Iress uses GenAI to help it respond to procurement requests

Iress, a maker of financial services software, used generative AI to help it to respond to lengthy requests for information (RFIs) and request for proposals (RFPs) used in procurement processes.

Head of business intelligence Jeff Gibson told the Databricks Data Intelligence Day in Melbourne that the company wanted to “solve a real business problem” with GenAI from the outset.

“We didn’t want to do another lab [experiment], something that gives us some theory. We wanted to actually solve a real business use case and learn how to implement and govern AI,” Gibson said.

“We hear about all these great things [about GenAI] but we wanted to actually figure out what it actually takes [to utilise it], and we also wanted to make sure that we understood the costs of building these types of technologies and products.”

The company settled on a challenge faced by its commercial and product teams and brought representatives from those teams together with people from the BI and data innovation teams, and from Databricks, to try to build something over a two-day period last year.

“We found a problem within the business which was handling our requests for proposals/requests for information,” Gibson said.

“Many businesses get clients asking for this type of information. This could, for us, be between 50 to 950 questions. We have to source this information from all across the business, from specialists and documentation, and it can be a very difficult process. 

“We’re often under strain and stress to get it to the prospect or client, and we don’t have a lot of time to spend on the sales process, so it can take up to three-to-six weeks, involve multiple teams and can incur significant costs.”

Gibson said that the exercise was purposely limited to two days - and one night - and that the team “didn’t want to walk out of the room and not have something to show back to the business.”

In those two days, it produced an app called ‘Bella’ that provides an interface where questions can be asked, and data is queried to return a response.

Gibson noted that there were other possible ways to query data - such as by putting a search engine over top of a spreadsheet or intranet site - but he said that GenAI added the dimension of context.

“Why start with GenAI? Because what it added to the whole process was context - context about the question, the client, and how that information can be presented back,” he said.

Gibson said that Bella was produced using Databricks, AWS for the underlying storage, and Hugging Face “to go and look for the [AI] models and bring those in.”

He noted that Databricks’ platform had evolved significantly even in the short space of time since Bella, and that additional capabilities available today would be beneficial to future work.

Gibson said that aside from solving a business problem, the two-day sprint had produced “invaluable learnings” for those involved and had helped Iress to chart a path forward on GenAI.

“We had a cross-functional team that got to learn a huge amount in such a small period of time,” he said.

“This has informed our internal GenAI roadmap, not just for these internal use cases, but we build products, and we can see now how we can start to build this into our product and deliver this to our clients.”

He added that the two-day project had also highlighted the importance of good data management when pursuing AI.

“We came to the realisation that if we don’t get the way we manage our data right, we’re not going to be able to do this properly or manage it ongoing,” Gibson said.

“So, it’s not only informed our GenAI roadmap but also informed the way we’re going to do data management and our data governance.”