This Software CEO Was Paying Her Customers' AI Bills on Purpose
Vanta's Christina Cacioppo said in October that charging for AI tokens invites a comparison with AI models that software companies will lose, so she was accepting lower profit to get customers using agents.
Invest Like the Best with Patrick O'Shaughnessy · Christina Cacioppo - Building an Enduring Software Company in the Age of AI - [Invest Like the Best, EP.496]In an interview published October 20, Christina Cacioppo, the co-founder and chief executive of Vanta, a company that helps businesses handle security and compliance paperwork, was explaining why she would not charge customers for the AI behind her product's new agents.
“We're taking lower margin and really concentrated on usage,” she said.
By usage, she means customers actually using the agent, not the computing cost itself. At the time of that October interview, she said Vanta was paying the AI bill, which comes from tokens that AI labs charge for, because she wanted more customers working in the new way. She described it as eating the cost for now.
Her reason is blunt. Customers are smart, she argues, and if a software company charges by the token, they quickly decide it has just marked up the price of a token from Claude, a leading AI model, and ask if the software is as good as Claude itself.
That is a comparison she says no software application wants, because the application will lose. Anyone marking up tokens and passing that cost to users will lose on that comparison, she said, so Vanta was accepting the lower margin then.
Her example is what happens when an employee leaves a company. Vanta's agent can write a custom cleanup that removes the person from accounts, then asks for permission to turn it on and to remove access for other people, too.
In that October interview, she said about 25% of Vanta's customers were in that kind of flow right now, in which the agent does work and asks if it can proceed. She wanted to reach 75%, because she believes the experience is much better and will bring much better retention.
The tradeoff is easier for Vanta than for some others, she said. She said coding agents use lots of tokens, with maybe legal work and document drafting next, while the decision and action work in security and compliance uses relatively few tokens.
That lets Vanta "optimize for quality versus being as concerned about cost as some of the other companies," she said.
“This is a usage phase and like a deployment phase of this technology and just go nail that,” she said. “And you can figure out commercialization after.”
Questions people ask
What is a token in this context?
A token is a small chunk of text that an AI model reads or writes. AI labs charge software makers based on how many tokens their products use. When Vanta's agent does work for a customer, Vanta pays that token charge rather than passing it on.
Will Vanta eventually charge for AI?
In the interview, Cacioppo did not lay out a future price. Her words were that if Vanta needs to sort out pricing later, it will sort it out later. In the interview, she called this a usage and deployment phase to get right.
What does she mean by outcomes versus outputs?
She said everyone wants to charge for outcomes, an idea she likes from her background in economics, but defining the unit is hard. She described a closed support ticket as an output and contrasted it with an outcome tied to a working Sonos network, but the example is unclear in the interview. The broader point was that companies are still searching for a unit of AI work customers can agree on.