The function of an ERP system is fast-changing to being a system of insights and action from a system of record, to meet the changing needs of businesses.
So, how does ERP software leverage generative artificial intelligence (GenAI) and large language models (LLMs) to achieve those actionable insights?
Kerrie Jordan, Group Vice President of Product Management for Epicor, leading ERP vendor for the manufacturing, distribution, retail, and auto industries, says, “At Epicor, we have a long history of investing in and launching a variety of AI solutions for the specific sectors we serve in the manufacturing and distribution, building supply, retail, and automotive aftermarket.”
Due to the unique needs of these customers, the AI offerings require a deep dive into the sub-vertical within each industry Epicor serves. For example, Epicor offers solutions such as Material Requirements Planning (MRP), an expert system for discrete manufacturers that learns from past production runs to make better predictions each time. AI-enabled solutions also include modeling and forecasting tools as part of Epicor Financial Planning and Analysis (FP&A), Inventory Planning and Optimization (IP&O), and Accounts Payable (AP) Automation.
However, Jordan and Arturo Buzzalino, Epicor’s Chief Innovation Officer, believe that the way forward lies with solutions that use GenAI and LLMs, like the company’s most recent AI innovation, Epicor Prism.
The process of AI adoption
When businesses decide to adopt AI into their processes, Epicor recommends that they go back to their basics to understand their return on investment (ROI) of adopting AI. According to Jordan, businesses must also understand how an AI solution aligns with their corporate objectives and vision. “We have been working with many customers to understand their business challenges and what they want to achieve by implementing AI as it is a complex solution.”
Specific issues always get the top priority in these conversations. “Our engineering team works one-on-one with customers to figure out their current business practices, if their data is in order and the software they use, as ERP is at the core of many of these businesses and needs to be implemented correctly,” says Buzzalino. He notes that knowing who accesses and changes data for accuracy is also essential. “Additional value can come in once the right data is in place,” he believes.
Buzzalino explains that processes that might take up a large amount of human time, like inventory projection, are ripe for AI disruption using LLMs. Implementing these AI models in manufacturing reaps some interesting results.
“Productivity improves, and certainly there are cost savings through inventory optimization and risk management,” Jordan says. This is because things that were previously done manually are now being managed by AI models, leaving employees free to focus on other, more strategic jobs.
Buzzalino gives the example of AI models taking over a job like order entry. “This job, done manually previously, can now be completed within seconds by LLMs using the data from the ERP, taking the human effort out of the loop in the process. The employee now only needs to ensure that what is going out is accurate, and that saves a ton of time and leads to significant savings,” he says.
Jordan agrees, saying: “Sometimes [the manual work] is arduous and repetitive and can be easily automated with AI. Not only does this result in great savings, but it truly benefits an employee’s well-being, too.”
Advancing AI in ERP
Even though organizations are excited about AI’s possibilities, there’s also some hesitation, according to Buzzalino. This is mostly because they don’t necessarily trust all their data and, as a result, do not want to make business decisions based on AI recommendations that use that information.
To help users overcome this hesitancy, Epicor invested in its Epicor Grow solution in 2022, which allows customers to connect, exact, and visualize their data. A cloud-based, full-stack Business Intelligence (BI) solution, Epicor Grow uses low-code/no-code to make it intuitive.
Epicor has now built on its offering to include an interactive data pipeline experience that the company calls the Epicor Grow Data Platform. “The solution is much larger now and is highly tuned to manage the types of data from a velocity and variety perspective to ensure speed and a seamless experience,” Jordan says, adding that in 2025, the platform will be further enhanced to make AI easily accessible to Epicor customers.
The latest enhancement will be available through Epicor Prism, which, according to Buzzalino, promises to “revolutionize the way humans interact with their software, particularly ERP.”
“We are about to officially launch Epicor Prism, which is essentially an inference pipeline for LLMs,” says Buzzalino. In this solution, the company is taking some of the world’s largest LLMs, like ChatGPT, and customizing them to leverage these AI models to solve specific problems. “We don’t want to solve generic issues like making summaries of long texts but would rather solve concrete steps in a workflow or manufacturing process,” he reasons. “With Epicor Prism, users can converse with their ERP system for insights and to determine processes or decisions through the Retrieval Augmented Generation (RAG) strategy we have developed to help LLMs learn from the data provided by the ERP.”
A use case for inventory optimization
According to Jordan, inventory optimization is another area in which Epicor has made great strides with AI. “Our inventory planning and optimization solution runs thousands of simulations across the supply chain to determine the best stocking level,” she says.
This model uses three patent systems to achieve this. The first patent system clusters and visualizes demand profiles of resources by leveraging machine learning to enhance supply and demand planning. The second one simulates demand and optimizes control parameters, allowing probability modeling to generate a future daily probability distribution for random lead times and account for multiple trends and seasonality levels. This can allow businesses within the make-move-and-sell economy to optimize control parameters for types of seasonality.
The third patent system is an automated system control with data analytics using a doubly stochastic model that exposes hidden drivers within the supply chain or inventory. It then forecasts those drivers to help organizations model their what-if scenarios to choose the best stock level.
“When we tie these into FP&A by leveraging industry-leading machine learning models at the back-end to forecast the budgeting and planning process, we have the full picture of the operations, which, along with the AI models, help businesses make the best decisions,” Jordan concludes – touching on what all enterprises want to do in the here and now with their ERP systems.
