“The way we start looking at every unit of work in a workflow flips from being something that's truly human-led to something that can be increasingly human-reviewed and driven, but not necessarily initiated as a human." - Nick Heinzmann, Head of Research at Zip
Procurement teams continue to face shrinking resources yet expanding responsibilities, a challenge that demands smarter, resource-efficient ways of working.
This dynamic is fueling the growing interest in Agentic AI: technology capable of analyzing information, taking adaptive actions, and significantly enhancing procurement productivity.
Nick Heinzmann, Head of Research at Zip, joined me to demystify Agentic AI and unpack its transformative potential for procurement workflows.
Here, in Nick's own words, are some stand-out moments from our conversation:
Agentic AI Defined
"We want to be able to show a degree of what we'd call agency. That's where the term comes from, right? Showing that instead of just having concrete and repeatable tasks being done, there's something more adaptive and intelligent behind it.
“So really when AI becomes Agentic, it's about injecting increasing levels of agency and decision-making. Maybe not necessarily full autonomy, but more awareness of what's being done instead of just doing."
Procurement teams leveraging Agentic AI can expect smarter workflows that allow AI tools to respond dynamically, taking defined actions and freeing human teams from repetitive, mundane tasks.
Moving Beyond Simple Automation
"In the last year alone, we generated 4.6 million AI insights with tools we've already developed. But what this changes now is we're moving from insight and maybe some more concentrated, specific actions to a world where the actions that AI is taking are much more complicated. The insights they're delivering are much more impactful to the level of not just accelerating and helping people make more informed decisions by collecting and surfacing more information than ever before, but also helping people take action."
Instead of merely automating simple actions, Agentic AI is elevating the scope and complexity of tasks that procurement can confidently delegate to AI, enhancing their efficiency in the process.
Applying AI to Real Procurement Pain Points
"A universal theme we see is a desire to work on certain kinds of tasks or activities that always get pushed aside, and everyone has this in their daily jobs (the stuff that's important but not necessarily urgent). In procurement especially, given the conditions they face these days, people are forced to work on things that are both urgent and important. But the stuff that doesn't get done, or even the stuff that could get done but doesn't get done, could be really valuable.
“The excitement I hear from people about this is that twofold element: one of getting freed from either monotonous or very painful work that prevents you from doing the work you want to do, but then also the aspect of reimagining what could be done."
Procurement teams leveraging Agentic AI now have the potential to explore value-generating opportunities previously left untouched due to resource constraints.
Strategic Starting Points for Agentic Adoption
"People who work with LLMs a lot are finding that it's great with text, it's great with summaries, it's great with synthesis. In limited scenarios where the right data is being applied, you can potentially use it for some calculations, or you can use it to determine, 'Oh, I need a tool for this. I should be using a Python script, or I should be using a calculator?' So when you're looking at use cases and tasks to automate, it's a question of: Is there a lot of legwork that is being done here that involves specific skills that we know LLMs are really good at?"
For CPOs wondering where to start, Nick’s practical advice is clear: begin with use cases ripe for intelligent automation, building interest and understanding incrementally throughout the enterprise.
Agentic AI and Data Quality
"Just because your data isn't good doesn't mean that you couldn't also find a way to get the data somewhere else where it's already good or higher quality or enriched, and then through orchestration you should be able to bring that into an agent's task or workflow to be able to do that more effectively. So ultimately, it's of course a concern, but you do not want to go into learning about, piloting, and ultimately implementing AI agents with this knee-jerk reaction that the data quality and volume have to be there, because even in some cases it doesn't necessarily need to be."
Data still presents hurdles, but Agentic AI significantly reduces its impact by extracting insights from external validated sources and automating robust data-enrichment processes.
