Birth of a Salesman: OpenAI Sheds Its Lab Coat to Seek Big Deals
THE TAKEAWAY
• OpenAI is building a sales team and aggressively courting customers to use its AI
• OpenAI is leasing servers to customers like Salesforce to run its models at a cost of up to $156,000 per month
• Khan Academy is paying OpenAI over $20,000 per month to test a tool with 1,000 users
• OpenAI is building a sales team and aggressively courting customers to use its AI
• OpenAI is leasing servers to customers like Salesforce to run its models at a cost of up to $156,000 per month
• Khan Academy is paying OpenAI over $20,000 per month to test a tool with 1,000 users
OpenAI has for years relied on a small, nimble team, composed primarily of researchers, to develop impressive artificial intelligence models rivaling those from Google and Microsoft. Now the seven-year-old startup is acting more like a traditional enterprise software company by building a sales team and aggressively courting customers to use its AI.
The company’s transformation from research-focused startup to deal closer began last year. Months before its ChatGPT chatbot seized the world’s attention, OpenAI began approaching companies to demonstrate its then-unreleased GPT-4 model, pitching them on paying to use the model in their software, several companies told The Information.
OpenAI’s sales push has brought new revenue as it strives to become profitable. Its models are new, and customers are still learning how much it costs to use them. But in interviews with The Information, a half-dozen OpenAI customers provided an early look at their spending with the startup, which ranged from cents per day to tens of thousands of dollars per month.
“It’s not cheap,” said Khan Academy founder and CEO Sal Khan. The education software nonprofit in March launched a tutoring chatbot powered by OpenAI’s GPT-4 model and is already spending more than $20,000 per month on OpenAI’s tech to offer the tool to 1,000 users. But Khan said the AI-powered technology is worth the cost.
OpenAI is now on pace to generate hundreds of millions of dollars in revenue per year, thanks in part to the launch of a paid version of ChatGPT in February, said a person with direct knowledge. An OpenAI spokesperson declined to comment.
OpenAI customers are charged a fraction of a cent per “token,” a word or part of a word generated by OpenAI’s models. Its newer models like GPT-4 cost more per token, while older, less sophisticated models like GPT-3 are cheaper. That means a small company with a dozen employees using OpenAI’s models all day may spend less than $100 per month on OpenAI, while large enterprises using OpenAI’s models in applications available to thousands of users can pay thousands of dollars monthly. In February, OpenAI sent a letter to prospective customers offering dedicated instances—essentially, a copy of its software on a group of servers set up for a single client—with prices ranging up to $156,000 per month. A copy of the letter was first posted to Twitter by developer Travis Fischer, who said he received it secondhand, and The Information independently verified it.
Hiring Sales Leaders
OpenAI’s sales push began in earnest last June, when it hired former WalkMe Vice President Aliisa Rosenthal as its head of sales. Former Stripe sales leader James Dyett joined in February as head of strategic accounts. To assist the sales team, OpenAI has assigned technical staffers to work with customers using its technology, customers say.
OpenAI first approached Khan Academy last August to pitch using OpenAI’s GPT-4 model in an educational chatbot, Khan said. In the following months, developers at Khan Academy worked almost daily with six engineers and sales staff from OpenAI to build and refine the tool. By the time GPT-4 officially launched in March, Khan Academy was a flagship customer, alongside bigger enterprises like Stripe, Salesforce, Zoom and Morgan Stanley.
Khan Academy, which offers its software for free and relies on donations, is asking users of its OpenAI-powered tool to donate $20 a month to offset its costs. As its users—and costs—increase, Khan said the nonprofit may consider switching to the cheaper GPT-3. “Right now we’re focused on making this experience as magical as possible for users, and once we’ve accomplished that we might look at making it more cost-effective,” he said.
Before OpenAI had built out much of its sales operation, it was approached in late 2021 by Morgan Stanley, which has a team in San Francisco tasked with identifying promising makers of AI software. Morgan Stanley leaders including CEO James Gorman and Andy Saperstein, head of wealth management, met with OpenAI CEO Sam Altman, who demonstrated how an early version of GPT could eloquently answer written questions about Morgan Stanley based on past public statements made by executives. Morgan Stanley executives were impressed, according to Jeff McMillan, the company’s chief analytics and data officer, who attended the meeting.
Over the next year, staffers from the two companies built a tool to train an instance of GPT-4 on roughly 100,000 Morgan Stanley internal market intelligence documents. Roughly 300 Morgan Stanley employees began testing the tool in March. McMillan said OpenAI’s support impressed him, because the company historically focused on research rather than building products for customers.
In the past, he said, OpenAI would have
referred big customers to Microsoft, with which OpenAI has a partnership. “They’re not really equipped to manage customers; they want to be a research firm that develops models, period,” McMillan said, referring to the startup. “That being said, OpenAI has kept to their word. They’ve dedicated resources to us, and if I need anything, I pick up the phone, and they work to solve it.”
McMillan declined to say how much Morgan Stanley is paying OpenAI, citing a contractual agreement not to disclose spending. But he predicted that the cost will prove trivial compared with the hours of labor the GPT-4–powered tool will save for Morgan Stanley employees.
Plug-in Partners
In another prong of its sales strategy, OpenAI has formed partnerships with companies that integrate its software with their products. The company last month announced 11 new plug-ins for ChatGPT, including ones from Expedia, Wolfram Alpha and Instacart, that allow the chatbot to query those companies’ databases to provide more factually accurate information. OpenAI first approached these potential partners less than two months ago, several of the companies said.
The plug-ins, currently available only to paid subscribers of OpenAI’s ChatGPT Plus, help fix one of ChatGPT’s biggest shortcomings—identifying factual information. For instance, users can ask ChatGPT to help them plan their itinerary for a trip, and ChatGPT will pull real-time data from Expedia to supplement its answer. OpenAI didn’t pay the plug-in makers to incorporate their data into ChatGPT. Instead, the partnerships are a form of data exchange, where plug-in makers get insights into the types of queries people post to ChatGPT.
At companies that are paying for OpenAI, leaders say the software’s cost is worth the value it brings to customers. Kraftful, a seed-stage startup offering a GPT-powered tool that summarizes customer feedback for product managers, is spending around $8,000 per month to use OpenAI’s models in its software, founder and CEO Yana Welinder said. She said the costs are rising as Kraftful adds customers, who can use the software to automate the labor-intensive task of summarizing feedback. (Welinder is married to OpenAI Vice President of Product Peter Welinder but said her company doesn’t receive any discounts from OpenAI.)
Another big customer that’s renting a dedicated instance from OpenAI is Salesforce, according to Clara Shih, CEO of Salesforce Service Cloud. The sales software giant announced in March that it would begin using OpenAI’s models in its Einstein GPT suite of sales and marketing software, which helps Salesforce customers generate sales emails, parse data and summarize customer feedback. Einstein GPT also uses Salesforce’s proprietary AI models and other models from AI startups Anthropic and Cohere, both of which Salesforce recently invested in, Shih said.
Salesforce is still assessing how much the new services will cost to run and will use that information to set prices for its customers, Shih said.
OpenAI’s sales push is still in its infancy and isn’t as aggressive as some larger software companies. “I’m sure they’ll get there, but they don’t fully look like an enterprise-oriented company yet, which is fine with us,” said Daniel Marcous, co-founder and chief technology officer of tax software startup April, which uses OpenAI’s models to help complete its users’ tax forms. “They’re not like Databricks, which has a thousand salespeople asking you to try out their new products. It’s more like, ‘We have these great products; you have developers; go play with it and see what works for you.’”
While large companies rack up high OpenAI bills, smaller firms who use its software for internal purposes—like generating website text or summarizing meetings—say the cost is marginal. Ignacio Semerene, co-founder and head of product of Bags, a 14-person startup that helps businesses identify and manage loans, said many staff members had been paying $20 per month to use ChatGPT Plus, OpenAI’s subscription-tier chatbot. The startup recently switched to paying for a single corporate subscription for GPT-4, which lets anyone at the company use the service on a per-token basis. Semerene said the company’s OpenAI costs now rarely exceed $2 per day.
In other cases, OpenAI has offered its models for free. Danish company Be My Eyes, which makes a mobile app to assist vision-impaired people, uses OpenAI’s GPT-4, but “there wasn’t any out-of-pocket [cost] for us,” said Mike Buckley, the company’s chair and CEO. He said OpenAI called him earlier this year to discuss the company’s image-to-text technology and ask if Be My Eyes was interested in being a launch partner.