Companies need a crystal clear, accurate data set to get something worthwhile out of AI.
Emily Reaney, Head of Treasury at Travelodge Hotels
The hype around AI is unquestionably in overdrive, dominating discussions from boardrooms to conference floors, as I experienced firsthand at the recent Association of Corporate Treasurers (ACT) conference in Liverpool. No one can avoid the buzz around the technology’s potential to revolutionise the treasury function.
It became clear in my conversations with treasurers that they are feeling the AI pressure ramping up week on week. They are being overwhelmed with questions about AI readiness, whether it’s their CFO asking for an AI roadmap or a treasury management system (TMS) vendor pitching an AI-powered cash flow forecasting tool.
While treasurers are keen to adopt AI in theory to drive productivity gains and efficiencies, they remain sceptical about its use within a highly regulated finance and treasury environment, as they don’t yet trust the accuracy of its output.
For example, one treasurer spoke about using an AI-powered cash flow forecasting tool embedded in their TMS. However, they were running their standard process in parallel, as the tool wasn’t mature enough to be completely relied on. Another was exploring deploying an AI bot that allowed different parts of the business to query the status of a payment, but voiced concerns about potentially exposing sensitive data.
Many feel paralysed in the face of these challenges as to where to start their AI journey in practice. But treasury teams that have begun meaningful AI deployment were consistent on their view of where to begin: treasury data.
Data quality and accessibility is commonly cited by treasurers as the biggest barrier to effective AI adoption. Treasury data is often siloed, fragmented, manually retrieved, inconsistently formatted, duplicated, and ownerless. This is due to the complex operating environment in large corporate treasuries, with multiple back-office systems and numerous banking relationships. The variety of payment and bank statement data formats also adds to the complexity.
An environment where there are data gaps, no standardisation, no consistency, and no clarity over who has control and what’s out of date undermines trust in the output. As one group treasurer put it: “If ‘bad data’ is plugged into an AI model, treasury could get a very quick answer that is very wrong.”
Treasurers need clean, connected, and consistently defined data that teams can rely on without caveats or hours of manual reconciliation. Many practitioners at the ACT conference talked about the need for a “single source of truth” and end-to-end automation. But this can only happen when the data is cleansed, centralised and standardised.
Getting the basics right
A best practice approach is to standardise the capture of clean data from the outset. But often treasurers are forced to deprioritise an important initial step in ensuring their data is clean and accurate: establishing a reliable, automated connection between their back-office systems and their banks. Many are still manually transferring data between banking portals and TMS and enterprise resource planning (ERP) systems. Automating this information flow should be the first stage in ensuring data accuracy.
Yet while connecting back-office finance systems to banks may seem simple in theory, it’s often challenging in practice, with numerous hurdles from technical issues to formatting limitations. Fundamentally, TMS/ERPs are not designed to handle the full operational complexity of corporate-to-bank connectivity.
In creating a clean data foundation, bank connectivity solutions have a vital role to play. Using a specialist connectivity layer, treasurers can automate payments and bank statement data flows by centralising feeds between back-office systems and banks, and crucially converting the data into the required format for different systems.
AccessPay, for example, has automated statement retrieval, format normalisation across more than 16,000 banks, and a single connection point regardless of how many banks or payment schemes are in play. We solve the last-mile problem in ERP/TMS/bank connectivity, facilitating end-to-end automation.
As well as improving cash visibility and reducing the risk of fraud and errors, adding an agnostic connectivity layer creates the clean, real-time, governed data feed that is the prerequisite for AI deployment in treasury to produce trustworthy outputs rather than fast wrong ones.
Rising to the AI challenge
The direction of travel is clear. A global McKinsey survey of CFOs in late 2025 reported that 44% of respondents had identified five or more AI use cases, compared with just 7% in 2024. In addition, recent Gartner research demonstrates how AI agents are starting to feature in finance workflows.
In response to an AccessPay survey last November, 28% of corporates said they have implemented AI enhancements in finance operations to a high degree, while another 38% partially implemented AI in select workflows or where vendors included it in their product. Interestingly, 68% stated AI would be embedded in at least 50% of their finance operations within the next 18 months.
Yet, as evidenced by the discussions at the ACT conference, many are not yet ready for AI adoption. Most organisations point to the need to get their data under control before progressing. If this isn’t done properly, AI adoption is meaningless. As one treasurer said, “The focus has to be on walking before we can run – and that centres a lot on the data.”
The data foundation must come first. In the next 12 months, treasurers should look to:
- Simplify
- Standardise
- Control
- Integrate
- Automate
A key takeaway in Liverpool was that it’s essential that the treasury team is part of the data governance process, as IT teams can’t be expected to know what treasury needs from those data flows.
In the near term, AI implementations are expected to focus on specific tasks within finance rather than full-scale automation of operations. Looking further out, finance professionals will increasingly work hand-in-hand with AI agents to perform repetitive tasks.
However, for this approach to be successful, the underlying data needs to be in order. AI agents require structured, real-time data to function effectively, which is where manual workflows for transferring data between systems fall short and where an automated data layer will pay dividends.
