As the AI industry continues to release a plethora of new models, companies hoping to stay competitive are racing to adopt them for their own purposes. In fact, according to technology consulting firm Searce, roughly 10% of companies plan to spend a whopping $25 million on AI initiatives this year.
But while a lot of money is being spent on AI, the ROI is unclear: According to Gartner, half of AI leaders don't know how to calculate or demonstrate the value of their AI projects.
Former Airbnb data scientist Chetan Sharma argues that calculating AI ROI isn't that difficult with the right tools. Sharma is co-founder of Eppo, an experimentation platform that helps customers evaluate and customize AI models for their specific use cases. In addition to a model evaluation suite, Eppo also offers a general A/B testing platform and services for apps and websites.
“With new AI models being released every week and companies pouring millions of dollars into them, A/B testing is a cost-effective way to evaluate their effectiveness without overspending,” Sharma told TechCrunch. “Eppo helps companies identify which models truly deliver value, enabling smarter, more sustainable decision-making in an environment of rapid innovation and rising costs.”
Eppo competes with a number of experimentation and A/B testing startups on the market, including Split, Statsig, and Optimizely. Large tech companies such as AWS, Microsoft Azure, and Google Cloud also offer a growing number of model fine-tuning and evaluation tools.
But Sharma says what sets Eppo apart are features such as its “contextual bandit” system, which automatically discovers new variations of a client's website, app, or AI model and then actively studies their performance by subjecting them to increased load and traffic.
Eppo's backend dashboard. Image courtesy of Eppo
“Experimentation increases speed and accelerates growth by eliminating bureaucratic (and often erroneous) decisions by committee, while firmly tying initiatives to growth metrics, quickly weeding out bad ideas and justifying good ones for reinvestment,” Sharma said. “Eppo's approach to live 'online evaluation' testing of AI models provides the answer to whether a premium model improves metrics.”
Sharma said Eppo, which emerged from stealth in 2022, now has “hundreds” of enterprise clients, including Twitch, SurveyMonkey, DraftKings, Coinbase, Descript, and Perplexity. Alexis Weil, head of data at Perplexity, told TechCrunch that Eppo has enabled Perplexity to “significantly scale” the number of experiments it runs simultaneously.
Investors seem to be pleased: This week, Eppo closed a $28 million Series B round led by Innovation Endeavors, with participation from Icon Ventures, Amplify Partners and Menlo Ventures. Sharma said the new funding values Eppo at $138 million post-money, bringing its total funding to $47.5 million, and will be used to strengthen Eppo's marketing and AI experimentation capabilities, enhance its analytics services, and expand its go-to-market activities.
San Francisco-based Eppo currently has 45 employees and expects to have 65 by the end of the year.
“The demands of lean growth combined with the rise of AI are forcing companies to become experimental, an adapt-or-die mentality,” says Sharma. “And traditional vendor gaps have led most of the experimental market to assemble large in-house teams and choose to build rather than buy. With frequent employee turnover and layoffs, these in-house teams are no longer sustainable, and companies are turning to Eppo to replace expensive or isolated in-house builds.”