Playing To Win
Investment Strategy & Artificial Intelligence
Verticality vs Horizontality
Last week’s piece discussed a way to think about Artificial Intelligence (AI) as a mechanism for advancing knowledge quicker from vexatious mysteries to ubiquitous heuristics. In this week’s Playing to Win/ Practitioner Insights piece, I am turning to a way for companies to think about their investments in AI, which is important because they are all investing aggressively in the domain. It is called Investment Strategy & Artificial Intelligence: Horizontality vs Verticality.
Wasted Investment
When anything arrives new on the scene in the world of business, vast sums of capital will inevitably be wasted because it starts as a mystery — which, as discussed last week, is at the top of the Knowledge Funnel. A knowledge domain is a mystery when we don’t yet know how to think about it. AI is a mystery because we don’t know how it is going to play out and are just starting to figure out how to think about it (the subject of last week’s piece). When it is exciting and new, a business mystery is a flame to which capital moths are drawn.
For example, automotive OEMs are completely wasting billions of dollars in autonomous vehicles because it is the exciting new thing. Ford invested $1 billion in Argo because autonomous vehicles were ‘the future’ and Volkswagen bought into Argo at $2.6 billion. Then in 2022, when they figured out it was a black hole, they shut it down five years after the first investment (the shutdown was a wise decision, by the way). Microsoft and LG wasted many billions on smartphones when ‘you had to be in smartphones.’
This is not new. Investors wasted billions (in current dollars) in the early days of rail transportation and automobile manufacturing. Most recently, billions were tossed away in the early days of the Internet. So, it always happens. When the new new thing comes along, vast sums will be wasted because if you aren’t willing to invest, you can get left behind — FOMO in spades!
The real enemy is not inevitable waste; it is avoidable waste. I really hate pissing capital down the toilet when there is no chance of it ever producing a positive outcome.
The Context for AI
To have a chance to identify avoidable waste in AI spending, it is important to understand the context for AI. It is part of the high-fixed-cost world of modern tech, as I have written about before. In modern tech, especially in software and Internet services, the cost of producing and selling an incremental unit is zero or close to it. These are businesses dominated by their fixed costs. The company spends often an enormous amount on creating the first unit of its product — working software — and subsequent variable costs are miniscule.
This means that scale is an imperative. The greater the scale, the more broadly the fixed costs are spread and the lower the overall cost position of the company’s offering. That is the encouraging side of scale dynamics. The intimidating side is that in the definable domain in which you compete, if you don’t have the greatest scale, you will have a cost disadvantage to any company that does.
Then the only way to survive competitively is for your offering to be demonstrably superior to competitors in the eyes of your customers. But that is hard to do because you will have fewer resources to spend on your differentiation because your underlying cost position is inferior.
Apple iPhone provides an excellent lesson on this front. Apple’s iPhone business is legendarily successful. But how could it be? It was never the market share leader in smartphones until 2023, sixteen years after its release. Samsung had #1 market share for the previous twelve years. And it categorically lost the operating system battle to Android, with iOS facing a 71% to 28% deficit at last count. Shouldn’t have Samsung and Android been able to spread their fixed costs wider and crush iPhone?
No. It is because fixed costs are both incurred in and spread across dollars — that is revenue dollars, not unit sales. And despite having only 20% unit share, iPhone has 50% revenue share, as much as everyone else combined and three times next best Samsung.
But that isn’t even the truly relevant story. Apple doesn’t compete in the smartphone business; it competes in the premium smartphone business (the definition of which keeps changing as phone prices keep going up, but the latest statistics tend to use $600+ as the definition). In that business, iPhone has 71% market share, over four times the closest competitor. And that is why it rakes in approximately 85% of smartphone profits. That is, the companies producing 80% of the units get to share 15% of the profits while Apple hogs the rest!
Apple, with its vastly higher revenue share, can spread every fixed dollar of R&D, design, and advertising spending across its business such that investments it can make with barely a thought would be financially painful if not entirely infeasible for its competitors. As a result, Apple can keep enhancing its differentiated position in premium smartphones. The biggest threat to iPhone is not a competitor, it is hubris — which brings down all great companies if unchecked.
The Key Distinction
The key thing to understand in any business, but more so in software/Internet services due to the high fixed cost cost structure, is the degree to which the use-case is horizontal or vertical.
At the extreme horizontal end of the spectrum, the use-case is common to all customers (whether companies in B2B or individuals in B2C). A software product near the extreme horizontal end of the spectrum is Excel. There is a use case for it in every company in the world. Not so far out but still near the end of the spectrum is SAP — for all medium to large size companies. At the extreme vertical end of the spectrum is a use-case that is unique to a single company, for example, a piece of customized software written for a company’s unique use. A case in the middle but nearer the vertical end would be a point-of-sale system for taking food orders for wait-staff in restaurants. It is for restaurants only, not all companies as with Excel or SAP.
The distinction is important because anything toward the horizontal end of the spectrum will be dominated by companies whose business is to sell the offering to as wide a set of customers as possible. Why? It is because their competitiveness will depend on spreading its fixed costs across as wide a customer base as its strategy can support. If you try to create spreadsheet software or an ERP for use of your own company only, you will waste every single dollar — because every company needs one and someone will invest massive capital to build and widely distribute a great one.
If, on the other hand, it is something only your company, or a few like yours, will ever use, you will need to build it yourself. It may be expensive, but it could well be worth it if it is important to your competitive advantage.
What Does that Mean for AI Investment?
AI is a whole stack of elements; not one monolithic thing. The stack is described in many ways by AI aficionados. At its simplest, the stack is seen as Infrastructure, Models and Applications. But you can get much more complicated (e.g. here and here) if desired.
There will be two kinds of successful investments in the AI stack of elements. One kind will be an investment in an element of the stack that is more horizontal. To be successful, that investment must aim to be the biggest player in that element with the biggest R&D budget and the broadest distribution across users. The other kind will be an investment in an element that is more vertical that is aimed at creating something uniquely customized to you that helps your core business outperform your competitors. Investments in-between will fail. Moderately high investment with moderately broad distribution will get killed by very high investment with extremely broad distribution and by narrowly targeted investment with no intent to distribute.
My prediction is that most parts of the stack are going to be primarily horizontal — like much of the world of software. Sadly, companies will waste their money on elements that are truly horizontal but aimed solely at their company and not on a broader market. And given the huge investments flowing into AI, things are going to go horizontal faster than with previous classes of software.
Practitioner Insights
Because it is the hot new thing in business, everybody is talking about AI— to the point of hyperventilating. The result has been the moral equivalent of the declaration of force majeure — an excuse to spend like a drunken sailor on anything related to AI.
Don’t. Instead think first.
Ask what aspects in which you are contemplating investing are largely horizontal. Do lots of companies need the utility/tool in question? If the answer is yes, do not attempt to build it yourself, even if you think you can. Make a choice like P&G did in the mid-1980s by being one of the first large American companies to implement an SAP ERP system. P&G was easily big enough and sophisticated enough to build its own ERP. But it was a horizontal element — so horizontal that now SAP advertises being installed in 99 of the top 100 companies worldwide. When it is substantially horizontal: get help!
What other aspects in which you are contemplating investing that are largely vertical? If you have a unique database — build it in a way that suits you and your distinctive use for it. If you need to ask your LLM/AI specific questions in a specific way to get the unique outputs you need, invest in working on that yourself.
However, even vertical aspects can have horizontal elements. Your data may be unique. But if the kind of database that you need to organize it is similar to that of many other companies, a horizontal provider will figure out how to provide that service cheaper and better than you could do it yourself.
This implies that it is important to invest in becoming skilled at figuring out who are going to be the horizontal winners in the various elements of the AI technology stack. You should aspire to be like P&G when it was able to pick the eventual horizontal winner in ERP software.
That is the number one reason why you will need AI expertise in your company. It is not to build AI things, but to figure out who you need to build AI elements for you and how to manage those providers. If you aren’t good at that, you will get both ripped off and left behind. But if you are good at it, you will be able to get the most out of AI without breaking the proverbial bank.
