When the treasury team at Moody’s Corp. found themselves constrained by fragmented bank connectivity and Excel-based forecasting, they turned to machine learning to modernize the process. By implementing a forecasting solution within ION’s Reval TMS, they automated manual workflows, improved forecast accuracy and freed the team to focus on value-added analysis—winning Treasury Management International’s 2025 award for Best AI Automation. - At NeuGroup’s recent Treasurers Symposium, chief treasury officer Deepali Chawla and treasury VP Amit Bhatt shared that the project delivered more timely, accurate forecasts while allowing treasury to dedicate more time to strategic decision-making.
- “The teams working on the forecasting are now able to shift their workload, going from cleaning up errors and cleaning up Excel to doing work that’s more value-add,” Mr. Bhatt said. “The teams doing the work felt that they now won’t be spending hours going through the files.”
- They are now using forecasts to inform decisions around liquidity, repatriation timing and working capital—decisions that previously would have depended on manual analysis and lagging data.
A complex mandate for modern forecasting. Moody’s treasury team supports over 120 legal entities across Asia, Europe and the Americas, with hundreds of bank accounts, some regional cash pools and an internal dividend distribution strategy—managing billions of dollars. The company operates in a highly regulated space due to its ratings business, reporting to the SEC and other authorities globally.
- Among treasury’s key tasks is a rolling 12- to 18-month cash flow forecast submitted quarterly to regional regulatory supervisors. The forecast must support not only regulatory obligations but also decisions around capital returns, including shareholder dividends and internal repatriation.
- “We have to show cash flow forecasts every quarter as part of board submissions,” Ms. Chawla said. “It’s important we make directors comfortable with the dividend decisions they’re signing off on—and that it’s supported by the forecast.”
- Before the transformation, the forecasting process was fragmented and manual, built on legacy systems and inconsistent assumptions across regions and functions, eventually compiled via basic linear regression models run in Excel. Forecasts often had to reconcile input from FP&A, tax and business units, with a wide variation in forecast logic.
- “We had a mishmash of models—some direct, some indirect, some detailed, others not,” Ms. Chawla said. “Different teams had different views of cash, and the data was coming from too many disconnected sources.”
Early adopters of AI. The transformation began in 2022, just as AI momentum was accelerating. Moody’s CEO Rob Fauber emphasized that generative AI should be seen as a tool for human empowerment, not replacement, a perspective that resonated strongly with the treasury team.
- “We wanted to be an early adopter; we wanted to be in the journey,” Mr. Bhatt said. “So what can we do to add value to this piece? When we looked at ION’s product, it solved two problems: they can make it a one-stop shop, and the capability to have machine learning that takes over some manual work for us.”
- After an in-depth process evaluating a number of options to implement AI-driven cash forecasting, the treasury team, which already relied on Reval as a TMS, opted for a module within it that uses machine learning.
- “We wanted a system that could support straight-through processes, work with our ERP and banking data, and provide machine learning capabilities,” Mr. Bhatt said.
- For Ms. Chawla, one of the key takeaways from the process is that a technical implementation can be managed within treasury. “We didn’t need IT, though we got light guidance from our financial services team on managing data on the back end,” Ms. Chawla said. “But in a lot of this we just worked with ION directly and relied on those with the skillset within our team.”
From spreadsheets to smart models. To start, the team uploaded three years of historical AP and AR data to train forecasting models. The models initially used various types of linear regression analysis, then advanced to a neural network—an AI algorithm modeled after the pathways of a human brain—as the system absorbed more data and detected patterns. - The AI-based engine helps the team build forecasts by account, entity or region. In some cases, the team needed to forecast at a high level for a group of accounts; in others, they dug deeper—especially when modeling working capital drivers, like customer collections.
- “A real benefit of this is we’re able to compare the forecast to actuals on a routine basis,” Mr. Bhatt said. “As our FP&A team’s forecast comes in, we’re able to marry the two.”
Data cleanup. While the AI-enabled system offered promise, treasury quickly realized that the transformation would only be as good as the data feeding it. “We didn’t fully appreciate what ‘garbage in, garbage out’ meant until we did this,” Mr. Bhatt said.
- The team developed tagging and classification logic across accounts, working closely with controllership and tax partners to interpret transaction narratives and clean up categories. “For machine learning to work, the system needs to understand each transaction,” Ms. Chawla said. “The more we clean the data, the better the output.”
Payoffs for Moody’s and others. One of the most significant results of the initiative has been how it reshaped the role of treasury staff. Rather than waiting for forecast variances to show up after the fact, treasury can now anticipate issues—such as liquidity constraints, mismatches in netting or shifts in repatriation—before they materialize.
- “We’ve built something that’s standard, scalable and forward-looking,” Ms. Chawla said.
- It’s also something other corporates will benefit from. Mr. Bhatt noted Moody’s worked with ION throughout the project to develop features for the module that include reconciliation workflows, advanced tagging logic and cash classification models based on behavioral learning.
- “We effectively helped ION shape what this module looks like now,” Mr. Bhatt said.
