Looking at averages can lead you astray. Yes, averages can be useful to draw certain conclusions. But very often you need to look within the subsets of an overall population to really understand what’s going on.
I’m a trained molecular biology and have a tendency to think in terms of the scientific method. It’s been a very long time since I worked in the lab, but I’m still interested in following development in the life sciences field. In particular, I’m fascinated by advances in the genomics. So I’m going to give you some examples why averages are often not useful in drawing conclusions when the individuals within a population are not homogenous. One of the things that people study is gene expression defined as the degree to which individual genes are turned on (i.e. are activated) or not. With some rare exceptions, each cell within our body contains the same DNA sequence as every other cell. Differences in function, responses to external or internal stimuli, and structure are largely due to differences in the specific gene expression patterns in individual cells or groups of cells. Some genes are turn on, some are not; furthermore, this expression changes temporally and in degree. So if the genome is like a piano with a fixed set of keys then each cell has its own piano and has the capability to play its music by controlling the playing of specific keys in its own manner. The music you hear is dependent on which keys are played, how often, how hard, and in which combination with other keys.
There are many examples of why a scientist would study gene expression to draw conclusions about a molecular pathway, or some form of pathology, or some immune response, or some cancerous disease. In many such studies the scientist will extract some living cells from a subject. This could be a biopsy or a blood draw (a liquid biopsy). Now, in the past the scientist might extract RNA (the transcribed form of DNA that is present in the cells due to the DNA being turned on) of the entire sample (or even subset of the sample) that can contain millions of cells. This RNA, which is the totality of the RNA in all of the millions of cells, is homogenized and the RNA is extracted. You now have a bulk sample comprised of the RNA from millions of cells----this is an AVERAGE of the gene expression of millions of cells. This is analogous to combining the individual music from millions of pianos. The notes are all combined and unless all of the pianos are playing the same song then it is impossible to deconvolute which notes are being played by which piano. About a decade ago, single cell analysis and single cell genomics started taking off. Single cell genomics enables each cell’s gene expression to be measured in isolation so rather than looking at the average of Gene A, Gene B, and Gene C’s expression (from all the cells in a bulk sample) you are looking the individual gene expression measurements from each of hundreds, thousands , or even millions of cells. What is going on biologically is unmasked by looking at all the individuals within a heterogeneous population of cells rather than the average of millions.
Why am I writing this? It’s because as investors we can also be fooled by averages. Averages can be very misleading. People say all the time that the market is over- or undervalued. So what? Who cares when you are analyzing individual companies and investing in those firms that you have decide will do better. If you don’t believe that it is possible to pick stocks that can outperform then you should not be on this board as that’s what we do here.
Now, let’s drill down to an individual company. Again, we can be misled by using an average to draw our conclusions. Here’s a post by me from 2 years ago:
Back then, UBNT growth had slowed and I argued that looking at overall growth would be misleading for trying to determine the company’s future prospects. The enterprise business was small but growing very rapidly while the larger service provider business was slowing. If you just looked at overall growth then you would have missed it.
I see something similar with NVDA today. They have 2 small but very rapidly expanding business lines: Datacenter and Automotive.
Now let’s get to the reason what prompted me to write this post. Yes, Twilio. People have said that revenue growth has slowed. First off, this was a reference to guidance so it hasn’t actually slowed yet. Now, I say that someone who says that revenue growth has slowed is being fooled by an average. So what’s the specific problem in this specific case? There are several.
- The figure used was y/y growth of total revenue. I would just use the recurring revenue and not total revenue. They call this Base Revenue and it is more indicative of what can be expected in future periods.
- An isolated population within the revenue number is declining while the rest of business is growing. This isolated declining business is the Uber business. Management has said that they are not seeing such declines in the other customer accounts. So if you include this declining revenue portion in projecting future growth, you will get a number that’s underestimating the longer term growth opportunity. This is particularly true when one considers that decline in Uber business will be complete in the next several quarters.
- Even within the Uber business, the revenue is not homogenous. You have lower value (to Uber) TWLO costs and you have higher value (to Uber) TWLO costs. From what I can gather is that Uber has been able to bring some of the easier communications functions in-house. It also appears that Uber has extracted some additional discounts from TWLO for other lower value communications. It was also reported that TWLO and Uber continue to work on other use cases. I would assume that this ongoing work is for higher value or more complex communications. So in 2-3 quarters the decline will be complete and you will be left with Uber once again growing its contribution to TWLO’s revenue.
So what do I think this all means? We will likely see a big overall decline in TWLO’s Uber revenue in Q3. We will continue to see decline in Q4. But remember that within this decline there is growth in higher value revenue and there is growth in overall usage as Uber’s communication messages are increasing as Uber grows its network of drivers. So we have TWLO’s largest customer temporary cutting back on its TWLO usage but once the cutting back is done then Uber’s overall business will grow again. In the end Uber will be a large customer but smaller in relation to the overall business. So I see it as a temporary drag on revenue growth because even the Uber business to TWLO will resume growth within a year.
If you think that Uber is a canary in the coalmine (other TWLO will do the same) then show me the evidence. You would see gross margin pressure. You would see decline Net Dollar-based Expansion. You would see TWLO management confirm that when asked on the conference calls; they were asked after Q1 and after Q2 and they denied that a) other customers are getting any non-standard price concessions, and b) other companies are moving business in-house. In the future if someone makes a claim then they should really back it up with evidence. And don’t be fooled by the average!