Laurent Bonhomme and Sébastien Schweickert share their insights with Redbridge about Groupement Les Mousquetaires’ success in improving its treasury operations (cash flow forecasting) using a predictive artificial intelligence model. Initiated five years ago, this groundbreaking project has delivered tangible benefits, driven by the dedication of an expert three-person team. It represents a compelling example of the opportunities, but also the challenges, that integrating artificial intelligence in treasury processes can involve.
Artificial intelligence in treasury: hype or reality?
The mention of artificial intelligence in treasury often conjures up visions of a utopian future – a world in which machines take over mundane tasks, freeing up treasurers to focus solely on analysis and strategic, value-adding activities, and in which they finally earn the recognition and favor of company management. But is this likely to be the reality?
Groupement Les Mousquetaires: pioneering the use of AI in cash flow forecasting
The story of Groupement Les Mousquetaires stands out as a rare example of how artificial intelligence can deliver a real, measurable positive impact to a treasury’s operations. Groupement Les Mousquetaires, the umbrella organization for retail brands such as Intermarché, Netto, Bricomarché, Brico Cash, Bricorama, Roady and Rapid Pare-Brise, has successfully integrated artificial intelligence in its cash flow forecasting processes. The initiative, which was launched in 2019, was spearheaded by Sébastien Schweickert, data analyst, and championed by Laurent Bonhomme, Director of Financing, Treasury and Investor Relations at Groupement Les Mousquetaires.
At a client breakfast hosted by Redbridge’s Treasury Transformation team in mid-November, Schweickert and Bonhomme shared some insights on the project. They emphasized the challenges of integrating artificial intelligence in cash flow forecasting, including the need for significant long-term resource allocation and ongoing oversight of the data set. This kind of vigilant monitoring ensures that the model can become more accurate over time and avoid stagnation or the propagation of inefficiencies.
Thanks to their efforts, Les Mousquetaires has successfully reduced the margin of error of its 12-week forecasts from 3% to just 1% – a level that Laurent Bonhomme describes as “acceptable for making financial decisions.” The Group estimates that it is making annual savings of €1 million thanks to its more efficient allocation of available cash and more precise calibration of bank borrowing and commercial paper issuance (NEU CP).
The background to the project
The project to enhance cash flow forecasts using a predictive AI model known as Cash Flow Management (CFM) began with the establishment of a data lab aimed at “realigning the planets between business expertise and data“, according to Laurent Bonhomme, who elaborates:
“Data was the first step in launching our project. The second was determining whether the effort was justified. For Les Mousquetaires, the answer was a clear yes given the significant difference between the cost of short-term financing and the return on cash investments. The third step involved ensuring the sustainability of the cash flow forecasting model through robust governance and making it accessible to all of the group’s departments. Finally, the fourth step focused on testing the model against our historical forecasts.”
The CFM project involves direct modelling of no less than €15 billion in cash flows over three months. Beyond this timeframe, the Group’s cash flow becomes too unpredictable to forecast due to various external factors. The project relies on a dedicated team consisting of 1.5 full-time equivalent data scientists and 0.5 data engineers. They are guided by the “third musketeer,” Sébastien Schweickert.
According to Schweickert, the group’s previous cash flow forecasting model, which was built in Excel, struggled to detect recurring patterns in cash inflows and outflows. “Although it was time-consuming and prone to long-term errors, the old model worked. Our challenge was to make the data more usable and eliminate the inaccuracies,” he explains.
Three years after the launch of the CFM project, Les Mousquetaires continues to compare its AI- and machine-learning-based predictions against the backtest results of its legacy Excel model.
Interface with treasury tools, enterprise resource planning (ERP) software and external data sources
The team began by developing an interface with its treasury software (Kyriba) to provide the model with streamlined, recurring access to raw treasury data. Without any external assistance from a software provider, the team developed two forecasting models.
The first is a business model, which replicates the firm’s previous Excel-based approach. This model relies on predefined business rules, such as annual pricing provided by central purchasing bodies and estimates for more volatile elements like how oil prices affect fuel sales.
The second is a predictive AI model that incorporates a much larger volume of external data. It includes variables such as oil prices and basic commodity prices, enabling the system to refine its forecasts through machine learning.
“The project’s success required several important elements,” explains the team. “These included efficient daily cash pooling, cash flow modeling with a focus on the income statement view (excluding inventory) and a seamless gateway between the accounting ERP and the treasury management systems”.
To ensure the model’s accuracy and reliability, the team monitors it daily by comparing its forecasts with actual cash flows from the previous day. This involves analyzing cash flow patterns and how data is sequenced, particularly payment timings by flow type. They also verify the stability and granularity of data allocation rules, with a minimum of five years of historical data required to support the model.
10,000 input lines processed every day
After a successful proof of concept in 2019, Groupement Les Mousquetaires officially launched the CFM project during the 2020–21 lockdown period. Its new cash flow forecasting model became operational in 2022.
“The COVID crisis was an exceptional event that we excluded from the learning phase,” notes Sébastien Schweickert. “However, it is still reflected in the evolution of behavioral data as consumer habits have changed profoundly in response to societal shifts like remote working.”
In the current model, COVID-related data can be re-entered by the treasurer at any time, treated as a replicable event. The system processes 10,000 input lines every day, and the team is now focusing on incorporating new explanatory variables into its predictive models. These include factors such as the impact of the energy crisis following the outbreak of war in Ukraine and the acceleration of inflation.
“We caution against making hasty changes to the data tags as this could cause the model to lose its ability to learn,” warns Schweickert. “Once the rules for classifying different flows are established, they must remain immutable and protected. This is especially important for the classification of operating costs and the analytical breakdown of food supplier costs.”
He also emphasizes that artificial intelligence is not infallible. “There are times when discrepancies persist for several months before being identified and corrected by humans.”
Greater accuracy and compelling financial benefits
The predictive model has evolved alongside the group’s business practices, particularly with the introduction of new rules on rebates. Initially confined to the group’s operations, the scope of the model was expanded at the beginning of 2024 to include entities affiliated to Les Mousquetaires.
What have the results been? The model has helped reduce gross debt by providing more accurate forecasts. Specifically, it has lowered the margin of error in cash flow forecasts to a maximum variance of €100–150 million – less than 1% of sales, which amount to €15 billion – compared with 3% previously.
“However, ROI isn’t just about financial savings,” notes Laurent Bonhomme. “Since our cash flow is cyclical, with significant downturns between January 15 and June 15, followed by a rebuilding phase from June 15 to January 15, it’s both interesting and reassuring to be able to accurately predict these low points. This enables us to notify our creditors in advance.”