.
One market they serve is people who want to learn English and achieve a recognized certified result (DUOL’s certification is now widely recognized); and who cannot afford to pay for Gemini’s best solution. When Gemini’s free solution gets better, the next barrier will be certification of learning.
They also serve people who want a “productive waste of time”. Gemini and other AI may eat at that market, in the same way that casual gaming eats into it as well, along with many other players. They’re all competing in the “attention market”.
It’s also worth noting that in the last conference call, the DUOL CEO went into some detail on how AI was dramatically reducing the cost of developing curriculum for them, and how they feel they can now aspire to actually teach language learners to the point of proficiency, beyond English. So while it hadn’t been a goal before, it is becoming a goal – not something they currently do.
Whether or not they succeed is immaterial to my investment thesis at 80% discount to the 12 month high, for a business with last 3 yr revenue CAGR of 44% and forward 2 yr revenue CAGR of 30%.
I realized about a month ago that SaaS was heading for trouble. I’d been building a program to auto-calculate Manual J and Manual D for my cousin who works in HVAC. Manual J handles the heating and cooling load of a house, and Manual D covers the ductwork. After spending 10–20 hours developing a solid product, I decided to set up a ChatGPT GPT as a quick comparison to see why mine was better. It took me five minutes, and it was a 95% solution. My cousin could even tell it how his region deviated from the national standard, and it adjusted accordingly.
I’d originally planned to launch this as a SaaS where contractors upload a PDF of the building and get a Manual J and D instantly. But once I saw what five minutes of prompting could do, I couldn’t convince myself any contractor would pay for a dedicated tool when AI keeps getting better and can already handle this with minimal setup.
To be clear, I think some SaaS products will be fine, especially in fields that require compliance or auditability. But SaaS aimed at small businesses is going to look radically different in two to three years.
Thanks, @drew1618t, for bringing us back to the central point:
A product is not a business.
AI can build some kinds of products – a single function product like you describe sounds like a great application for AI. @Smorgasbord1 earlier in this thread brought up a couple reporters building a dashboard similar to Monday.
Those are only products. There’s much more to building a business, let alone one that takes share from established players.
As someone who has built numerous software products over the years, and built large businesses based on those products, as well as invested in software businesses for many years, here’s what I think you can look for to help you distinguish between the software companies that will likely benefit from AI, versus the ones that are threatened by AI.
The analysis lens I suggest using comes from Hamilton Helmers’ Seven Powers, which talks about seven different ways that businesses build persistent competitive advantage. Here they are, along with my commentary for software businesses and AI:
Scale Economies: unit costs fall as volume increases. For SAAS businesses, the incremental cost of serving another customer is often thought to be $0 – but when you consider the payback period for customer acquisition costs for a typical SAAS business, it is far from $0. Payback of 2-4 years is not uncommon. This should serve as an indication that there are many more factors going into building a successful software business than just the cost of the software. That point aside, I don’t think this is a primary field of competition for existingplayer.SAAS vs upstart.ai. If anything, it would argue in favor of existingplayer.SAAS, with thousands or tens of thousands of existing accounts, where the CAC has been paid off already.
Network Economies: the product gets more valuable as more users join. This is an important part of my thesis with INTU: they have tens of thousands of small and midsize business customers, which means their AI engine can learn things that carry across businesses; and they can put that learning to work on behalf of all their customers. A new entrant will struggle to acquire the same learning (they have many fewer customers). That’s not the typical application of network economies but I think it’s relevant for AI learning. Note that it’s related to Process Power (below).
Counter Positioning: A newcomer adopts a business model that an incumbent can’t copy without cannibalizing their existing business. This is one of the central points that @Smorgasbord1 is making: if the cost of developing the software for a new entrant is very low, they can charge much less for their solution. Incumbents will struggle to match on price because they can’t jeopardize their existing revenue. This is certainly a risk – but there are often other considerations at work for customers besides just price.
Switching Costs: Customers incur real or perceived penalties—money, time, data loss, retraining, risk—when changing providers. An example: NOW empowers users throughout a customer organization to use it to build workflows. A competitor would have to make it worthwhile for the customer to retrain all those users and rebuild all the workflows that are automated via NOW. That’s a non-trivial task.
Branding: A trusted reputation built through consistent experience and messaging creates demand preference and often a price premium beyond functional features. Enterprise software businesses typically don’t build strong brands, but of the software businesses I’ve mentioned, INTU and DUOL both stand out for brand.
Cornered Resource: Exclusive access to a valuable asset—IP, talent, data, supply, distribution, licenses—that others can’t get on comparable terms. This is one of the primary attributes of the kinds of niche software businesses that the serial acquirers I mentioned earlier, like CSU, go after. My earlier example was about bus dispatch software for municipal transit operators. There are only so many people who know how these systems work within transit companies, or who can explain how the operator in, say, Calgary, wants to do things differently than the operators in Toronto or Ottawa. Detailed insider knowledge of a small industry is a perfect example of a cornered resource.
Process Power: Unique, often tacit ways of working that improve quality or lower cost and get better with scale and learning. Embedded in culture and routines, these processes are complex, path-dependent, and hard for competitors to copy. AI brings a lot to the Process Power arena – but it favors the incumbent, with access to all the customer’s historical data and the ability to bring insights to bear that a competitor can’t match. In a product shootout, bet on the contestant who brings the bigger gun. DUOL is a good example of this: they’ve been learning how to get users to persist, to keep coming back day after day, to use their product. Contrast that with the online training course completion rate of around 1%, which is typical for online courseware.
You have to know the businesses you’re investing in to apply this kind of analysis, but I know that many of the posters on this board are deeply immersed in the businesses they invest in. I find it’s generally quite helpful to think about my portfolio in these terms. If I don’t know enough about the business to do this kind of analysis, and I don’t have the strong curiosity to learn, that’s a good indication that it shouldn’t be in my portfolio.
At any rate, I hope it’s clear that while AI will have an impact on all software businesses, what kind of impact that will be is very much dependent on the individual business.
That’s all backwards-looking: Disruption is here and the future will look very different than the past.
It’s not just that. You talk about “building a business,” but competition will, in some cases, come from in-house developed solutions that replace bought & configured solutions. One company I worked at replaced Jira with its own in-house project management solutions. What we wrote didn’t do near everything Jira did, and we didn’t have to turn it into business - it just did what we needed. Your example of “dispatch software for municipal transit operators” could be an example where the in-house expertise can be leveraged by AI to create the desired workflows in-house since they no longer need to know how to write software.
It’s an interesting thought to consider how a company may feed its existing customer data into its AI to improve it, but I wonder what customer companies feel about that? As a customer, I’d want to be compensated for use of my data to improve your system for other customers. Maybe this gets buried in contracts for now.
Switching costs only matter for existing customers, and I assume that companies in which we want to invest are growing at high rates, which means they have many new customer conquests, not just expansions of existing customers. Hence, switching costs are the same for these new customers.
I don’t doubt that some babies are being thrown out with the “AI eats software” bathwater. I do think it is hard to predict which existing SaaS companies will leverage AI well, change their business models, and find new sweet spots of value for their customers, as well as making more money than before.
It’s unfortunate that Clayton Christensen is no longer with us - it’d be interesting to read his take on the AI disruption and how good companies not just avoid failing, but thrive.
And yet AI logistics company Algorhythm (RIME) shot up the past two days, while CH Robinson, XPO, Dominon Freight slid between 5% and 24%. RIME claimes its “SemiCab” freight platform is deployed with live customers and is enabling a 3X to 4X scaling of freight volumes without adding headcount. Basically each individual operator can now manage 2k loads/year instead of the 500 average previously is the claim.
So, yeah, AI finds its way into even niche applications.
The thing about moats is, once you build one, the castle is harder to attack, but it also can’t expand across the landscape.
Dumb metaphor maybe, but this is a growth investing board, so I really think we should focus on growth. We’re not in the business of looking at DOCU or TWLO or others now with ~10% YoY growth and trying to catch falling knives (ie, figuring out when the price gets low enough). It’s similar for MNDY which will soon be growing at 20% or less. What’s the right price? Hard to say…whether there’s a moat or not…because whatever the “right price” is, how much higher will it be 2-3 years from now?
It is of course possible to catch falling knives, but that’s not what this board is for.
So I think our question is not about moats, or about who will survive. Our question is who will benefit and grow. Not who will keep their customers, but who is in the right stage of growth that they will get more and more (and larger) customers.
Reddit is interesting. I guess they have a moat, because while an AI (or a human) could easily create something similar…the hard part is getting users to show up. But the real question is how much more they can grow. They have hundreds of millions of eyeballs already, but maybe they can sell more ads, charge more from them, etc. It’s not my highest confidence position, because that will run out at some point, but with 70% revenue growth this past quarter, it’s a decent bet IMO.
Does AppLovin have a moat? I dunno…if someone can recreate their Ai-powered platform, they can probably offer a similar dollar for dollar return for their customers, but people have to know about it. AppLovin just grew revenue at 65%+, so while any human or AI could have created a competitor, they either haven’t achieved one as good or better than APP, or they haven’t marketed it well.
Maybe even more exciting are the direct beneficiaries of AI like ALAB and CRDO. They are growing even faster. Many others too (LITE? MU? SNDK? etc).
Duolingo was mentioned. I tried to catch the knife on it myself but gave up and took a loss. I just have never understood how they’ve kept growing like they have. Now that growth is starting to slow, is it really a good time to try to figure out the “right price?” What are reasons to believe the growth will continue, even as DAU growth decelerates?
I don’t think we can figure out which kinds of companies (if any) are undisruptable. But we can talk about a company’s growth drivers. We can debate a company’s prospects.
I think the actual moat for Reddit is the history it has - the data that Google wanted to tap at $60M/year. Besides the money, this also drives users to the Reddit platform, so drives even more fresh data coming in. This all seems good, but I wonder how much more room to grow there is, whether the future is actually using AI to summarize and rank the results of Reddit’s data. Would that AI be coming from Reddit or from Google? I don’t know, but I’d be surprised it Google wasn’t eyeing that closely.
Once you’re getting the data via AI, getting to the Reddit thread sources seems much harder, so I would think this is less likely to drive users to the Reddit platform, and so Reddit still needs to figure out how to keep that growing. Yes, I know today AI results include links to the source(s), but clicking through those, especially when the AI is (as it should) summarizing from multiple sources, that seems less and less likely.
What, if anything, is Reddit doing in terms of either directly adding AI to its business, or rethinking its business model in terms of being AI first, if not AI native?
I’m not here to defend the moat idea — I think you’re right that these “clicks” could be disrupted. But what we do know is, Reddit data is already summarized in AI and search “results,” but that doesn’t appear to be keeping clicks from growing now. Also, I don’t expect that growth to really continue much longer. Revenue growth will have to come from other ways like ARPU and targeted ads (maybe using AI, haha).
Also, those AI and search “results” exist because there’s discussion on Reddit in the first place. This is something that seems like it will persist. I could be having this discussion with chatGPT, but I’d rather have it with you. Maybe AI (agents) will jump in and augment the discussion here or on Reddit, but I feel like we’ll still want to contribute.
There are many variables at play, and it’s hard to know which will dominate in the future.
My thinking about AI search results is that less and less they’re going to be single-source dependent. Today, Gemini might summarize a couple/few Reddit threads to answer your question, but in the near future, those Reddit threads are just a piece of information, with additional information coming from Facebook, multiple single-purpose dedicated forums, product reviews, YouTube videos, etc. And so, the AI will either not include all the links, or bury them in some relatively hard to find and isolate manner. Right now the links are needed since we (rightly) don’t trust the AI, but as the AIs get better, we’ll care less and less.
Of course, ads will remain a significant contributor to revenue, but who will be capturing that? Google’s paying $60M/year to Reddit, so I assume Google’s ads are Google’s revenue. Maybe the deal requires Google to link to the source Reddit threads, but again, that’s getting further removed from the process. I think about that old meme:
But with AI, we’ll get disconnected from the source - and hopefully AI will sort out bad information for us so we’re not seeing it in the first place, meaning we’ll want to connect less often.
Targeted ads using AI is definitely a growth area. Depending on what data you let Google keep (eg, if you’re using incognito windows or not), Gemini will be able to figure out what you’re doing and present ads that actually make sense. Today, for instance, Amazon does a lousy job of this - I may search and add to my wishlist different versions of items I’m looking to buy. But, after I buy one, you’d think Amazon would stop showing me these items - but it doesn’t. If anything, it shows me more of them, as if I need two torque wrenches or yet another toilet flapper valve. What it should do is look at all my purchases and figure out what projects I’m working on and suggest either other replacement parts, or maybe a whole new replacement thingy.
Can Reddit use what you’re commenting on as well as reading to determine what you’re actually doing to suggest either other threads to read and/or ads for products you might want to buy? I sure hope so. Have they done this yet?
I thought their click growth is evaporating, growing their most valuable audience by 5% YoY, down from the 24% growth last year (logged in US users). That growth is so bad they are going to stop reporting it going forward, claiming they gain the same value per user on a per click basis.
I felt like RDDT is moving the goal posts. A year ago the CEO said “We’re working on converting logged-out to logged-in” and after a year of not being able to do that he says “phase out reporting on logged-in and logged-out later this year”.
