Every procurement team goes through a "ChatGPT phase." This is the moment where you realize you can get answers to almost every question instantly with generative AI.
But when you move from ad hoc queries to systematic workflows, from answering questions to executing procurement work, general-purpose LLMs like ChatGPT or Claude hit a wall.
At Suplari, we’ve been helping enterprise procurement teams move from AI pilots to full-scale implementation since 2017. The teams getting the most value from AI aren't the ones with the best prompts. They're the ones who understand the fundamental difference between AI that assists and AI that executes.
The ChatGPT phase is real - and necessary
A couple of weeks ago I hosted a roundtable with two procurement leaders using Suplari's AI Agent. Both have deeply experimented with ChatGPT, Copilot, and other off-the-shelf LLMs. What made this conversation interesting wasn't that they chose Suplari. It's that they stress-tested off-the-shelf approaches against it. Side by side. To see which actually delivered.
The journey they described is happening in procurement organizations everywhere.
According to research by Wharton Human-AI Research, weekly use of generative AI within procurement increased 44 percentage points from 2023 to 2024. Today, 94 percent of procurement executives use generative AI at least once a week. That's the highest adoption rate of any business function.
The reasons are clear. The Hackett Group's 2025 Key Issues Study puts numbers to the pressure: procurement workloads are projected to increase by 10% in 2025, while budgets grow just 1% - creating a 9% efficiency gap that has to close somehow. With layoffs continuing across the enterprise, it won't be additional headcount. Procurement teams need to get used to doing more with less.
So teams turn to ChatGPT. They start with simple questions. Get good answers. Build confidence. Then they try to scale, moving from ad hoc queries to building systematic workflows.
That's when the limitations of off-the-shelf AI solutions become apparent.
Key limitations of ChatGPT and general-purpose AI in Procurement
One of our roundtable participants spent two months benchmarking Suplari's agent against off-the-shelf tools before rolling it out. Ran the same procurement tasks through ChatGPT and Copilot. Compared outputs. Catalogued patterns.
Her conclusion: "For advanced analytics of our spend data, ChatGPT and Copilot can't compare."
The moment you move into actual spend analysis, contract-to-PO reconciliation, or savings calculations, general-purpose LLMs break. Results vary between queries. Numbers don't reconcile. Memory disappears between sessions. Every analysis requires re-explaining your supplier taxonomy, your contract structures, your organizational context - all while watching for inevitable drift.
These aren't edge cases. They're fundamental architectural limitations of off-the-shelf generative AI tools.
- ChatGPT cannot access your proprietary data securely. Without your data, it provides generic advice that may not apply to your situation. It cannot identify real cost savings opportunities or analyze your actual procurement performance.
- ChatGPT suffers from hallucinations. It generates confident-sounding but incorrect information. This creates serious risks for financial calculations and analysis where accuracy is non-negotiable.
- ChatGPT's training data may be outdated. It often misses current market conditions and pricing. And entering confidential supplier information, contract details, or strategic plans into a public AI model creates unacceptable security risks that eliminate most high-value procurement use cases.
As another roundtable participant put it: "Copilot does a better job with artwork than Claude does. But if you're trying to do a really in-depth contract negotiation, you want to go with the Suplari Agent."
Different tools. Different strengths. But none of the off-the-shelf options could consistently handle procurement's full complexity.
The gap between adoption and transformation
The research confirms what we're seeing in practice. According to the 2025 EY Global CPO Survey, 80 percent of global CPOs plan to deploy generative AI in some capacity over the next three years. But currently, only 36 percent of procurement organizations have meaningful generative AI implementations.
That gap - between individual experimentation and organizational transformation - is where most procurement teams are stuck.
MIT's 2025 State of AI in Business study delivers the stark reality: despite $30-40 billion in recent investments in generative AI, 95 percent of enterprise pilots deliver no measurable ROI. Over 80 percent of enterprise firms pilot generative AI, but only 5 percent have reached mature production-stage deployment.
One reason: the emergence of "shadow AI." According to MIT, 90 percent of employees use personal AI tools at work, while only 40 percent of firms have official subscriptions. Individual productivity gains don't translate to organizational transformation.
Another reason: most general-purpose tools don't learn or adapt. Every conversation starts fresh. The AI never accumulates understanding of your organization's suppliers, contracts, categories, or decision patterns. This creates a verification burden that often exceeds the time saved - and explains why so many pilots stall before reaching production.
What purpose-built AI delivers differently
The fundamental difference between ChatGPT and vertical procurement AI comes down to one word: autonomy.
While ChatGPT assists with tasks, purpose-built solutions actually execute procurement workflows. They operate with direct connection to your spend data, contracts, and purchase orders - enabling real-time analysis of supplier performance and market conditions.
Unlike ChatGPT, which starts fresh with every conversation, purpose-built procurement AI maintains long-term organizational context.. It builds institutional knowledge that improves over time. Every decision, outcome, and market change improves the agent's decision-making capability and strategic alignment. The AI becomes more effective as it learns your organization's preferences and successful patterns.
Purpose-built solutions handle sensitive procurement and financial data securely. They provide audit trails and governance controls you won't find in generalist AI tools. They integrate with your existing security infrastructure while meeting industry regulations.
This security foundation enables the real data analysis that drives meaningful procurement improvements - the kind you simply cannot do with a public AI model.
The question that changes everything
The Hackett Group's fresh research shows that 64 percent of procurement leaders expect AI to transform their roles within five years. Early adopters are already seeing productivity and cost improvements of up to 25 percent. The opportunity is real - but capturing it requires moving beyond the experimentation phase.
If your team is in the ChatGPT phase right now, you're not behind. You're learning. That experimentation builds essential understanding of what AI can and cannot do.
But when the limitations start slowing you down more than the capabilities speed you up - when you're spending more time working around constraints than getting value from outputs - it's time to ask a different question.
Not "how do I work around these limitations?"
But, as one procurement leader put it to his team: "If I could do this, what would that enable me to accomplish?"
Off-the-shelf tools provide helpful conversation. Purpose-built agents drive measurable outcomes. The practitioners who stress-tested both approaches already know the difference.
Ready to move beyond the ChatGPT phase? See what purpose-built procurement AI delivers at suplari.com
