Should you build or buy your next procurement AI agent?

A recap of the Pure Procurement x Pivot debate webinar

‍

Setting the Scene

If you work in procurement, you've probably noticed that AI agents are no longer a distant promise, they're showing up in vendor demos, budget conversations, and strategic roadmaps.

‍

But once you've decided that AI agents have a role in your operations, a second, harder question follows: do you build one yourself, or do you buy one off the shelf?

‍

That's exactly what Joël Collin-Demers and Romain Libeau tackled in their December 2024 debate webinar, hosted in partnership between The Pure Procurement Newsletter and Pivot.

‍

The Speakers

Joël Collin-Demers is the Consulting Principal and head editor at The Pure Procurement Newsletter. With over 13 years of experience implementing ProcureTech and supporting procurement functions, Joël brings a practitioner's perspective on what actually works (and what doesn't) when organizations try to get results from technology.

‍

Romain Libeau is the CEO and Co-Founder of Pivot, a global, full-suite, Source-to-Pay platform. As someone building procurement technology from the ground up, Romain brings a vendor-side lens on the complexity, trade-offs, and realities of deploying AI agents in enterprise environments.

‍

For the purposes of the debate, Joël argued the build side and Romain argued the buy side, while both acknowledged upfront that their positions were partly theatrical. The goal was to sharpen the arguments on both ends, not to represent their personal views in absolute terms.

‍

The Build vs. Buy Continuum

Before diving into the debates, the hosts were careful to establish one key point: build vs. buy is not a binary choice. It's a spectrum.

‍

• On one end sits the Full Custom Build with your own team, open-source LLMs, your own infrastructure, and full maintenance responsibility. 
• On the other end is the Pre-Packaged Agent, a vendor-led, out-of-the-box solution with the vendor choosing the underlying model and managing everything under the hood. 
• In between, you have options like building on a generic platform (think n8n or Microsoft Copilot Studio) or buying a procurement-specific platform like Pivot, where you get vendor infrastructure and support but retain meaningful configuration control.

‍

One thing both speakers agreed on regardless of where you land on that continuum: you cannot skip the fundamentals. Good financial hygiene, a solid understanding of your procurement processes, and clean data are prerequisites for success with AI agents, no matter how you acquire them.

‍

The Five Scenarios

The debate was structured around five real-world procurement scenarios, with the audience voting after each round on who made the more compelling case.

‍

Scenario 1: The Competitive Edge Question A services company with a highly specialized sourcing process wants to scale its advantage with AI. 

‍

• Joël (build) argued for maintaining full control over the agent's logic, policies, and workflows — and for protecting the IP that underpins a genuine competitive advantage.
• Romain (buy) countered that adoption speed matters more than custom code in a fast-moving AI landscape. 

‍

The audience sided with build, largely convinced by the IP ownership and control arguments.

‍

Scenario 2: The Control vs. Accountability Question A global multinational fails a compliance audit and turns to AI agents for supplier risk monitoring. 

‍

• Joël (build) argued that complex, regulated environments often require custom controls that vendors simply can't support, and that internal teams can encode compliance rules with precision. 
• Romain (buy) pointed to vendor SLAs, pre-certified frameworks, and the EU AI Act's audit trail requirements as reasons why buying is actually the faster path to compliance credibility. 

‍

This one was closer to a 50/50 split.

‍

Scenario 3: The Scalability Question A company wants to roll out an intake and triage agent across 40 countries. 

‍

• Romain (buy) made the case that global deployments involve far more than language localization as they require 24/7 monitoring, cross-ERP integrations, and a level of operational complexity that most internal teams aren't built to sustain. 
• Joël (build) argued that large enterprises often already have the infrastructure and engineering capacity to handle this, and that vendor pricing at that scale can be punishing.

‍

The audience still leaned towards build, though the buy argument landed well.

‍

Scenario 4: The Speed vs. Fit Question A fast-growing tech company needs AI agents to handle a surging backlog of purchase requests. A vendor offers 80% use case coverage in 8 weeks; the internal team promises 95%  in 6 months.

‍

This scenario raises a question worth sitting with: when operations are struggling today, is the 15% coverage gap worth a 6-month wait?

‍

That gap could be trivial, or it could be exactly where your most complex, high-value requests live.

‍

Scenario 5: The Talent and Knowledge Question A SaaS company built a custom spend analytics agent two years ago. Now the architect is leaving and a key developer is interviewing elsewhere. 

• Romain (buy) used this as proof that building creates dangerous key-person dependencies and unsustainable maintenance liabilities. 
• Joël (build) countered that the real failure was underinvestment in team depth and documentation rather than the build decision itself. 

‍

The audience remained pro-build, though the discussion around documentation and governance added meaningful nuance.

‍

The Verdict

Despite the audience voting for "build" in almost every round, the hosts' shared conclusion leaned the other way: for most organizations, buying is more appropriate than building — unless the AI agent capability is genuinely core to your competitive differentiation.

‍

The five reasons they cited:

• Risk protection over control. Buying sacrifices some autonomy, but modern platforms offer substantial configurability and deliver contractual accountability in return.
• Speed to value. Vendors have solved implementation challenges you're about to face for the first time.
• Predictable TCO. On the build side, you risk sporadic reinvestment to keep your stack functional as components evolve. Vendor costs are more foreseeable.
• Reduced key-person risk. Building concentrates critical knowledge in a small number of people. When they leave, you're exposed.
• Scalability. Vendors have already worked out how to operate across regions, languages, and compliance environments. Building means answering those questions yourself.

‍

The exception, both agreed, is when the agent capability truly differentiates you in the market. In that case, building may well be worth the investment and risk, but the advice was to be honest with yourself about whether that's really true.

‍

Final Thoughts

The framework for thinking through the classic build vs. buy question made the webinar, not the verdict itself.

‍

The right choice depends on context: how critical is control to your strategy? How quickly do you need to generate value? What's your realistic internal capacity? And are you willing to own the long-term maintenance burden?

‍

The build vs. buy question is ultimately a risk and resource allocation decision. The scenarios made clear that the same organization could reasonably land on different sides depending on the use case. An intake orchestration agent rolling out to 40 countries is a very different beast from a spend analytics agent serving a single procurement team.

‍

One closing thought from the conversation that's worth sitting with: Romain suggested that regardless of how an agent is acquired, companies should ultimately own it — understanding how it works, what data it touches, and how to evolve it over time. A black box, whether built internally through poor documentation or bought from a vendor without scrutiny, is a liability either way.

‍

This post is based on the December 2025 webinar "Procurement AI Agents: To Build Your Own or Buy One Off the Shelf?" hosted by Joël Collin-Demers (The Pure Procurement Newsletter) and Romain Libeau (Pivot).