Analyst Perspectives

There have been a multitude of potential use cases for artificial intelligence (AI) and generative AI (GenAI) dreamed up over the past 18 months. ISG-Ventana Research describes AI as the development of systems and software capable of automating tasks that have previously required human intelligence. It encompasses machine learning (ML), deep learning and GenAI to deliver capabilities including predictions, recommendations, personalization, speech and visual recognition as well as translation and summarization. Purchasing is ripe for the application of AI because there is a great deal of rote, repetitive work that can be automated or assisted by the technology, from sourcing to processing invoices to making payments. ISG-Ventana Research asserts that by 2027, almost all procure-to-pay software suites will use AI and optical character recognition (OCR) to automate data ingestion from external documents and emails, saving time and ensuring all necessary information is captured accurately at the source. For now, the snag is that while the practical use of AI technology has made remarkable advances over the past year, there’s still a lot more that needs to be developed and field-tested. And purchasing departments must prepare to make full use of AI as the technology matures.

AI will be a transformative technology for purchasing, as well as finance and accounting, but like all technology, it will not be a magic bullet that fixes an underperforming department. One important reason is that AI is nothing more than an adjunct to core business software capabilities. Only by methodically addressing any organizational and process issues that hamper the department’s performance can AI enable purchasing to achieve the full potential of technology. Another factor that must be included in any departmental AI initiative is ensuring that the data necessary for training the system is available on the software platform. Making purchasing departments AI-ready means addressing those business issues sufficiently before implementing software or, alternatively, addressing the people, process and data gaps as an explicit part of the project.

One common mistake that enterprises make when buying software is not starting from the top of a logic tree to identify root causes of the issues they are trying to fix. That is, they make a set of assumptions about “givens” that aren’t necessarily immutable, and that leads them to a seemingly reasonable but incorrect or suboptimal conclusion. In this case, an important first step is to understand and define what sort of purchasing department the business needs and identifying the organizational and process gaps that exist in meeting those needs. Admittedly, this should be obvious to those in the department, but that’s never a given, so it pays to make this explicit. Doing so will expose differences in the mission of the department and its priorities. Ensuring that everyone is on the same page from the start requires clear communication within the department and with everyone else on the specific needs of the business.

Making certain that the basic assumptions about how the department operates reflects the fact that different business models require different definitions of the department’s functions. For example, sourcing materiel and parts for manufacturing, especially for high-value products, is a more rigorous process than buying non-strategic items such as office supplies, so the software used in supporting the process should address the specific business requirements of the enterprise and the full range of people involved in vendor selection. Enterprises that have gone through multiple mergers, divestitures and reorganizations may no longer have the department structure and processes necessary to perform well.

Determining how to AI-enable the purchasing department should start with identifying where existing and near-term AI capabilities can improve performance in the core responsibilities. These departmental functions typically include:

• Sourcing: Defining the requirements for goods, materials and services, and finding and evaluating reliable suppliers to meet these requirements.
• Negotiating: Establishing pricing, terms and conditions with suppliers.
• Procurement: Ordering, tracking and scrutinizing the delivery, timeliness and quality of purchases.
• Management: Handling budgets, payments and inventories.
• Control: Setting and monitoring compliance with policies and regulations.
• Review: Periodically assessing suppliers' performance to inform future sourcing and purchases.

Each of these functions has the potential to use AI and GenAI to increase the productivity and effectiveness of the department. However, there probably are already multiple ways to use existing, conventional technology to achieve results that also lay the foundation for a higher return on investment when implementing AI systems. Existing procure-to-pay technology can:

• Improve operational efficiency.
• Increase visibility and spend control.
• Ensure policies are followed.
• Reduce process errors.

For example, having end-to-end process workflows in place helps ensure that procedures are followed and there are many fewer delays and dropped balls. Using virtual cards, especially for indirect spend, improves compliance, control and visibility because these cards can limit purchases to specific purposes and vendors and for limited amounts. Integrating a dedicated procure-to-pay platform with an enterprise’s ERP systems ensures accuracy and timeliness of information. And using OCR systems for ingesting documents like invoices can enhance accuracy while boosting productivity. None of these require AI; therefore, reviewing and, wherever necessary, restructuring roles, responsibilities and processes should be undertaken before deciding how best to use AI and GenAI in purchasing.

The quality of results from AI systems is heavily dependent on the accuracy, completeness, breadth and timeliness of the data used for training the systems. For all but the most trivial of AI-enabled tasks, the system must have all the necessary data resident on the application platform to support the ongoing training necessary for the software to learn and adapt as new information or changing conditions require. This means that departments must assess the quality and completeness of available data as a precursor to AI adoption, something that can be done immediately to ensure that closing gaps between data needs and data availability will at least begin to be addressed to achieve the best results from AI. As my colleague Matt Aslett has pointed out, the good news is that technology itself, including data operations, data management and data intelligence, have reduced the effort needed to address data-related issues.

In the past, at the start of major transitions in information technology (including client-server, the internet and cloud computing), we’ve been treated to a litany of this-will-fix-everything claims for that technology. Each time, the claims have come up short. The common denominator each time is that technology by itself will not address poor process management, data deficiencies and, especially, change management barriers. I strongly recommend that finance and operations leaders that are looking to use AI to improve the performance of the purchasing function in their organization first address fundamental issues before layering in AI.

Regards,

Robert Kugel