AI can return an answer quickly. In finance, speed doesn’t mean much if you can’t trust the answer.
If a system confuses billings with net sales, or a channel partner with an end customer, it can give you an answer that sounds perfectly plausible and is still wrong.
That was the standard we set for ourselves when we began building an AI-ready finance data foundation inside Ericsson Enterprise Wireless Solutions.
We didn’t start with the chatbot. We started with the data.
Our finance information lived across two separate ERP environments and several other systems. Answering questions about billings, plan variance, security attach rates, active devices, or gross margin could mean exporting information, joining it manually, and reconciling definitions before anyone could use it.
So, colleagues across Finance, AI and Analytics, Data, and IT began building the foundation underneath the experience we wanted to create.
We brought finance-relevant information together in a governed data environment and created seven semantic views across the areas our teams use to understand the business. More importantly, we encoded the definitions behind those measures so the system could distinguish between concepts such as billings, net sales, and orders booked.
That foundation now powers three production AI agents: Crystal for finance, Phineas for the commercial team and Aurelius for the product team. All three utilize the same governed data, helping teams work from consistent business definitions regardless of where the question originates.
One example brings the difference to life. Åsa, the head of Enterprise Wireless Solutions, asked why one of our sales regions was underperforming compared with the prior year. Getting an answer back took eight people and two weeks. The work involved multiple meetings to agree on definitions, carry out the analysis and align on how to present the findings.
Today, that same analysis can be done in minutes. The data is connected, and the definitions are already agreed. Our teams can spend more time discussing what the findings mean and what to do next.
One of the things I’m proudest of is that we did not treat AI as a substitute for expertise.
When we needed to bring SAP data into the foundation, for example, we used an AI coding agent to accelerate the modeling work, with finance and engineering experts reviewing the results. Work that had been estimated at more than eight weeks was production-ready in three days.
That experience reinforced something for me. The visible AI experience is only the last mile.
The harder work is underneath it: connecting information, agreeing on what business measures actually mean, putting the right governance around the data and bringing together people who understand both the business and the technology.
Our core team was small. Three people led the build, working closely with colleagues across finance and IT and using AI itself as a force multiplier. Today, Crystal and Phineas have served 88 users and answered more than 12,000 prompts.
That work has now been recognized with the 2026 Gartner Finance AI-Ready Data Breakthrough of the Year Award.
I’m proud of the recognition. I’m even more proud of how the team got there.
If there is one thing I would tell another finance leader thinking about AI, it is this: before asking what AI can do with your data, make sure your business agrees on what the data means.
That is where the real work starts.
Learn more: https://www.gartner.com/en/about/awards/finance-awards
