Microsoft is among the companies building AI copilots for corporate use.
David Paul Morris/Bloomberg
Much of the early speculation around new forms of artificial intelligence surrounds its impact on jobs. A handful of early adopters within the payments industry are providing a different view of AI’s impact — how it helps workers perform everyday tasks.
Payments fintech Stax, for example, has not implemented AI fully into its daily internal processes, but Mark Sundt, who recently became Stax’s chief technology officer, said he and his team have been using new, advanced forms of AI copilots internally.
The payment company uses AI to create meeting transcripts and summarize key takeaways or to-do lists for those who either missed the meeting or were unable to take notes. “This removes barriers some may face when attending meetings about extremely technical subjects and facilitates a quicker understanding of company objectives,” Sundt said.
“We have found that it increases productivity while saving time on day-to-day workflows and mindless tasks,” Sundt said.
Stax and other companies such as Tipalti and Stripe are using chatGPT-style generative AI, large language models or some equivalent such as AI copilots, which are LLM-powered digital assistants. These programs are often being tested or deployed in parallel with customer-facing AI tools for customer service, product development or risk management.
Australia-based ANZ Bank recently reported that a Github copilot helped developers write code 42% faster than a control group that did not use the copilot, according to Economywatch, a local technology news site. The bank, which did not provide comment for this article, was testing the ability of the AI copilot to help its development staff produce quality code faster while maintaining security.
The bank reported that the copilot-aided code had fewer errors and bugs, but the details on security were inconclusive. That said, it was unlikely the copilot introduced security risks. ANZ overall found the copilot “compelling.”
“This will be a generation-changing technology,” said Sarah Spoja, CFO at digital payment company Tipalti. “While it will take time to fully adopt gen AI, it’s important for people to start using these tools right now to start learning what can be done.”
Generation changing
Tipalti’s Spoja refers to her company’s product as “copilot-like.” For CFOs in general, Spoja said competitiveness is on the line regarding embedding AI into their organizations. She likened it to the introduction of Excel in the mid-1990s, noting that many CFOs were resistant at first, but now every financial office runs on Excel.
“It can save time and find gaps,” Spoja said. “There are a number of ways that this can be used.”
Tipalti is using gen AI both internally and in customer-facing situations. Tipalti recently embedded OpenAI’s GPT-4 technology to bolster its AI capabilities for several activities. Auto coding analyzes the context of a purchase order, an invoice and a ledger tied to a payment to avoid manual coding. Another product, an AI-powered chat feature, also analyzes data to enable what the company refers to as an “intelligent assistant.”
Chat GPT-4 is a newer form of OpenAI’s Chat GPT that is designed to improve the quality of responses to user queries with upgraded “problem solving” capabilities. Programs that use LLMs, such as ChatGPT, are designed to produce original content by analyzing existing data and instructions from users.
“You can query the database quickly using normal language that all of us are used to using, like using Alexa at home,” Spoja said.
For workplaces, that can mean helping employees do research, schedule their time, or develop content to inform a conversation with a colleague or client.
To use the chat, staff and clients ask questions that are directly tied to business functions. Examples include: “Who are my 10 slowest approvers?” or “How much volume was spent last year on educational projects for employees?” or “Can you identify major trends on budget that were spent on travel and entertainment year or year?” or “What are the major reasons sent back to accounts payable for corrections?”
Virtual helper
Microsoft and other firms sell their own AI copilots. And other gen AI programs can perform functions similar to a copilot, including AI software that uses natural language AI.
“It can be a copilot or a virtual assistant. What it really refers to is an interface that you can interact with in writing to produce content that can complete a task,” said Jason Wong, a vice president and analyst at Gartner.
Gartner’s research found that 42% of bank CIOs have deployed or are planning to deploy gen AI in 2024, using it more for internal applications as opposed to customer-facing ones. The most common uses are coding, personalized product development and marketing content generation, where AI can gather and analyze data faster than humans.
“It can give the user something more contextualized to act on,” Wong said.
Payment companies have embraced gen AI and similar technologies over the past year, creating momentum for experiments with copilots.
Stripe, for example, has added AI for internal use, external use and acts as a payment provider for OpenAI. Klarna has distributed gen AI to about half of its employees and plans to have nearly all of the company’s staff use gen AI to aid in their jobs. Klarna has also changed its recruiting policy, reportedly deemphasizing jobs outside of engineering due to gen AI’s impact on automating jobs. Klarna, which did not provide comment for this article, has reported that AI has saved the company millions of dollars due to reducing inefficiencies.
“Our early experiments with LLMs last year led us to introduce copilot-style offerings like Sigma Assistant and Radar Assistant,” said Emily Sands, head of data science and machine learning infrastructure at Stripe, in an email. “For businesses using Sigma (data analysis product) and Radar (fraud detection product), these new assistants make it possible to glean business insights and write custom fraud rules even faster, by asking in plain English.”
Stax is using Microsoft Copilot predominately and has been investigating ChatGPT for developer documentation. It’s also looking at IBM WatsonX for its financial services (Granite) LLM, called Granit, and governance capabilities, and at Hugging Face LLMs for internal Retrieval Augment Generation (RAG) solutions, Sundt said.
In addition to helping record meetings, Stax uses AI to aid in creating job descriptions for open positions.
“While I do not believe that AI should ever be used to replace hiring managers, it has been extremely helpful in communicating job expectations and processes to potential candidates way faster than I could,” Sundt said.
In terms of results, Stax has seen improvements in developer productivity. In the past, developers would have to look up codes and watch video tutorials, according to Sundt. “Now developers can prompt the tool to create the code they are looking for and copy and paste exactly what they need, significantly reducing time and enhancing efficiency.”
