Will AI lower profits?
There is a lot of controversy about whether or not AI has a any overall impact on businesses, either by reducing cost or increasing revenue.
But let's side-step that debate for a moment and simple ask what would happen if AI actually delivers on it's promises. What happens the day after AI actually gets good.
Most companies seem to somewhat blindly assume that any kind of innovation is good and that all increased efficiency somehow magically leads to more profit. It's not that simple.
Why Companies Think AI Will Make Them More Money
There seem to be two schools of thought on how "AI" is supposed to make more money for businesses. The first is that automation will reduce cost and therefore increase profit margin. The second is that adding "AI features" to a product will sell more of that product or enable the business to raise prices.
These would generally make sense when the innovation originate internally, i.e., when one company innovates while others don't. The competitive advantage enables you to either gain market share or raise prices. But that's not what's happening right now. Large language models are equally available to all companies and some of the low-hanging fruit is so obvious that everyone is doing it. Adding a sparkly icon to your app is hardly a differentiator at the moment.
What all this means is that innovation is happening at the level of industries, not at the level of individual companies and that dramatically changes the math.
Let's dive in.
Competition keeps prices low
The mechanism that keeps profit margins roughly constant is that competition will drive prices down to the point of the lowest acceptable profit margin. What that profit margin is depends on factors specific to an industry, like the cost of raising capital in an industry. There are exceptions like markets with near-monopolies. For example, Google could get away with increasing prices for every workspace because they are effectively only competing with Microsoft who are doing the same.
This means are generally two conditions under which lowering costs increases profits. Option one: you have pricing power, that is, the price of your product is not set by your costs and you can just keep prices high while you reduce costs. This is the case where you are somehow insulated from competition. This is true, for example, for some desirable brands and industries with near-monopolies.
Option two: You happen to be in an industry where lowering price increases demand by a large enough amount that increased demand offsets the lower price and, as a result, total revenue increases.
Caveat: Buyer behavior is complex. Buyers are people and do not behave rationally. There products where demand is constant when prices change. There are even examples of products where demand increases when price increases. It also takes time for new competitors to enter a market and for prices to react to changes in cost.
The point of this article is to make the argument that improving efficiency or lowering cost does not obviously lead to higher profits for companies. The opposite can be true, and under reasonable assumptions, the increase in profits can be 10x smaller than one may naively expect.
How demand responds to price
The relationship between price and demand is governed by the demand curve. This is usually plotted as the price the market is willing to pay for a given quantity:
In this figure I've plotted a demand curve with constant elasticity and the additional assumptions that there is a maximum price the market will support and a maximum demand beyond which people will not consume even at zero price.
How markets set prices
We usually assume that the marginal cost of producing one extra unit of something goes up as an industry tries to produce more units. This makes sense as producing more of something requires more of some input resource. The increased demand for the input resource pushes up the price of the input and therefore the cost of the final product.
This applies even in industries where we traditionally think of the marginal cost being zero, like software engineering. The marginal cost of producing an extra copy of existing software is nearly zero. However, the cost of producing new software packages, or extra features does go up with quantity as this would require attracting more people and thus raising salaries.
The actual quantity produced and the corresponding price is where the supply and demand curves intersect.
The plot shows two supply curves with two different assumptions about the cost of producing a product. The more efficient production process leads to lower prices and higher demand.
Revenue and profit
The key question is if higher demand at lower prices will increase or decrease profits. This really depends on if the demand increase is large enough to offset the lower prices. The ratio between demand increase and price increase is called the price elasticity of demand (PED).
If we assume that competition keeps profit margin to a constant percentage of revenue. That means that profit is proportional to revenue and in order to see if profit will increase or decrease we only need to check if revenue will increase or decrease. Revenue is given by
Where R is revenue, P is price and Q is quantity. The combination of P and Q must always lie on the demand curve so we can compute R as a function of Q by reading the price, P, from the demand curve.
The figure shows revenue as a function of quantity for three different values of elasticity. Over the "normal" range where price and quantity are not capped, revenue either always increases for greater quantity or always decreases for greater quantity depending on the value of the PED.
If the elasticity is greater than one, revenue increases, if it's less than one revenue decreases.
This worked example also shows some examples of how much profit actually increases depending on the value of the elasticity. Let's assume a company has a profit margin of 10% (90% of revenues go to costs). Then one might naively expect that reducing costs so that costs now account of 80% of revenue would double profits.
Not so! Competition pushes prices down so that the new profit margin is once again 10% (or whatever the benchmark is for that industry) and any profit increases have to come from increased demand at that lower price.
Even if the elasticity is 2.0 (a very high value) the profit increase is not 100%, it's only 10%. If elasticity is equal to 1, profits are constant. If elasticity is less than 1, profits decrease.
What do industries look like where lower prices increase revenue?
It's worth looking at the cases where a hypothetical price decrease would lead to overall revenue growth to try to understand what an industry like that would look like. That is, how can you tell if you're in a situation like that.
Typically, goods with high elasticity are goods for which equivalent substitutes exist or goods that are not essential. In those cases, buyers can easily choose not to buy those products if the prices become to high. Luxury goods are a typical example of this.
On the other side of the spectrum are goods that are essential, that is, people buy as much as they need, regardless of what it costs. Typical examples would be food, water, electricity, gasoline, and so on.
The question then becomes, what category do the businesses that are most aggressively pursuing AI right now fall in to?
Demand is not infinite
Another pitfall here is assuming that if demand increases a little in response to a small price drop, demand will increase a lot in response to a large price drop. Maybe if, say, the price of smartphones significantly drops, I would buy a nicer phone but I would not buy two; I certainly would not buy 10.
While I would like the quality of some of the software I use to be better, I don't really see how my life would be better if I had twice as many phone apps.
The world can probably not support twice as many cars (where would you even park them?). We don't have the time to watch twice as many TV shows.
Twice as many sales people?
A key (and wrong) assumption a the heart of all this is that what works at small scale still works at large scale. For example, imagine you're running a sales organization. You know that one salesperson can book 20 sales calls in a week, and two of those calls result in a sale. That means you sell 2 widgets per salesperson per week. If you hire another salesperson, you get more sales because you've just increased the fraction of sales people in the market that are working for you. You just doubled your sales force while your competitors were standing still.
But what if every company in your industry simultaneously doubled their sales force as well? Sure, as an industry you might ship a few more units due to the increased customer awareness, but not twice as many.
Your buyers have finite budgets. Your buyers have finite needs. Your buyers have finite attention. A lack of marketing/sales if often a bottleneck at small scale but not at large scale.
There may even be a point where your industry as a whole gets a reputation for being spammy and people stop taking your calls altogether.
In other words, the logic that made sense for internal innovation -the kind that helps one company out compete others in the same industry, simply does not apply if the same innovation is introduced industry-wide.
Final thoughts
What really appears to be happening is that industries are actively working to undermine their own revenues under threat being run out of business entirely. They feel like they have no choice but to go along with the hype out of fear of being left out.
But this is a trap for all companies in an industry. Actual product innovation is hard and so companies focus on automating processes to cut costs. Lower costs lead to lower profits which means lower investment in innovation. An industry can bleed out.
The only path to increased prosperity for all is to develop novel products that solve valuable problems for real people. That is, the winners in the coming years will be companies that invest in building new products, not just cheaper versions of old ones.
A pure drive for cost reduction is ultimately a race to the bottom where everyone loses.
The big challenge with LLMs is that while some product innovation is possible, the scope of problems you can solve with purely digital, text generators that are sometimes wrong is pretty limited.