Many CPOs question the transformative impact that generative AI (gen AI) can deliver. The arguments range from overly niche functionality to concerns about the accuracy and security of using the large language models (LLMs) that underpin it. More often, though, we hear skepticism about the start-up cost and initial duplication of existing work. Sixty-six percent of CPOs surveyed in McKinsey’s CPO 100 survey in H2, 2023, believe gen AI is still years from generating substantive business results.
What we have observed, however, not only changed our minds but also those of some of the most skeptical procurement leaders. Gen AI has already started to drive measurable impact in four ways: 1) generating content, 2) enabling synthesis, 3) augmenting engagement, and 4) accelerating software programming.
Content generation
A key procurement use case is automating document creation to reduce manual work. Many CPOs believe that this type of content generation is not yet mature. “Gen AI is a great toy. It’s good to play with, but our day-to-day work remains unchanged,” a senior procurement executive of a Fortune 100 medtech company told us.
That daily work could be changed. LLMs can be trained on specialized data to generate highly tailored content. Some of the more common examples include custom-designed requests for proposals (RFPs) and multiparty contracts.
One McKinsey client team recently developed an RFP engine, leveraging sanitized templates and cost drivers from more than 10,000 RFPs and their responses.1 The technology replicated complex “best of best” analyses in a fraction of the time. It also learned what drove winning bids and redesigned future RFPs for optimal bid structure and cost granularity. Finally, it predicted, and prevented, omissions and mistakes in the bids.
A second use case was applied to contractual terms and conditions, one of the most mature applications of gen AI. LLMs trained on a database of contracts jump-started the first draft and created custom clauses that reflected a specific supplier agreement. We have already seen multiple contract life cycle management solution providers in the market. They have cracked the code on practical, effective solutions and offer off-the-shelf tools that require little customization.
Information synthesis
The magic of gen AI lies in its ability to retrieve, summarize, and extract insights from multiple unstructured data sources simultaneously. It can solve unstructured, end-to-end problems better than any prior technology, but is often perceived as a next-generation search engine that requires a savvy user to pose the right questions. This perception underestimates the power of gen AI for procurement, as it can be quickly deployed in two very common activities: category strategy and supplier sourcing.
A good category strategy reflects insights and trends from the external market and internal priorities. It includes supplier performance attributes, demand specifications, and risk management. A multilayered gen AI tool can generate a robust end-to-end category strategy in far less time than was previously possible, coordinating input from many sources and contributors. Over the past year, we saw gen AI models formulate strong strategies in categories ranging from highly specialized precision-machined parts to more mundane temporary labor. The resulting strategies were impressive and delivered with accuracy, security, and fairness.
The traditional route to supplier identification—searching through industry publications and database queries—is long gone. Gen AI combines keyword and capability-based searches with highly specific prompts to deliver superior results. An LLM can ingest publicly available information and cross-reference it with seed companies (incumbent suppliers with known capabilities that serve as exemplars for the model). The model scans broadly for similar patterns and identifies clusters of potential suppliers for further evaluation.
For example, a prompt of “suppliers for high-pressure injection molding based in Southeast Asia that are ISO 9002 certified” yields three times the results of traditional search engines. This approach improves success by scouring substantially more territory, augmenting available data, and filtering the data through proven supplier scoring criteria.
Interactive engagement
Interactive engagement is another frontier for gen AI–enabled procurement. Chatbots commonly guide category managers and business partners through their journeys. “Great procurement is based on relationships—you cannot replace that with a chatbot,” more than one CPO has told us. In practice, gen AI enhances internal and external relationships by eliminating friction due to different time zones, busy schedules, and overwhelming data. Instead, it aids conversations with talking points that better reflect the counterparty’s priorities.
Examples include negotiator pilots that craft scripts based on multiple scenarios. Gen AI can play different roles in mock negotiations to pressure test a strategy by examining arguments and counterarguments. The latest models are experimenting with iterative outcome evaluation based on different supplier responses—much like chess. The output is a recommended negotiation approach—confrontational, collaborative, fact based, or leverage based—that is most likely to succeed.
Software programming acceleration
Lastly, we have seen applications started in the IT department flow over to procurement. Most modern sourcing data cubes are built from a myriad of sources, starting with the enterprise resource planning and financial systems. For companies that have grown through mergers or have discrete business units, these systems are often not synchronized and require substantial manual work-arounds. Today’s gen AI models can take script code from one system and bridge it into others, greatly accelerating digital sourcing transformations and tool adoption. For some CPOs, this has taken years off their technology road maps and shifted the ROI in favor of greater automation and analytic horsepower. It is a terrific illustration of gen AI and traditional AI working together to create leapfrog progression in procurement capabilities.
From skepticism to certainty
To a skeptical, untrained eye, gen AI applications in procurement may appear niche or gimmicky. In reality, LLMs are trained across multiple knowledge domains, which allows them to handle broad, layered questions and draw implications that lead to good conclusions. It is one of the key capabilities that procurement organizations need to evolve and strengthen, moving from a pure executing function to a strategic, business-leading function.
1 Based on comparison of supplier discovery for direct materials suppliers in a traditional method versus AI-driven scouting across sectors (industrials, consumer, and life sciences).