ChatGPT scrambled the Chegg

An interesting thought experiment of mine has been trying to figure out which businesses will eventually be killed by generative AI. So far, education tech company Chegg appears to be the biggest loser, with its market cap collapsing from $14 billion in February 2021 to just $191 million in November 2024, including a 49% single-day drop in May 2023. Over the weekend, The Wall Street Journal published an interesting (almost) epitaph on Chegg:

“Since ChatGPT’s launch, Chegg has lost more than half a million subscribers who pay up to $19.95 a month for prewritten answers to textbook questions and on-demand help from experts. Its stock is down 99% from early 2021, erasing some $14.5 billion of market value. Bond traders have doubts the company will continue bringing in enough cash to pay its debts…

A survey of college students by investment bank Needham found 30% intended to use Chegg this semester, down from 38% in the spring, and 62% planned to use ChatGPT, up from 43%.”

As someone who was an undergraduate student from 2015 to 2019, and an MBA candidate from 2022 through 2024, I’m in the unique position to have used Chegg to “help” with my undergraduate finance classes and ChatGPT to help with my graduate-school finance classes. In hindsight, Chegg’s death by GPT was one of the more predictable outcomes in public markets.

While Chegg’s management may disagree, Chegg’s primary utility has been helping college kids cheat on their assignments. In January 2021, when Chegg sported a $12 billion market capitalization, Forbes published an excellent feature story on how the company’s growth exploded during the pandemic as colleges turned to remote classes. The big takeaway: kids were using it to cheat. On everything. Forbes interviewed 52 students for the piece, and 42 of them straight-up admitted to using the site for cheating. The thing is, Chegg’s ability to be used as a cheating tool was dependent on the answers in Chegg’s database. While Chegg launched (or acquired companies that provided) a variety of services, its cash cow was Chegg Study, which had a database of 46 million textbook and exam answers, and most of those answers were supplied by freelancers from India. From Forbes:

“Chegg is based in Santa Clara, California, but the heart of its operation is in India, where it employs more than 70,000 experts with advanced math, science, technology and engineering degrees. The experts, who work freelance, are online 24/7, supplying step-by-step answers to questions posted by subscribers (sometimes answered in less than 15 minutes). Chegg offers other services students find useful, including tools to create bibliographies, solve math problems and improve writing. But the main revenue driver, and the reason students subscribe, is Chegg Study. 

‘If I don’t want to learn the material,’ says a University of Florida sophomore majoring in finance, ‘I use Chegg to get the answers.’

‘I use Chegg to blatantly cheat,’ says a senior at the University of Portland.”

I mean, we shouldn’t be shocked by this. Students (not me, of course) were using Chegg before the pandemic was a thing. Remote learning removed any remaining friction from just looking up your answers online. However, while Chegg’s database was useful for finding solutions to questions that had previously been answered (or that closely resembled questions that had previously been answered), it was less effective for answering novel questions, because Chegg itself couldn’t solve anything. It outsourced that to India.

ChatGPT, on the other hand, does solve things. Instead of hoping that your question previously appeared on Chegg, you could just upload a screenshot to ChatGPT and let it cook. In April 2023, realizing that generative AI posed an existential risk, Chegg announced a partnership with OpenAI to build GPT-4-powered “CheggMate,” an “AI conversational learning companion.” Chegg then pivoted in August 2023 to a partnership with Scale AI, a platform used by companies like OpenAI and Nvidia to help train and build machine-learning algorithms, to build its own proprietary large language models. 

The issue here was that Chegg’s models were never going to compete with OpenAI’s. OpenAI had a multi-year head start on training large language models, and GPT-4 was rumored to be trained on 1.8 trillion parameters. Was Chegg really going to build a more powerful model from its data set of 46 million answers?

The takeaway here, I think, is that if your business model is predicated on cheap overseas labor quickly answering customer queries, there’s a good chance that a generative-AI model can accomplish that goal cheaper and faster.