 “Every generation laughs at the old fashions, but follows religiously the new.”― Henry David Thoreau, Walden In my last post, we talked about why AI automation is the new industrial revolution. At its core, such revolution represents a major landscape shift that will give birth to a new breed of tech companies — what I called the AI-first species. Just like any emergence of new species in mother nature, AI-first companies, although share some common traits with traditional software/ SaaS counterparts, carry very different characteristics. More interestingly, the new species behave, in many ways, more aggressively than the incumbents. In this post, we will examine a few intriguing aspects of this AI-first specie: from the gross margin, product form, business moat, to ultimately the billion-dollar question — how do we adequately measure the size (value) of these AI-first companies. Let’s start. 1) Unlike typical software companies, the “AI-first” companies might have a non-software-like low margin (at first). And it might be a good thing.  Investors love the 80% margin business. So do traditional software companies. Over the last decades, we have seen powerful high-margin software companies — such as Salesforce, Oracle, and SAP — created and dominated the respected fields for a long time. One of the advantages of having a high-margin product is that it allows you to stack up on your well-oiled go-to-market machine continuously. In other words, it enables the incumbents to be extremely good at doing what they have been doing. And it’s a comfortable situation. For the new species of “AI-first” companies, they wear the cover of “low margin,” at least initially, to enter different business territories. And this is not by choice. It’s the embedded nature of AI companies at its initial phase for the following two reasons: a) In the field of AI, it’s easy to get to 80% accuracy; it takes investment and resources to get to 95%, but it’s near impossible to get to 100% because of the diminishing returns of solving the last 5% edge cases. Because of such dynamics, AI companies still employed humans on two fronts — data labeling/training and edge-case solving. And these humans are often data scientists who are not cheap. The cost of such eats away margins. b) Contrary to popular belief, AI computing actually costs more than the traditional one, especially as more training data accumulates *and* as your algorithm is getting more complicated. Ironically, AI companies’ cloud computing costs are likely to be much higher as the business becomes more successful. And this is not even counting the often-ignored fact that AI companies, unlike traditional software ones, will need to retrain or “refresh” their data-algorithm periodically to refresh the latest reality (e.g., mapping data, latest business wording, etc.) It’s worth noting that both constraints above are not unsolvable, especially as we make more progress on the improvement of AI-computing infrastructure. It will just take some time. Therefore, these AI-first companies will usually carry a non-attractive low margin, at least initially. Believe it or not, this might be a good thing, which leads me to my second point. 2) “AI-first” species’ low margin is its deceptive camouflage. Underneath, it is often a highly attractive ROI solution or service. It should get incumbents worried.  As Clayton Christensen — author of the classic book “Innovators’ Dilemma” — would point out, disruptive new entrants usually come in the form of “non-attractive low margin” business that’s “easy to ignore.” Such dynamics happened when Minicomputer (DEC, Wang, etc.) disrupted Mainframe Computer (mostly IBM), and then quickly Personal Computer (Microsoft, Apple, etc.) made Minicomputer absolute. In both cases, the emergence of Minicomputer and PC came with audiences that were rather niche but were proliferating nonetheless. The nature of a “niche” market made the “customer-driven” incumbents turn the heads sideways because their main customer base didn’t demand either Minicomputer or PC at the time. More importantly, the “low margin” characteristics of the new entrants essentially stripped incumbents’ incentives to be “lean” and to be innovative again**. Now with such framework in mind. Let’s look at what’s happening now. AI-first companies are attacking new territories and replacing older software the same way illustrated in the PC era. Yet AI-first companies are making the moves more aggressively in a pattern we have rarely seen before. So it’s incredibly tricky and dangerous for incumbent tech companies on two folds: 1) Unlike traditional software companies that usually offer one specific value (i.e., Salesforce for CRM, Box for storage, etc.), a robust AI-first company might propose itself as a full-service company with better ROI for customers. For example, we have seen AI companies “disguise” as a virtual lead-gen / sales force service agency that could potentially replace companies’ in-house business development or sales staff. Under the hood of these “service” companies, most of the operations are automated through AI. Therefore, it results in lower monthly costs to the customers, even though the initial COGS might be higher than that of AI-first companies. On top of that, it would be easier to match sales/lead gen performance against the cost: it’s not hard for an executive to choose $100K on sales staff vs. $20K “service” fee to this AI solution. Other territories that are being trembled are: customer service/call centers, executive assistants/ scheduling, and legal document review/ compliance, just to name a few. The customers who are responding well and quickly are often the ones who value cost-efficiency the most. They are usually startups and SMBs. They are AI-first companies’ first “nitch market” that is often underserved (and overcharged) by incumbents. They are the beachheads. By establishing an ultra-strong ROI case on a particular business function (i.e., sales, call center, EA, etc.), AI-first companies are quickly building up the beachheads before attacking incumbents’ primary market — fortune 500 customers. 2) On the incumbents’ side, the existing high margin software business offers little incentive to explore AI-service-like products with a lower margin. More importantly, much incumbents’ existing customer base — the stable large-cap companies — might be willing to try out some AI-first products, but they do not demand them. Why?  First, the mainstream market, by definition, is conservative and usually likes to wait until fully adopt a new product. Second, thanks to software incumbents’ decade-long effort in building high-switching costs to their products, the customers are even less likely to be proactively looking for new solutions. Therefore, the “conservative” customer base, plus the high margin “lifestyle,” will immobilize incumbents to react and defend the inevitable take-over from the new breed of AI-first companies. Such is life. However, this is not to say that AI-first companies don’t have shortcomings. Other than the “low margin” factor as mentioned above, AI companies will have to figure out a few things, including the sustainability of the moat. This lead to my third point. 3) Unlike typical social network or marketplace companies, the moat for “AI-first” companies might not be deep. Therefore, the initial data advantage and an ongoing low-cost data acquisition would be the key.  Because AI algorithms are widely open-sourced, the keys to building a defensible business are proprietary data and the method of keeping acquiring quality data cheaply***. Because of such dynamic, it’s not hard to foresee the successful breed of AI-first companies will be heavily vertical-integrated and even service-oriented (at first.) Therefore, there won’t be such a thing as horizontal AI (with the exceptions that maybe Alphabet or Microsoft can morph into horizontal AI players). Instead, we will be seeing AI companies build their data pools, users, and brands in specific verticals such as healthcare, payment, media, automotive, and manufacturing. We are still in the early days. I don’t think we have figure out the best practices of obtaining sustainable data moat yet. On the one hand, it could be a capital intensive endeavor as we have seen Google/ Waymo invest billions in its self-driving AI as it keeps updating the mapping and driving data on its own. On the other hand, we have seen some startups being creative by striking partnerships with hospitals, payment processors, and OEMs to get the initial data advantage started. However, is either approach sustainable and scalable? We don’t know. We haven’t figured out the formulas as much as we have on many of the SaaS calculations. One might imagine a very different pricing model that is offered by AI companies. We have discussed AI-first companies might charge customers on a service basis (initially). But the pricing model might eventually mutate into data-discounted pricing where the customers might enjoy much of the discount in exchange for the training data they contribute. So that would be very different than the typical SaaS companies’ per-seat model. Parting thoughts: the next mega-giants?  How do we, the investors, properly evaluate these pricing models and the underlying business created by this new breed of “AI-first” companies? I don’t have the definitive answer, but I would expect it to be different from a typical Cloud/ SaaS multiple. My guess: In the coming ten years, we will be witnessing a new breed of tech companies that are much more automated, intelligent, and ever more embedded in business functions. And this unleashes value. The market valuation for these AI-first companies will likely be much higher than the “1-billion-dollar unicorns” over the next decade — perhaps we will see a few trillion-dollar companies across the US and China. What would be the AI-first equivalents of Amazon, Apple, and Alibaba? If history is of any guidance, we should be well prepared by studying the trend and then embrace it. Fully. ** Another example would be Netflix vs. traditional media companies. Started as a DVD rental business (i.e., you actually have to mail out physical DVDs and then expect consumers to mail them back via US Postal Office), Netflix’s original business model was heavy with DVD warehouse and intense sorting and mailing labor operation, thus, relatively low margin business. Because of the initial “low margin” characteristic, many media senior executives were blindsided. The gravity of short-term quarterly profit pulled them away from the long-term threat of Netflix. Many in the media industry failed to imagine how Netflix could morph into a digital-first streaming product that left giants such as Disney, Warner to play catch-up.
*** This is assuming everything else being equal. It’s worthwhile to point that compared to traditional software companies, AI-first companies might have to be better at sales and marketing as the customer market will experience a learning curve on its own. Within the AI companies — which are often staffed with engineers and data scientists — the ones who have the best GTM approach /team will dominate, since the end market doesn’t care about the technology but rather the ROI-proven solutions.
0 Comments
|
