As the generative AI craze rages on, Ramp acquires customer support startup Cohere.io

Want more fintech news in your inbox? Sign up here.

Finance automation company Ramp has acquired Cohere.io, a startup that built an AI-powered customer support tool, the companies told TechCrunch exclusively.

Founded in 2020, New York-based Cohere.io (not to be confused with Cohere, another AI startup that recently raised capital) raised $3.1 million in a seed funding round led by Initialized Capital, later tacking on another $400,000 in funding. Other backers include Y Combinator, BoxGroup, Soma Capital, Shrug Capital and Chapter One. Angel investors include Plaid CEO and co-founder Zach Perret, Elad Gil, Naval Ravikant, Opendoor’s Eric Wu, Prasanna Sankaranarayanan, Ramp co-founders Eric Glyman and Karim Atiyeh, and Jack Altman, among others. 

The deal marks Ramp’s first acquisition since it bought Buyer, a “negotiation-as-a-service” platform that claimed to save its clients money on big-ticket purchases such as annual software contracts. in August of 2021 and second since its 2019 inception. Financial terms were not disclosed.

As reported previously by TechCrunch, Cohere’s three co-founders, Yunyu Lin, Jason Wang and Rahul Sengottuvelu (CTO), first met while attending Duke University. Interestingly, Lin later left to work for Ramp, but the others graduated during the pandemic.

After one year, Lin left Ramp to help start Cohere.io. And now Lin, Sengottuvelu and three other engineers have joined Ramp to drive AI-related projects with the goal of solving problems for Ramp customers “in entirely new and radically more efficient ways.” One salesperson is joining the Ramp team as well (Wang left Cohere in 2021).

Cohere.io told TechCrunch at the time of its raise in 2021 that its goal was to improve on the remote desktop and screen-sharing experience. With Cohere’s technology, it claimed, businesses could help customers “in seconds” by taking instant control of their screen without any downloads or setup on the customer’s end.

That ease-of-use gained the startup an early over 50 paying customers for its product, including TechCrunch Disrupt 2020 winner Canix, CopyAI, Ramp and others. Since then, though, Cohero.io switched focus from screen sharing to LLM (large language mode)-powered support automation. It says it has been using generative AI and LLMs “since before the advent of ChatGPT” to extract companies’ historical customer support data and apply it to similar questions in the future.

The company’s automation product was used by teams like Ramp and Deel to generate “high-quality support knowledge” from existing customer interactions, and automatically resolve up to 60% of tickets, the company said. As part of the acquisition, Ramp said all customers will continue to be served “in maintenance mode,” but it will focus its efforts on adapting the LLM-powered automation product to help Ramp customers automate their workflows and better use and understand their spend data.

While Cohere’s founders declined to reveal hard revenue figures, they said the company had seen 150% year-over-year revenue growth and had grown to over 200 customers (also including Rippling, Loom and SecureFrame) at the time of the acquisition at the end of last month. Ramp itself has over 15,000 customers.

Ramp CEO Glyman told TechCrunch that Cohere.io’s approach stood out from other chatbots in the market that require employees to manually define and answer different variations of customer questions to ensure accurate responses.

Said Yunyu Lin: “There are a lot of vendors doing some automation out there. And historically, a whole bunch of them have been targeted towards simpler use cases, like for instance, e-commerce, or simple consumer SaaS that the total number of questions that people can ask for answers to is not that many.”

Cohere’s differentiator, Sengottuvelu said, is its ability to generate content for “even very infrequent questions” and for “complex enterprises with complex products such as Ramp.”

Specifically, he added, Cohere’s automation product uses existing help center and ticket data to create workflows and pinpoint areas for improvement in self-service systems. Cohere also provides “comprehensive” performance metrics, giving businesses insights into the effectiveness of their self-service in different product categories.

Initial projects include making workflows “even more automated,” Sengottuvelu said, by completing multi-step processes on behalf of customers, as well as improving Ramp’s ability to analyze large sets of both unstructured and structured data “so customers always get the best price on software.” For example, Ramp uses Sengottuvelu’s Jsonformer project – which lets developers set hard limits on what LLMs can output. Since Ramp deals with massive sets of unstructured data, Jsonformer helps the company put that data into a structured format “that’s easier to analyze and use,” Ramp said. In fact, the company’s new price intelligence feature uses the technology to extract details from contracts, invoices, and receipts, in an effort to gets its customers a more fair price from their vendors.

Notably, Ramp wasn’t Cohere.io’s only suitor, according to the founders.

“We got other offers from some players in the space who saw a lot of value from generative AI but we believe that a lot of the value of generative AI will belong to companies that have access to this customer data and Ramp is out of these companies, basically. And it’s one of the fastest growing companies in tech, especially in this macro environment,” Lin said.

For Ramp’s Glyman, it was evident from the early on that once his company started using Cohere.io, “suddenly the majority of tickets were being answered properly in an automated fashion.”

“It actually really worked,” he said. “The technical sophistication of the team was far beyond anything we had ever seen.”

Got a news tip or inside information about a topic we covered? We’d love to hear from you. You can reach me at maryann@techcrunch.com. Or you can drop us a note at tips@techcrunch.com. Happy to respect anonymity requests.