The Right Dynamics: Why Private Equity and Venture Capitalists Invest in the Language Industry

The SlatorCon San Francisco 2019 Investor Panel brought together three recent investors into the language industry, representing both venture capital and private equity. Despite having different investment strategies, they agree on one thing: the language industry is highly attractive to investors.

Tomasz Tunguz is Partner and Managing Director at VC firm Redpoint Ventures, which manages some USD 4bn across multiple funds and which has backed such companies as Netflix, Twilio, and Zendesk. A leading authority on SaaS, Tunguz joined the company from Google, where he ran AdSense social-media products and internationalization. At Redpoint, Tunguz led Lilt’s 2016 seed round and participated in its 2018 Series A.

Sri Chandrasekar is Partner at Point72 Ventures, an early-stage VC backed by hedge fund legend Steven Cohen. He leads investments in AI and machine learning for Point72. Chandrasekar joined the company from In-Q-Tel, the strategic investment arm of the CIA, where he led an AI lab.

Among Chandrasekar’s recent investments are PolyAI, an NLP company that builds tech to support contact center agents, and Unbabel; the latter’s USD 60m Series C led by Chandrasekar was announced a week or so after he spoke on the SlatorCon SF Investor Panel.

Charles Stubbs is Partner at MSouth Equity Partners, a PE firm based in Atlanta, Georgia, which manages over USD 2.2bn in assets across four funds. He sits on the board of Big Language Solutions, in which MSouth recently invested, as well as a number of other portfolio companies. His firm’s investment sweet spot: companies with an enterprise value of between USD 25m and USD 200m. Typically, MSouth makes investments of USD 25–75m.

Feeling the Love From Investors

Although not many VCs, as yet, know about it, according to Point72’s Chandrasekar, the language industry is a “very exciting industry” from the point of view of an investor with some of the “biggest enterprise customers in the world.”

“They spend not insignificant amounts of money on making their content available to a global audience. So it’s a really exciting industry that has the best kinds of customers; the stickiest kinds of customers, who spend a lot of money. And those are all things that we as venture investors love to look at,” Chandrasekar said.

The language industry presents investors with the right kind of dynamics, according to Chandrasekar: “big industry, lots of labor, lots of awesome, big customers who have a high willingness to pay, and a technology chain that should enable new kinds of business models and an expansion of opportunities that players can chase after.”

MSouth’s Charles Stubbs, too, cited the huge demand for services offered by “an industry that’s growing 7% year over year, double or triple GDP — that always interests us — and low capex investment. We traditionally love business services.”

Another thing Stubbs loves (as do other investors, VC or PE), is that the language industry is still highly fragmented. According to Stubbs, “We love fragmented industries [because there is] lots of opportunity for buy and build.”

While PE firms, such as Stubbs’ MSouth, are executing a roll-up by buying incumbent players, VCs, represented on the panel by Tunguz and Chandrasekar, are financing seemingly riskier bets on companies that are attempting to use the latest advances in AI to outcompete established LSPs and open new markets.

The strategy behind MSouth’s backing of Brink’s Big Language Solutions is simple, Stubbs said: “We invest in people, fundamentally, and we had the great fortune of being introduced to Jeff Brink, who has 30 years of experience in this industry and is incredibly well-networked. Brink had previously told Slator that he and MSouth will “assemble a portfolio of high-performing and complementary language service providers” — beginning with their first acquisition, Miami-based ProTranslating, which has around USD 25m in annual revenues.

For his part, Chandrasekar said he has seen a large, interesting market emerge “very, very rapidly around rapid translations for customer service,” where there is “a higher sort of switching impetus” compared to segments with a high quality bar — ergo, with stickier customers who will not switch LSPs so easily, “unless the vendor screwed up,” Chandrasekar said. (At the time of the panel, Unbabel’s USD 60m Series C had not yet been made public. The tech-enabled LSP has been laser-focused on building services based on neural machine translation to enable multilingual customer service.)

According to Chandrasekar, “In customer service, they’re always upset with their customer service [provider], so they’re always willing to try something new. I think that allows people to break in and bring tools to the industry.”

As for Redpoint’s strategy behind investing in Lilt, it has to do with what Tunguz predicts will be the rise of the so-called “AI Agency.” Silicon Valley-based Lilt pioneered ITP or Interactive Translation Prediction and launched in 2015 as an ITP-powered translation productivity tool. Lilt has since transitioned toward more managed services.

Rise of the AI Agency

“We’re running an investment thesis right now called the AI agency,” said Redpoint’s Tunguz. The thesis is based on the fact that, according to Tunguz, there are “large swathes of the US economy dominated by agencies” (e.g., accounting firms, law firms, debt collection and, yes, language service providers).

“Today, most of that revenue is actually labor. It’s basically marked-up labor and a meaningful fraction across all that work is rote. It’s repeatable. It can be automated through machine learning,” Tunguz said.

“On the other end of the spectrum, our perspective is that translation is actually quite a bit more rote than most other categories” — Tomasz Tunguz, Redpoint Ventures

He added, however, that the degree of automation possible varies depending on the agency category. “I would argue graphic design agencies are very difficult to automate because there’s a lot of creativity. On the other end of the spectrum, our perspective is that translation is actually quite a bit more rote than most other categories.”

So how does one invest based on that thesis? By looking at next-gen, machine-learning companies, which will begin automating rote work and then taking their services to market by masquerading as an agency, Tunguz said. Case in point, Lilt; another, Unbabel.

Florian Faes, Charles Stubbs, Sri Chandrasekar, Tom Tunguz (left to right)

“They’re going to look like an agency, but they’re going to have a different engine; a fundamentally different engine that is far more efficient. They’re going to go and sell exactly the way an agency would sell, but under the hood they’re going to be much more efficient — which means they’ll have higher gross margins,” Tunguz said.

Tunguz predicted these AI agencies would go down one of two paths: “They can either run the business at a higher net income margin or they can underprice the market.”

Unbabel’s backer, Chandrasekar concurred. He said, “I absolutely think that, over the course of the next five years, there’ll be a tremendous amount of pricing pressure in the meat of the language translation market — which is the really high quality translations for healthcare, financial services, and legal, etcetera.”

However, it will be more than made up for by technology making possible a wider array of use cases than ever before, Chandrasekar said, adding, “As a result, the best companies are not going to shrink. They’re actually going to grow, because they suddenly are able to deliver far more to an enterprise customer than they were delivering before.”

NMT v.a.v. Humans

Given the looming prospect of technology disruption, where does that leave more traditional LSPs and, for that matter, more traditional PE investment targets?

“There are certain parts of this industry that we thought had a sufficient moat around them over our investment time horizon. It’s really around financial services, legal, and particularly litigation, patent services and life sciences, where there can be no errors” — Charles Stubbs, MSouth Equity Partners

According to MSouth’s Stubbs, they believe there is a “sufficient moat” around the businesses they build over their typical investment time horizon. He pointed out that their target businesses will still require “an incredible amount of project management and handholding of the customer […] even as software accelerates.”

He added, “The next owner of the business we build will [still] do quite well. It’s really around financial services, legal, and particularly litigation, patent services and life sciences, where there can be no errors. [There is] a lot of complexity around those things.”

In Chandrasekar’s view, much of translated content will remain human-powered. “At the top of the industry is content that has to be human-translated; because, let’s be honest, no one’s ready to throw their content into Google Translate and then make that available to their customers,” he said.

Technology enables a whole new class of language translation, expanding the size of the market — Sri Chandrasekar, Point72 Ventures

What technology does, according to Chandrasekar, is to enable “a whole new class of language translation, expanding the size of the market. Perhaps NMT can enable faster turnarounds for a set of things that big enterprise customers want translated, [but] were unable to go to a language service provider for. Because of technology, [it can] now be translated in two minutes instead of two days, or two hours instead of two weeks.”

He continued, “Is there an opportunity for lower quality translations that NMT can handle 99% of already? So you can imagine all of us have been thinking about that: Where are the opportunities for technology to impact the translation industry, specifically deep-learning-based translation technology?”

On the contrary, Tunguz said, “I think NMT will actually price at a premium to humans. So there will be a deflationary force — lots of content where you can have some marginal error and that’s going to commoditize, and it’s going to go from 24 cents a word to 15 cents a word, maybe 10 cents a word. But, I think there’s going to be a category of translation where it absolutely has to be right and it absolutely has to be consistent, and machines are way better at doing repetitive work than humans are. If the style has got to be consistent across a whole bunch of different languages, then maybe, in 10 years, you see NMT systems actually pricing a premium, two to three cents a word above.”

As the SlatorCon SF 2019 Investor Panel drew to a close, Tunguz said, “I think the other massive trend in the industry is you’re going to have massive consolidation. Technology is going to create economies of scale. So it’s going to make sense to be as big as you possibly can; because, once you have the existing customer, you can serve more and more of that customer base in an increasingly efficient way.”