Is This a Moat: Mailchimp and PayPal’s Algorithmic Advantage - Commoncog

Two case studies of competitive advantage, and a question: was this a moat or not?

I’m going to do something different with this piece. Most weeks, the Commoncog essay is something that I’ve spent a great deal of time thinking through, often to the degree where I’ve worked out most of the implications. This week, however, I’m going to present two business cases, and then ask a series of questions for members to answer in the forum.

This is a companion discussion topic for the original entry at
1 Like

Please leave your answers to this puzzle here! I’ll be picking snippets from the ones I like the most and then updating this piece with them.

At a quick glance, these both seem like scale economies as the critical Power. Both Mailchimp and PayPal got a jump on competitors, possibly due to the lack of VC funding you mention, and then scaled up their capabilities around spam and fraud detection. With greater reach (due to e.g. Mailchimp offering the freemium model), they could amortize their spam/fraud prevention efforts over more users, allowing them to widen the gap on their competitors. It’s expensive to close because to build similar capabilities would require much higher per-user costs which would affect margins, so they could lower their pricing below what other competitors could match (e.g. Levchin saying the fraud rate was way below anybody else in the industry).

I suspect Mailchimp also had a dose of switching costs as a moat. Once you get your email list set up and tuned for Mailchimp, another provider would have to be much better to justify switching, and I’m guessing there’s no functionality that justifies that switch. Plus, there grew to be a whole ecosystem around Mailchimp of consultants and classes who teach people how to use that system and make it easy for them to get onboarded, which another competitor would have to replicate.


First of all, what a great read - thank you!

  1. In the two cases that you’ve just read, is the competitive dynamic that is present a type of moat? Or is it a temporary competitive advantage, one that erodes over time?

I think its a temporary competitive advantage related to one power and one meta-moat (sorry to be the obnoxious one using another meta tag).

The reason it erodes is because the environment is dynamically changing. Not only a competitive and funding environment but also computation costs, availability of GPUs, availability of knowledge of how to actually write and setup the anti-fraud / spam also translatable to: “Can we hire a person who knows his sh*t fast enough and fix the spam issue.”

  1. If this is a moat, what is the nature of the Barrier? And what is its Benefit? Also: which one of the 7 Powers is it?

I believe its a cornered resource:

  • Benefit: In Paypal’s case, the benefit compared to other companies was probably the talent it had at the time to tackle the problem quickly enough for it to not destroy the margins and not to day. Essentially it also touches a theory of constraints in Mailchimp where instead of hiring zillions of moderators at higher cost and major OPEX increase, they essentially had knowledge resource to tackle the problem in a different way by utilising on probably high risk / high reward idea of doing so via technology.

  • Barrier: As mentioned in 7 powers, the cornered resource is a bizarre thing. In Pixar’s case in the case the barrier to hold talent and team together was personal choice. In the case of Paypal and Mailchimp I believe it was even more bizzare. Perhaps it was the time and survival constraint and the urgency of do or die kind of situation that prompted them to take the risk and implement something. They didn’t have time to search for more funding to hire people to moderate.

Contrary to @eric - I don’t believe the switching cost was an issue. Even as an entity extracting 500k users out of Mailchimp is quite seamless and you could have gotten another provider setup within a day. I did it with the list of 5k and it took me around an hour.

  1. If it is not a moat, what lessons can you derive from the cases? How did this advantage ultimately affect the business? And what does it tell us about non-moat competitive advantages?

The reason I believe it is not really a moat is that we can see that eventually new entrants did gain a good market share in the market (Stripe, Mailjet) but they did it based on other conditions that were unrelated to that particular point in time. As mentioned, the market (and life) is fluid. Every 5-10 years the market shifts, changes, sculpts or introduces something new that gives an edge to faster and leaner companies to compete.
Stripe targeted a completely different group of people and infrastructure play which was still a baby market when Paypal started, but it evolved into a high-growth one.

One thing I wanted to add in relation to meta moat is based on this article by Morgan: Sustainable Sources of Competitive Advantage · Collab Fund

that is perhaps a different way how to look at a competitive advantage and the one that I picked from there is:

The ability to learn faster than your competition

1 Like

So I’ve given this business puzzle some thought, and here’s my current take:

  1. In the two cases that you’ve just read, is the competitive dynamic that is present a type of moat? Or is it a temporary competitive advantage, one that erodes over time?

The way that I phrased this question is malformed: most moats decay over time. But I’m about 70% leaning towards this is not a moat, at least not the way it’s defined in Helmer’s 7 Powers.

To sort of break this down a bit, I asked myself — what is the Benefit here, and what is the Barrier?

  1. The Benefit (as is usually the case in such analyses) is quite clear: the company in question has cost advantages. These cost andvantages enable it to a) survive, b) undercut the competition.

  2. But what of the Barrier? In both cases, Paypal and Mailchimp’s advantage is something of an ‘earned secret’ — a temporary advantage that can be copied if given enough time. I don’t think it is a barrier, because there are multiple competitors today that go toe-to-toe with them.

I’m somewhat on the fence on this last bit, at least when it comes to PayPal. I think there’s a possibility that PayPal’s advantage is that of scale economies — you’d notice that both Wise and Stripe targeted new and different use cases in order to build the transaction volume to feed its fraud detection systems. That’s what you’d expect if you were trying to go up against a scale incumbent.

But the Mailchimp advantage doesn’t seem to be as solid. Sure, you could say that it was the first to use an unorthodox approach, and that it had an extremely competent understanding of the nature of the ESP ignorant spam problem, but these are advantages that (as @kirso has pointed out) should leak out to the broader environment.

  1. If this is a moat, what is the nature of the Barrier? And what is its Benefit? Also: which one of the 7 Powers is it?

Possibly scale economies for PayPal. But not really a moat in Mailchimp’s case.

I think if you look at the cases for scale economies, you’d realise that one way to get around a scale economy advantage is to find scale in some other part of the world/some other part of the industry, and then parlay that into direct competition. E.g. GM using a ‘multiple cars for multiple product segments’ to go up against Ford’s ‘any colour as long as it is black’ approach.

  1. If it is not a moat, what lessons can you derive from the cases? How did this advantage ultimately affect the business? And what does it tell us about non-moat competitive advantages?

One of the most surprising things I’ve taken away from these two cases is the notion that the capital environment affects the impact of the advantage. In simpler language: it’s totally possible for temporary competitive advantages to wipe out the competition when the competition is highly dependent on a particular type of capital environment. The common, naive assumption is that ‘the company that raises the most venture capital wins’. But these cases illustrate a corner case: when venture capital is easily available, companies become adapted to relying on cheap and easy capital; when that capital environment dries up, even temporary competitive advantages can allow certain companies to wipe out everyone else.

I like thinking of business environments as biological ecosystems. So there’s some analogy here where say a whole bunch of dinosaurs live happy, carefree lives, feasting on flora and each other, and then an asteroid hits and the tiny mammals with minor competitive advantages (regarding not relying on some environmental surplus) turn out to be the creatures that win in the end.

The other thing that perhaps we should take away is: don’t laugh at temporary competitive advantages. Every little secret helps, even if it eventually leaks out. But then I think this is intuitive to most business operators — even when I was running things, I’d notice that we’d had a small number of competitive advantages that we tried to keep close to our chest, in order to prolong their effective lifespan.

1 Like

I’m going to take a different tack on this, as the idea that came to my mind reading both of these cases was the notion of “developmental stage theories” of cognition, specifically Michael Commons’ Model of Hierarchical Complexity. It’s an unusual fit and may just be an overreach (or seem too obvious to mention), but hear me out.

One of the insights I’ve gotten from studying the MHC is the notion that much of what we call insight comes from when we can better deal with two situations: each member of a population looks indistinguishable, or each member looks unique. In either case, all we can do is process each situation as rapidly and with as high a quality as possible based on its own merits. That leads to the inevitable tradeoffs of quantity vs quality we all know too well.

However, if we can instead create reasonable groupings of 2-7 categories that account for a significant portion of the population, we can develop and use heuristics for each category. That allows us to increase quality (by taking consistent action that is shown to yield better-than-random results) and quantity (because we can act based on category markers rather than a full investigation of each situation) at the same time.

In mailchimp’s case, what led me to this thought was the following paragraph

So our abuse desk decided long ago that we had to change the way we think about handling abuse. We began experimenting and analyzing massive amounts of data in 2008, which led to our list activity score feature. The idea here was to stop classifying customers as good or bad (and giving them access to special IP ranges for better deliverability), and start looking at their list management practices instead.

Competitors were either treating each customer as an individual, or putting them into one or two static categories that were not very helpful. Analyzing list management practices is what enabled mailchimp to create algorithms that improved both quality and quantity of work. Competitors unaware of this concept of looking at list management practices only had bad options:

  • Try to increase the number of staff to process manual entries (which became untenable once VC funds dried up)
  • Try to create algorithms based on an inadequate understanding of the population (and get blacklisted when they didn’t work well enough)
  • Limit the volume of signups to what they could effectively manage (losing out on growth opportunities, and also likely getting a higher mix of poor customers getting rejected by mailchimp’s superior algorithms)

For PayPal, the part of the case that triggered this thought was the following:

Frezza and Levchin’s efforts to apply this technique to patterns of fraudulent activity yielded another breakthrough: now, PayPal could match not just numbers to numbers but patterns to patterns. They augmented this with computer-generated rules that triggered an alert if one pattern resembled an earlier fradulent one. If such a fraud pattern registered frequently enough, the team could write a blanket rule in the system to prevent it from recurring again.

“A simple layman’s explanation is that we started fighting patterns—more than [fighting] fraudsters,” “observed engineer Santosh Janardhan.

I won’t belabor the point as the last few sentences quoted above are a pretty obvious match for what I described around mailchimp’s identification of list management practices.

Exactly which of the 7 Powers this fits is one I’m not clear on, so I’ll throw some ideas out and see if any of them stick. There is an argument for cornered resource and process power here, but there is also an argument that this is more of a synthesis of all 7 powers rather than an application of a specific one. One could say that mailchimp was able to create process power using this cornered resource to counter-position using scale economies from its superior insights that were further enhanced by network economies. This enabled it to deliver superior services that increased switching costs for its customers and create a superior brand that gave it an advantage in attracting new customers. As this way of thinking became better understood, it eroded the entire stack of power and has primarily just left them with the brand as their main power source (as best I can tell, at least).

Curious hear if this resonates with others or if it confuses more than it helps.

1 Like

I’ve been looking forward to replying on this for a while, but got swamped last week. Y’all have raised some good points, so I’m going to briefly just add a couple things I haven’t seen addressed.

One thing that muddles the question of “was this a moat” is the fact that Mailchimp actually had two distinct business units: “Mailchimp”, which was the marketing / email newsletter service most people are familiar with; and “Mandrill”, an email-as-a-service backend that Mailchimp utilized but that was also available as a separate a-la-carte service. Mandrill was high volume but comparatively low-margin as a business, but it was still a ludicrously profitable one. It’s been a long time since I looked this up, but at one point around five years ago Mandrill accounted for over half of all email sent on the internet. But “Mailchimp” never really knew what to do with this arm of the business, and repeatedly tried to shift Mandrill users into the core Mailchimp product. I don’t think any of the “moat” conversation really applied at the Mailchimp level, even though that’s the area where the freemium offering was most visible. But Mandrill had a moat that, if they’d managed it effectively, would likely have meant that even Mailchimp competitors were using Mandrill as their email backend. I think the biggest reason why Mailchimp lost their moat is that they failed to embrace this offering as a stand-alone service, so as competitors to the Mailchimp product emerged, they all ended up needing to learn this problem for themselves
so that their emerging competitors all ended up needing to develop their own email delivery backend systems.

Here are a couple anecdotes toward that theme:

  • Reputation is critical in the email business - because spam has become such a big issue, when an email is sent, there is virtually no way to guarantee that any given recipient will receive it, or even to confirm that they did receive it. Mailchimp was very careful about how poor behavior of customers might impact the reputation of the IP addresses they used to send email: for example, users on the free tier of Mailchimp used a separate pool of IP addresses from those on paid plans. They also ran their MTA infrastructure off of owned hardware in leased datacenter space, built off of custom MTA software at a time when most of their competitors were having to pay for volume-based delivery offerings from third-party vendors. All-in-all, this was a pretty high barrier to entry for new competition.
  • One other quirk: because Mailchimp embraced low-end, low-volume customers at a time when mature email services were focused on enterprise customers, the impact of losing any individual customer was minimal. (For a time, the company worked hard to streamline the process of customers migrating to other services when they “graduated” out of Mailchimp and needed services Mailchimp wasn’t prepared to offer.) This made it easier for Mailchimp to walk away from paying customers with a pattern of behavior that would decrease the reputation of Mailchimp’s email senders; in effect, they could choose to only keep the customers who they knew wouldn’t end up costing them more money. (And these customers would likely move to a Mailchimp competitor, creating a virtuous cycle for Mailchimp)

I was really looking forward to your response to this @peter, because of your history with Mailchimp.

The point about Mandrill is interesting.

What strikes me about this is that if Mailchimp’s competitors were able to build out their own Mandrill-equivalents, surely that must mean that the underlying technology was not a moat? I’ve been leaning towards calling this a temporary competitive advantage, one that could be caught up with if you spent enough energy and money to pursue it, and I think the detail you provide here just confirms that view even more.

Thank you for this, wow — this provides a little more colour to the dynamic that Mailchimp was taking advantage of.

As I see it, at any point over the past decade or so, anyone could have spun up their own email delivery service. They would have had to solve many of the same problems, and would have arrived at the same solutions, and could have created mail delivery infrastructure that operated with a fairly low cost - but still at a higher marginal cost that what Mailchimp incurred, and with little incentive to make the sorts of investments that would be needed to reduce those costs further. Mailchimp had built an email delivery infrastructure that cost very little to operate and maintain, and could have operated profitably at whatever price the market dictated, and while maintaining a higher quality product (that is - a greater reputation score)

Here are a couple more details I forgot to mention last night:

  • The mandrill infrastructure had virtually zero marginal costs - each year, as the busy season approached, the company would spin up a few dozen more servers and start warming up some new delivery IPs - but past that, the major cost drivers were data center space and network capacity.
  • I think this disconnect between “the software that generates the mail” and “the hardware that sends the mail” is a big part of this picture too. When sending email at volume, email servers generally are not just sending a single message and BCCing 2000 people on it - they’re generating and sending a unique message for each recipient using a template model. There are lots of systems available for sending mail programmatically, but IIRC circa-2017 there were only two vendors that offered systems to do this sort of bulk email templating, both with pricing tied to email delivery volume. Mailchimp started off with one of these packages, but after a multi-year skunkworks project, ended up writing their own in-house replacement.
  • I don’t know if this has been widely discussed publicly, but Mailchimp was able to negotiate peering relationships with some of the large cloud computing providers. IIRC, this improved reputation scoring with those recipients while also reducing network utilization costs.
  • I think the main challenging question in this is “if this was a good moat, then what happened to it?” I imagine the problem of medieval walled cities, where the wall that protects the city also inhibits its growth. Mandrill was always extremely profitable, but it was also going to be a business that peaked at eight or nine figures of annual revenue. The Mailchimp founders had not set out to create an email company, and chafed at being described that way - they were attracted to design, and when they had first started the business they had been doing full-suite web services including website design. I don’t think they ever were quite comfortable walking away from that vision of what they wanted to build. At some point, after Mandrill had spun up and become highly successful, they pulled it back in under the main company umbrella, stopped active development work on the infrastructure (for many years), and started trying to transition Mandrill customers into being core Mailchimp customers.

The reason this moat ended up only being a temporary competitive advantage wasn’t that it wasn’t a moat, but because it was a moat surrounding a castle that they didn’t want to live in. There may be an analogy here with PayPal too - where they seemed to really lose some of their edge after being bought by eBay…


I meant to reply to this too. This is an imprecisely remembered anecdote, but I recall someone telling me once that if Mailchimp lost its hundred largest customers, it would have less than a 1% impact on revenue. It was in some ways a highly democratic business, where it was far more important to maintain the aggregate health of the overall system than to please any individual customer. Contrast this with other business of similar size where there’s a persistent incentive to compromise the product architecture in order to accommodate the requests of a very large customer.


I really enjoyed the post. Thank you.

I jumped to Counter-Positioning before reading the earlier comments. Both Mailchimp and Paypal employed heterdox solutions to issues faced by competitors … and in ways that caused competitors to suffer.

Benefit: attract more customers at lower costs
Barrier: and here is why I think that this is not a lasting Power … competitors could deploy the unique solutions without harming themselves

1 Like