|
Welcome back! While artificial intelligence agents continue to generate mixed results, depending on how much assistance customers get from their AI vendors, some of them have gained a new skill: negotiating software contracts. Over the past year, data management startup Cribl has been using an AI agent from procurement software startup Zip that reads Cribl’s contracts, determines when it may be overpaying and automatically drafts emails to vendors asking for lower prices. To the vendors, the emails appear to be coming from Cribl’s procurement staff, who ultimately send the emails that the agents draft, said Meteb Alfayez, director of procurement at Cribl, which is privately valued at $3.5 billion. Cribl signs hundreds of contracts a year for software and other services, he said. While Alfayez’s team negotiates the largest contracts to try to winnow down costs, it doesn't have the bandwidth to negotiate smaller deals worth tens of thousands of dollars annually. Previously, Cribl would have just approved some of the smaller deals at whatever rate its vendors offered, he said. Cribl still has humans review the emails the AI agent drafts, and so far has only entrusted the agent with negotiating smaller deals. But the tool will save the company roughly $3 million this year alone. Alfayez declined to share how much Cribl has paid Zip for the agent but said the savings outweighed the cost of the tool. “Individually these contracts are not material, but if you combine them all, they become material,” he said. “[W]e don’t have time to nickel and dime all these vendors, but the agent does.” Alfayez said Cribl got hands-on help from Zip, a five-year-old startup founded by former Airbnb engineers, in developing the AI agent. Before Cribl’s finance staff started using the agent widely, they tested it for several months and provided feedback on its responses so that Zip could improve the quality of its answers. That tracks with a broader theme we’ve heard in AI, where AI pilots are often much more successful when companies get a high level of technical support from the startups selling the tools. Zip’s AI agent uses models from OpenAI and Anthropic, according to a company spokesperson. Crucially, it also uses proprietary contract data Zip has licensed from software management companies like Vendr and Tropic, which provide a baseline of how much companies have paid for specific types of software in the past. And Zip engineers worked to connect the agent to Cribl’s internal contracts data so it had more context on how much it would typically spend on software. The tool still isn’t perfect. Sometimes it can’t find applicable price data and gets stumped when drafting emails or simply tells users that there’s no data available, Alfayez said. Still, he said he’s optimistic the agent can eventually be good enough to handle even bigger contract negotiations. “Working with AI agents is like raising a baby but they grow 100 years every day,” Alfayez said. What about when vendors’ AI is good enough to beat back the AI hagglers from software customers? It stands to reason that Zip and Cribl, which both sell software, will themselves have to use AI to handle agents their own customers use to negotiate pricing! Will the dueling AI’s cancel each other out? Alfayez still thinks that’s a long way off but could lead to fairer deals if both sides use AI that knows the market rate. “I haven’t seen salespeople rely on AI yet, but if they do, that could mean we're going to minimize the back and forth,” Alfayez said. “I could see that being a win-win situation.” We like his optimism! But let us know what you think. AWS and Microsoft Vet Takes Aim at Enterprise AI Logjam For two decades, Baskar Sridharan ran some of the world’s largest-scale computing systems as an engineering executive at Microsoft, Google and Amazon Web Services. But in his view, cloud providers haven’t been effective enough at helping large companies, especially ones in regulated industries, get value from AI. That’s one reason he recently left AWS, where he was vice president of AI, machine learning, data services and infrastructure, to become president of Trase Systems, a two-year-old startup that sells AI for healthcare, oil and gas companies, and federal agencies. Sridharan’s role involves running Trase’s product and engineering teams, as well as business operations, AI research, and sales and marketing. At AWS, he oversaw key AI products like SageMaker and Bedrock. In a recent interview, Sridharan said Trase has gained traction with companies in regulated industries by helping them move at their own pace and understanding what it takes to get AI projects over the finish line. “When I speak to CIOs and CTOs, they don't actually know how to take their prototypes that they have successfully built on either AWS or Azure or Google Cloud and put them into production,” Sridharan said. “No cloud provider is ever going to come and spend the time that we spend with you understanding your humans and processes and requirements.” Trase uses technical staff known as forward-deployed engineers, a practice pioneered by Palantir to help big companies quickly get AI projects up and running. They study customers’ internal business operations to figure out where AI can have the biggest impact, and then help them build small AI models for specific applications. For example, Duke Health recently used Trase’s AI to upgrade the healthcare provider’s prescription refill system. The project has already made the turnaround times for refills three times faster and yielded nearly $200,000 in annualized cost savings, according to a Trase spokesperson. Another Trase customer is the U.S. Navy, though Sridharan declined to discuss its AI usage. Once customers familiarize themselves with AI, they can use Trase’s main product—a service for building agents that tap their internal data, Sridharan said. Trase is part of a growing trend of AI providers using FDEs to help big customers overcome the pitfalls that can stall AI projects. Trase recently raised $10.5 million in pre-seed funding from Red Cell Partners, a McLean, Va.-based incubator, and Virginia Venture Partners, the commonwealth’s venture firm. Trase CEO Grant Verstandig is the founder, chairman and CEO of Red Cell Partners, and the startup’s CTO Joe Laws is a former U.S. Army Captain who has held technical leadership and engineering roles at Google, Dropbox and Reddit, according to his LinkedIn profile.
Thank you for reading the Applied AI newsletter! I’d love your feedback, ideas and tips: aaron@theinformation.com. If you think someone else might enjoy this newsletter, please pass it forward or they can sign up here.
|
