Ask a room of corporate treasurers what their two greatest priorities are, and the answers rarely surprise: cash and liquidity management, and fraud prevention. Ask what is stopping them from doing both better, and the answer is almost always the same word: data.

Glisson Inguito, treasurer at Konica Minolta, has spent five years building an AI-powered forecasting and cash application capability that now runs largely without manual intervention. Anshul Patni, vice-president and treasurer at Bakelite, joined the company two years ago and found a treasury function that, as he described, was heavily reliant on spreadsheets. Since then, he has built the foundation needed to support future AI capabilities.

Together, at EuroFinance’s 30th Annual International Treasury & Cash Management Summit in Miami, they traced the full arc of what treasury transformation actually looks like in practice: from the unglamorous work of system integration and data governance to the moment when the forecast runs itself.

The foundation problem

What Patni found at Bakelite,  a leading global integrated producer of thermoset specialty resins, solutions and engineered thermoset molding compounds, was, in his own assessment, a treasury function in the first category of automation maturity. “We lived in spreadsheets,” he said. “Our cash flow forecasting was not as robust as it needed to be and didn’t consistently support  decision-making, particularly when informing leadership about expected cash positions over the next 13 weeks.” Bank connectivity was limited, there was no real-time visibility, and the company’s global cash pooling structures, critical for a multinational chemical company with entities across North America, Europe, and South America, were not yet operating at the level of efficiency the business required.

His instinct was not to reach immediately for artificial intelligence. It was to build what he calls a “treasury foundation”: standardised processes, a treasury management system (TMS) capable of integrating with the company’s SAP environment, properly configured cash pooling structures in its two primary regions, and investment tools connected to live bank data. “The most important thing is, are my treasury processes standardised? Do I have the right data in place before I begin this AI journey? Otherwise, it’s going to be a huge expenditure without  yielding the results that you want,” he said.

That story, he said, “really helped ensure all stakeholders were aligned as we moved through the TMS journey.” Two years on, Bakelite’s treasury has achieved a significant improvement in visibility. All bank accounts now report into its TMS, and the CFO receives a weekly cash position report without manual intervention. “On a weekly basis now, my CFO has the report at the click of a button,” Patni said. Investment tools and pooling structures feed into the same system. The next question is what artificial intelligence can do on top of that foundation.

What the foundation enables

For Inguito at Konica Minolta, a technology company that specialises in digital workplaces that question is already being answered. His team began its automation journey roughly five years ago, and the starting point was not cash forecasting but accounts receivable.

“DSO was the most critical thing during COVID that was impacting our cash flow,” Inguito explained. “It was going up to 90 days. Normally for us it’s 55 to 60 days.” The worsening situation, daily revenues multiplied by the extra days outstanding, made the cost clear to management. Konica Minolta has approximately 75,000 business-to-business customers, many of whom do not include payment details with their remittances. Collections staff were manually searching emails to match payments against invoices before posting transactions through the clearing account process. The company partnered with a fintech provider to automate cash applications using machine learning. The implementation took six months.

That project, once completed, created the appetite for more. Treasury followed, then credit and collections as an integrated cycle. “Because we know the ultimate vision we have is our cash flow, our basic metric or the top metric is net cash flow,” Inguito said. “So we had an insider’s view of where we are.”

Inguito noted that the current architecture is considerably more sophisticated than the prior state, which involved pulling bank data from portals, loading it manually into spreadsheets, running reports for accounts receivable and payable, and assembling an 18-week rolling forecast that he estimated consumed roughly 80 hours of staff time per month. Now, data flows via Secure File Transfer Protocol (SFTP) into the fintech platform on a daily basis. Bank reports arrive electronically. The forecast horizon has extended from 18 weeks to 13 months, and the system runs agentic jobs that generate the forecast each Wednesday without manual triggering. Variance analysis runs at least twice a month, again without human initiation.

The accuracy is impressive. “So far 97% accuracy,” Inguito said, measured against net cash flow relative to loan balances, a metric he regards as more meaningful for a net borrower than an absolute percentage. That accuracy has direct financial consequences. “Because of better forecasting, we’re able to reduce our excess cash and use that to pay down debt. We also size our credit lines with our banks and also the intercompany pooling borrowings that we do.” He estimated savings of close to $1.5 million annually in interest expense and commitment fees.

Current macroeconomic disruptions have been absorbed without losing discipline. Konica Minolta imports heavily from China. When tariff rates reached 105% this year, the team imported a data sheet into the platform to reflect the adjusted numbers. “Without those tools, I don’t think I would be able to go home,” Inguito joked.

Data lakes and conversational analytics

The ambition beyond the current state is to move treasury data into the enterprise. Konica Minolta is preparing to bring three years of cash management and forecasting data into a corporate data lake. The objective is to enable prescriptive and diagnostic analytics: to understand why events occurred, and to allow senior management to query the data conversationally.

“We could do prescriptive analytics, basically looking at the numbers, not just the variance analysis, but we’ll be able to do conversational analytics to have our CFO and other senior management prompt,” Inguito said. The initiative also aims to identify the real drivers behind collection performance, cost trends, and cash flow variability. The data lake, he argued, would answer those questions systematically. Internal audit and IT will monitor the data pipeline for integrity. “We own the data,” Inguito said. “Treasury is owned by us, so we need to validate that data as it comes in.”

For Patni, developing predictive forecasting models, whether invoice-based or regression-based, is a near-term priority, even as the team continues to stabilise its newly implemented TMS. He noted that cash forecasting is one area where TMS systems have inherent limitations and may require  additional AI-powered tools.

The recurring obstacles

Neither journey has been without friction. Both treasurers pointed less to technology and more to organisational factors

For Patni, the fundamental challenge was making the business case compelling enough to prioritise. “How do you create a business case for your CFO and other finance stakeholders so that they are aligned on moving forward on this journey?” he asked. IT alignment, he emphasised, is often underestimated. “I’ve seen treasurers lose their jobs when projects are not successful. IT is one of the most important stakeholders.”

Change management within the treasury team itself also required attention. Even after implementation, staff reverted to familiar spreadsheets. “People, even after going live, still trusted  their existing processes,” Patni said. His response was to run parallel processes for a period—but then to retire the old ones deliberately and completely. “After a certain point, you have to retire your existing processes. That’s very important, otherwise you will not be able to move forward.”

Inguito faced different issues around data ownership. When integrating the fintech platform with the company’s internal systems, the provider initially sought to manage the end-to-end data transmission. Inguito pushed back, arguing that the company’s Electronic Data Interchange (EDI) department, and ultimately its data analytics team, should own the pipeline. “IT just makes sure the data is consistent,” he said. “The business should own the process.”

Konica Minolta established a company-wide AI governance steering committee early in the AI journey, involving HR, sales, and other functions. The committee set the parameters for what AI tools could be used internally, leading to the adoption of established standard AI tools.

A staged view of maturity

Treasury automation is not a single project but a staged accumulation of capabilities, each dependent on what came before. Cash application at scale requires clean customer and vendor master data. Cash forecasting at 13-month horizons requires reliable daily bank feeds and structured Enterprise resource planning (ERP) data. Prescriptive analytics requires years of consistent, governed historical data in a format that an enterprise data lake can ingest.

The treasurers who attempt to skip stages by deploying AI on top of fragmented data and manual processes, are likely to find that data quality remains the primary impediment to value realisation. Those who do the less glamorous work first find that the later stages follow with greater speed and credibility.

“We are now able to put our feet in the water,” Patni said of his current position. “We feel that we are stable with what we have automated so far, and next up is looking into forecasting models.”

For Inguito, the stabilisation phase is already in the past. The question now is how much of the analytical burden can be formalised  and governed. The answer, he suggests, will arrive not through any single technology, but through the patient construction of the infrastructure that makes such technology trustworthy.