It began with a desire to answer customer questions and a dream to help them complete tedious tasks more efficiently.
Accounting services provider Xero began slowly and started small when envisioning its AI-powered platform, JAX, short for Just Ask Xero. It was meant to be an avenue for small businesses to ask their particular accounting questions. Now, with the help of OpenAI, JAX is a “superagent” that gives users access to financial data, insights and a clear picture of their finances, as well as the ability to perform tasks like bank reconciliation and draft invoices.
“The first thing we did was the ability in JAX to create invoices and answer questions about invoices. So it's very interesting that today we talk about automated actions,” Xero CEO Sukhinder Singh Cassidy told VentureBeat in an interview. “So that was the very first thing we prototyped and launched to customers. And I'd say the early feedback, which was about a year ago, was, hey, this is cool.”
JAX is currently in open beta, although Xero has not announced when it plans to release JAX in full to the public.
Xero knew it had to move fast to keep pace with the speed of innovation, but it also needed to balance accuracy and control, all things its customers want.
“Accuracy is important, but the other thing that’s important is our customers like the dopamine hit, believe it or not, of reconciling their own transactions,” she said. “Sure, you can magically say ‘it’s done,’ but people want control, so it has to work with one button, and those who want more control can have control.”
So Xero had to make JAX serve those two types of customers. It also needed to offer a way for JAX not to be simply a chatbot, but a platform that provides answers based on the context of the customer’s data and the larger economic world in which they operate.
Baby steps
JAX began its life as an AI companion for Xero customers. Businesses could open JAX and ask it questions about their invoices and receipts, summarize paid and unpaid invoices and create invoices.
Xero is updating JAX to include automated bank reconciliations, where JAX “will work alongside” customers to complete these tasks with full visibility. Users will also be able to ask specific questions about their cash flow and get custom graphs through JAX. For example, they can tell JAX to display profit trends for a specified period. The platform can also suggest actionable insights and bring in external data to give more context to financial decisions like loans or expansion.
Bank reconciliation involves comparing and harmonizing a company’s business records with those of its bank. This process typically involves accounting teams spending hours reviewing receipts, checks, and other documents to match them with what’s logged in their financial accounts.
However, Singh Cassidy said Xero had to bring in guardrails, mainly due to the complexities of accounting. The company developed JAX Assure, a way to monitor and control the flow of information to and from LLMs used in the JAX platform, so that JAX does not solely rely on models to prevent hallucinations.
To build JAX, Xero used models from different providers and is currently using OpenAI’s GPT models. It also utilizes open-source orchestration frameworks to manage the various agents operating under the hood on JAX.
Working with OpenAI
Beyond automating tasks, businesses also wanted to do more research on JAX. While Xero’s knowledge base is deep, Singh Cassidy pointed out that it needs to gather context from other sources.
To give customers a better understanding of their financial position and help them navigate the world, Xero collaborated with OpenAI to leverage the GPT models’ understanding of the world beyond finance. Singh Cassidy likened it to casting a wider net for information beyond the specific knowledge base of Xero.
Olivier Godement, head of product for OpenAI’s API platform, told VentureBeat that working with companies like Xero is part of OpenAI’s ongoing strategy of expanding its collaboration with more enterprise customers.
“We’re steadily ramping up our B2B efforts, especially in the enterprise arena, because they often have to deal with different and more complex types of services,” Godement said. “They are, of course, the industry experts in their fields, but we know how to optimize our models best, and we will help guide them to use our models best for their needs.”
He said OpenAI worked with Xero for a year to help them fine-tune models and build out a more context-aware chat on JAX.
The financial world has slowly embraced AI, understanding that adopting AI tools still requires them to meet regulatory and fiduciary guidelines.
Xero has to compete with companies like Intuit, which also work closely with model providers like OpenAI to offer AI services and agentic tools to businesses. Recently, Intuit outlined how it successfully integrated AI into its interface and built trust with customers.
Startups like Puzzle have also begun offering AI accounting applications, which could automate the majority of the tedious tasks accountants typically perform.