Ramp Connects Petabyte-Scale Unstructured Customer Data to Enterprise LLMs
Ramp, a New York City-based fintech firm, has made significant strides in leveraging artificial intelligence (AI) to make company data more accessible. The company’s goal was to democratize AI use, empowering people throughout the organization to query various types of company data autonomously. According to Ian Macomber, Ramp’s Head of Analytics, everyone at Ramp wants to utilize the power of AI to answer questions quickly, affordably, and accurately, without necessarily having to go through a data team. This is a sentiment that Macomber believes is echoed in many other businesses.
The Data Challenge
However, like many firms with similar goals, Ramp faced a data challenge. The data needed to answer questions was siloed and existed in varied formats, including sales call transcripts, customer emails, customer service tickets, purchase orders, invoices, receipts, memos, and expense policies. Sameer Gupta, a financial services AI leader at EY Americas, pointed out that such data obstacles have often hampered banks’ efforts to provide personalized recommendations or useful virtual assistants. The issue, according to Gupta, lies with the data. Most banking bots can only provide the bank balance before needing to refer the customer to a human agent.
An American Banker survey conducted earlier this year revealed that 29% of bank executives identified data silos and inaccessible data as a significant challenge to implementing AI.
Connecting Unstructured Customer Data
At Ramp, which provides corporate cards, bill payment, and other services to businesses, a valuable trove of customer data could be found in sales call transcripts. The company receives hundreds of thousands of calls every year, as well as a similar volume of customer emails and customer service tickets. A wealth of unstructured data is also contained in documents such as purchase orders, invoices, receipts, memos, and expense policies.
To make all this data searchable by an AI model, Ramp’s team fed it into Snowflake’s Cortex AI, a cloud-based data platform. This was done using Snowflake’s application programming interfaces and its Model Context Protocol Server that connects data sources to large language models and applications from Anthropic, CrewAI, Cursor, Devin by Cognition, Salesforce’s Agentforce, UiPath, and Windsurf.
Understanding Customer Complaints
The primary use of the system at present is to analyse customer data to understand customer complaints. Ramp’s product teams can ask questions of the Snowflake cloud in natural language and receive immediate answers. The ability to analyse customer feedback is a trend in the banking industry, with Wells Fargo and Citizens Bank also using AI for this purpose.
The Legal and Regulatory Considerations
However, Greg Ewing, an attorney with Dickinson Wright, warns that any regulated company looking to use AI to analyse customer feedback needs to be mindful of potential risks. The technology can make mistakes, and misclassifying data could lead to legal liabilities, litigation, regulatory scrutiny, and reputational damage. Another potential issue could arise from the discovery process of a legal case, where an attorney could request all feedback the company has received or all response summaries generated by its AI system.
Democratizing Access and Safeguarding Data
Notwithstanding these risks, Macomber reports that the new system has increased the volume of questions being asked, as it is no longer necessary to tag someone on the data team. Moreover, the system has proven useful for answering deeper questions that would previously have taken machine learning engineers a couple of weeks to address. To protect customer data privacy, Ramp does not put any personally identifiable data into the Snowflake cloud.
As Macomber puts it, “As we have made it easier to ask questions, as we’ve made it easier for people to analyze unstructured data and see themes across expense policies and calls and customer service tickets, more people have done that.” It seems clear, then, that as the technology continues to improve, we can expect broader LLM integration alongside stricter data-governance and legal discovery expectations.
Source: Here
