Freshworks' Efficient Frontiers
Freshworks is a strange sort of company: it planned to be a small business-focused software tool with a single product, a helpdesk tool, but has since evolved into a more comprehensive suite with two separate products at the $100m+ annual recurring revenue mark, and has also moved upmarket. The company has 52,500 customers, with average annual revenue of $5,900 per customer. But it's $50k-and-up customers are 37% of recurring revenue. Freshworks is, increasingly, a software bundle for big companies—but still one that’s mostly bought or sampled by small ones.
The company’s range of products should be familiar to anyone who has looked at Salesforce. Freshworks automates helpdesks and marketing, handles HR, manages internal and external IT requests, and otherwise acts as a switchboard for companies to convert questions and problems into items on someone’s to-do list. The inspiration for the company was either that the founder shipped a TV from the US to India and had trouble getting a refund when it broke (if you listen to his more recent interviews), or a Hacker News comment about how much everyone hated existing helpdesk software providers (the original story).1
Their S-1 has some other interesting nuggets. They've managed net dollar retention in the 110%-120% range over the last few years, which is impressive given the structurally high churn of smaller buyers. Part of Freshworks' strategy is something the CEO calls "Indian Democratic Design"—the manifesto makes it vague, but the basic idea is to build a product simple enough that a small business can adopt it, flexible enough that the product can grow with that business, and cheap enough that emerging market customers aren't excluded from buying it.
Freshworks is an interesting example of India's service export sector. The company is headquartered in San Francisco and plans to list on the Nasdaq, but it was founded in Chennai and 88% of its employees are in India. (In an example of the fractal nature of all tech company conventional wisdom, the CEO was advised that Bangalore was a better spot than Chennai, because the local network was denser and there was more venture funding available.) A low cost of living, a large English-speaking population, and good technical schools has made India a good place for software outsourcing, but this selects against building products locally rather than converting specs into software that's then owned by some third party. But a common development pattern is for countries to move up the value chain as they grow; a place that supported cheap outsourcing a decade or two ago will have the human capital and financial infrastructure to support product-based companies now.
The classic view of enterprise software is that there are several kinks in the demand curve: up to a certain price point, the product will be sold to buyers who are putting it on a corporate card and expensing it, so a light-touch sales process works. That's where Freshworks' roots are, and it's still something they highlight; on a podcast last year, their CEO was still proud of this post from ten years ago because it keeps driving traffic. He converted a negative tweet from a larger competitor into this single-serving site, another nice marketing tactic—the site is designed to be shared, at least by people who have strong opinions on whether or not the name "Freshdesk" is a ripoff of "Zendesk," and it closes with—a signup form for a demo and some customer testimonials.2
Most of Freshworks’ customers are small businesses, although the 23% of them doing over $5,000 in annual recurring revenue are responsible for 84% of total ARR (and growing). With those smaller customers, it's possible to close the sale entirely online, and marketing tactics like this are a good way to push the cost of customer acquisition down to zero. But Freshworks noticed over time that they weren't closing the biggest deals (one deal fell through because the UK-based client wanted an in-person meeting and no one at the India-based Freshworks had a visa), so, starting in 2014 they began opening local offices in major markets at a rapid clip: the US in July of 2014, Australia in April 2015, a Europe office the next month. This coincided with a sizable venture round; it's an interesting example of venture money both ratifying that an idea works and funding its refinement into a better version.
The higher-ticket the software, the more a company optimizes for sales and satisfices on product. But for a certain category of customer, this is particularly troubling: companies that are growing fast are pressed for time, and if the product takes a long time to become useful, it’s much more expensive than it otherwise would be. If a customer is growing fast enough, and it takes a few months to implement a newly-purchased product and get users used to it, the company is materially bigger by the time it actually gets the software it needed at its previous scale—and that much closer to no longer needing the product.3 Time from purchase decision to implementation is a useful metric, but over time enterprise software companies will probably shift to measuring how long it takes to go from purchase decision to roughly one daily active user per workday per seat, i.e the expected time at which churn driven by something other than budget cuts or bankruptcy rounds down to roughly zero.
The core problem Freshworks evolved to solve goes like this: as companies grow, they end up with multiple channels to acquire and communicate with customers. Which sounds great: they’re running a balanced portfolio of different marketing channels, and can easily tilt their resources towards whichever channel is performing best. But a problem arises when a customer who arrives through channel A, but makes purchases through channel B, complains through channel C: figuring out who this person is, what they bought, and how their complaint ought to be resolved is nontrivial. And it’s one of those nonlinear scaling problems that gets relatively bigger as a company grows.
Nonlinear expense scaling, in its various guises, is what ultimately turns a growth company into a company that’s reached its steady state, where it can replace the customers it loses but can’t profitably add more. This doesn’t just have a direct effect on growth: it has a corrosive effect on employee morale. If the biggest problems at a company involve solving problems for their customers, that’s exciting; if the biggest problem is dealing with self-inflicted mistakes, it’s painful.
This leads to a dynamic where a company that grows to its manageable limit will only stabilize temporarily. Over time, it will decline, as the best employees leave and as everyone else works a little less hard. So Freshworks isn’t just selling software; it’s selling the avoidance of a particularly drawn-out and annoying failure mode.
A few years ago, Facebook turned out to be overcounting video views ($, WSJ), an incident that has been blamed for the great media pivot to video and subsequent washout. It was never all that clear how the bug could have been that damaging, since video views are an input into conversions, which are what advertisers actually care about. To the extent that it mattered for videos monetized with branded ads rather than direct-response, Facebook was accidentally subsidizing video, then discovered the bug and stopped. But conspiracies remain: what if Facebook was deliberately fudging metrics? One good answer to this is that Facebook has discovered another bug, and this one undercounted conversions for app ads on new iPhones. That is both a more serious mistake and one that directly took revenue out of Facebook's pockets; gaming advertisers are the reserve bidders for Facebook inventory, and they care very deeply about user acquisition costs, so when Facebook missed high-spending customers, it directly lowered their bids and hurt its entire model. The size of the impact is hard to measure, but it's good evidence that genuine bugs are a better explanation than conspiracies.
A Money Market Modest Proposal
Aleph Blog has a suggestion for keeping money market funds from “breaking the buck” by going below $1: when their assets are worth less than a dollar, penalize people for redeeming them until the fund's asset value gets back up to $1. Money market funds have experienced bank runs before (which was a central claim in this piece about stablecoins), but preventing bank runs is difficult. The main cause of a bank run is people worrying about a bank run, specifically worrying that they won't get their money out on time. Anything that penalizes withdrawals once a run starts will force some holders to just hold on until it's over, while others will redeem earlier than they otherwise would have in order to avoid that outcome. Liquidity crises are partly a process of realizing that something that looked like money was not quite money after all, which makes stopping them hard: any policy that prevents liquidation causes exactly the problem that it's solving.
Restructuring China Tech
A few recent headlines show the general contours of the CCP's new approach to tech companies: they aren't being shut down, but they are being rearranged. Ant, for example, is selling a stake in its credit-scoring system to state-backed companies, which is a way to redistribute its data. And Didi and JD are unionizing, which will probably change the distribution of profits towards workers, and slow their growth. One way to look at this is that the old grand bargain between tech companies and the Chinese state was that they'd be left alone as long as they made the country richer and didn't try to get involved in politics. But the CCP, like investors everywhere, underestimated just how profitable a full-scale tech company can be, and turned out to give away more than it needed to. China's big tech companies are useful, and the state is well aware of that. Now they're in the process of becoming more useful to the CCP specifically.
The WSJ has an interesting piece on CVS, Home Depot, and other retailers' efforts to stop shoplifting ($), which has gotten more professionalized over time. One reason is that consumer goods are more liquid now that Amazon exists: if you can get lots of power tools for free, Amazon is a good place to sell them. The lifecycle of Internet-driven scams is sharp but often brief: there's a point at which a platform's scale is enough to make an existing illegal activity far more lucrative, followed by a response from the platform itself, which collects enough data to quickly identify people who are selling stolen property.
(Disclosure: Long AMZN, though fencing stolen goods from other large-cap companies is, of course, not core to the thesis there.)
Limits to Copying
LinkedIn is killing its stories product. Ephemeral content is useful in many markets, but in business communications it has a fairly bad reputation: it implies a deliberate effort to avoid future discovery. Of course, LinkedIn users were probably not sending one another short clips of material nonpublic information or suggestions to engage in price-fixing, but it was still an interesting effort. For a while, the surprising thing about the Stories format was how many products it worked for. Launching a stories-like feature turns out to be a good proxy for which apps are at least trying to keep up, even if they don't necessarily win.
This is a fairly benign version of the usual pattern where a company pivots its narrative as it gets more successful. ↩
I wrote here about what a positive signal it is when CEOs insist on doing individual contributor-level work past the point at which it makes sense. I don't know if their CEO is still writing blog posts designed to rank for long-tail search terms, but to anyone on the Freshworks marketing team, it's probably some combination of inspiring and terrifying to know that the CEO is perfectly capable of doing the job, and, in turn, plenty good at judging quality. ↩
This is one explanation for the stylized fact that a meaningful percentage of expensive enterprise software products never get used by their customer. Selling something useless is bad for user retention, but it still happens; if the company’s needs change faster than the implementation cycle of the product they buy to suit those needs, then implementation becomes a lower priority over time, and eventually drops off the to-do list entirely. ↩