Procurement AI: Yawning Gap between Hype and Reality

Everyone is talking about AI in Procurement, apparently even more than the LLM model developers. I do agree that the Procurement AI industry stands at a critical inflection point, but it's not because it is at a breakthrough stage. It is quite the contrary.

With the circular validation ecosystem, the Digital Procurement industry has created an echo chamber that barely resonates with actual business results within its organizations. The data backs it up, and I will share many such datapoints in this article.

Fancy events labeled as "community events" (which organizations should classify as personal expenses due to reckless spending), along with exponential ROI claims, inflated savings claims from every tool provider, pay-to-play awards from conference organizers, and motivational quotes featuring large individual portraits under the guise of Procurement content, are not helping Procurement gain any credibility. So, let's start.

While vendors promise autonomous procurement, 80% of companies have not seen any material contribution. With an average of 80% of GenAI projects being abandoned and a systematic failure to predict any major crisis (such as semiconductor shortages, trade tariffs, and shipping congestion), even Gartner moved GenAI for procurement from "Peak of inflated expectations" to "Trough of Disillusionment" in just one year! This has to be one of the fastest hype cycle collapses in history!

Chapter 1: Procurement AI claims far, far, far exceed the actual business value/impact (pardon my stress here!)

In fact, these AI claims from all the vendors (including the legacy vendors) have saturated the market. SAP Ariba markets "autonomous agents" executing workflows with "production-grade agent networks". Coupa promises autonomous procurement decision-making with minimal human intervention. They even said this will multiply your margins. Oracle claims its AI delivers 80% faster procurement cycles. GEP and Zycus both promise 10x speed and efficiency. Ivalua promises nearly 400% ROI. Like, how do you even measure any truth behind these claims? Can you attach their payments successfully, like they advertise? The answer is a clear No.

Against those very tall and bold claims by Procurement Tech providers, the reality tells a different story. McKinsey reports that 80% of companies see no material contribution to earnings from their GenAI investment. The average GenAI investment was $1.9 million (in 2024). Gartner predicts that 30% of GenAI projects will be abandoned after proof of concept by the end of 2025, and 40% of agentic AI projects will be cancelled by 2027. The reason? Unclear Business value and escalating costs. One reason stood out for me very clearly in this Gartner assessment as they stated that "current models dont have the maturity to achieve complex business goals autonomously".

Having spent 10 years in capex and raw material procurement and now 8 years in the Procurement AI domain, I wholeheartedly concur with this assessment. What you see as shiny AI solutions are the lowest-grade problems that hardly move any needle in Procurement operations. I'd classify this as a low-quality, low-importance GenAI segment, and this is where most of the CPOs are being lured into.

The ROI disconnect becomes clear when comparing vendor marketing to actual business outcomes. Vendors claim 2-4x returns, 80% faster cycle times, and 400% ROI. Meanwhile, CPOs internally would accept less than 50% ROI as success (a significantly low bar that itself signals minimal confidence). This is validated in the research done by Forrestor. Only 26% of companies have developed capabilities to move beyond proof of concepts, according to McKinsey and BCG, leaving 76% of companies stuck in "pilot purgatory," as McKinsey calls it.

I also liked the BCG report because they said only 10% of value derives from AI algorithms themselves, and 70% requires changes in people and processes. This fundamentally contradicts the AI revolution narrative. Yet, all the vendors continue to promote the dishonesty. We haven't even arrived at the most damaging claim. The actual adoption rates lag dramatically behind the hype. Only 23% of organizations use AI for Procurement functions, according to SAP's own survey - the second lowest adoption rate among all business functions! Deloitte's study validates what we have been discussing so far: 92% of CPOs plan to assess GenAI capabilities, only 37% are actually piloting, and a mere 4% have reached any large-scale deployment stage. 

The numbers tell us the actual story of Hype versus real business value benefits.

Chapter 2: Procurement AI does not tell anything new; it is rather a validation exercise today

Today's GenAI capabilities in Procurement tools are largely API wrappers built on your data and the OpenAI Api in the backend. Procurement Tech platform providers dont own anything new, and they are not investing any time in analyzing the data for you. They want to cash in on the gold rush of "Procurement AI" by luring CPOs to mega party conferences and selling AI pipedreams. 

The current AI capabilities are predominantly what technology experts call "validation exercises". They are just confirming what procurement teams already know rather than generating any new insight. This pattern appears across all major use cases that digital procurement vendors are promoting. Let's speak about some popular use cases

1.) Spend Classification: 

If you have been following my work that I do at Supernegotiate , you will know how passionate I am about the Spend analysis industry. Spend classification represents the most common application for AI, where one can expect AI to classify your product description and invoice descriptions into a proper taxonomy (hopefully, you are not choosing UNSPSC, as it is not the right choice for taxonomy). 

Promise: AI categories spend and create a "spend cube" and claim "97% to 98% accuracy". 

Actual: Every category manager and CPO knows that, despite the AI claims, their teams are required to validate each classification, and trust me, that is an extensive and time-consuming exercise. All the AI claims fall flat when it comes to classification accuracy. It is one thing to claim AI-based classification, but it means nothing if the classification is wrong. Most of the vendors still rely on rule-based engines to support their accuracy claims. This isn't anything new, as vendors claimed 95%-98% accuracy rates for rule-based classification systems built 15 years ago!

Now, you tell me, is this progress?

2.) Supplier Risk Scoring: 

This area also follows similar dynamics. 

Promise: AI assigns risk scores based on financial data and news, flagging suppliers with potential issues.

Actual: Procurement teams already monitor strategic suppliers closely through traditional scorecards using some weighted KPIs and financial ratios. The automation (or AI, save me from the warlords of Procurement AI platforms here for the misrepresentation of technology, lol!), so, again, the automation merely surfaces what category managers already track. AI merely creates dashboards that display the already known information rather than discovering any fresh risk by studying patterns. 

When the actual crises hit, like in the last 2-3 years, all of these AI-driven monitoring systems have failed to provide any early warning.

3.) Contract Analysis:

Promise: AI can extract key terms and pricing, and then perform supplier performance management. 

Actual: While natural language processing has improved the accuracy over older OCR systems, the critical contract terms are already documented in procurement systems and tracked by category managers/contract managers. AI extracts what's already known as important, looks for manual validation, and then merely creates a searchable repository of key terms. The core workflow - scan, extract, identify, flag, and alert remains identical to systems from the 2010s, just with better handling of non-standard formats.

This fundamental problem has been identified in multiple industry research papers as well. The tool solves problems that researchers and vendors think exist, rather than the problems category managers actually face. When they test these tools, they fail to discover any new value or insight, leading to non-adoption. The funny thing is, AI tools require high-quality data to function, but at the same time, these tools are being marketed as solutions to data-related problems! Talk about circularity!

Chapter 3: Every crisis exposes Procurement's predictive capability failure

The recent few years have truly exposed Procurement's GenAI claims, often leaving them with less credibility than before. All AI tools have failed spectacularly in predicting any major event. Let's look into some examples!

1.) Supplier bankruptcies: 

In 2024, close to 700 companies in the US filed for bankruptcy - an 18x increase compared to 2022. Silicon Valley Bank collapsed despite public data showing concentration risk and rumours of liquidity issues days before failure. A few more notable failures - Signature Bank, China's Evergrande, Sunac China holdings - all these were colossal bankruptcies that the IMF warned about in 2024! Despite that, I didn't see any AI procurement tool notifying the category managers about these collapses by sensing these entities in their suppliers' value chain.

2.) Supply chain disruptions: 

Probably, you were expecting this point already (See, you dont need AI here!). Supply chain disruptions increased 38% in 2024 despite all companies claiming AI-driven predictive analytics. Nike's overinventory problem was widely discussed! I also wrote an article on it (of course, post facto).

The Red Sea crisis is forcing the rerouting of commercial vessels, catching most of the companies off guard. Tesla halted Berlin plant production for 2 weeks due to part shortages, IKEA warned of product shortages, and shipping rates spiked 140% on China-US routes and 500% on China-Europe routes!

3.) Procurement Fraud: 

This gets me! Procurement fraud is on the rise as AI claims in Procurement are on the rise! The DOJ recovered $2.7 billion in False Claims Act settlements with 40 procurement fraud cases brought by whistleblowers. Procurement fraud is still detected and reported by humans and not AI. (Here comes a shameless plug - Supernegotiate has a rule-based engine to detect Procurement Fraud patterns - Say hi@supernegotiate.com to see it in action).

The most damaging insight - according to the Trustpair report, 96% of US companies experienced at least one fraud attempt in 2023 (up from 56% in 2022). These fraud cases were occurring precisely when AI fraud detection tools were being marketed and deployed. Even Coupa's own Chief Procurement Officer admitted, "We've seen companies defrauded for tens of millions of dollars". Quite ironic when Coupa claims AI-driven engines that detect and prevent Procurement fraud!

4.) Geopolitical Risk: 

Name one tool that predicted your supply chain disruption risk due to the Russia-Ukraine war before it happened! China's export controls on gallium, Trump's tariff war (as recent as the 10-Oct announcement), risk of looming semiconductor shortage round 2! What a remarkable, consistent failure rate of AI in predicting and modelling these risks. Maybe that's why even Deloitte reports that most of the CPOs (in fact 70% CPOs) see a rapid increase in Procurement risks despite having a state-of-the-art AI system to warn them proactively!

Chapter 4: Thought Leadership Becoming Marketing Theater

There are only a handful of procurement practitioners who post practical content that can be implemented. 90% of the procurement content creators have succumbed to the LinkedIn algorithm - often posting naive and low-effort content that is not even related to procurement anymore. 

Thought leadership has become spam, and ChatGPT has made it too easy to be a thought leader! Thought leadership is determined by the number of likes and shares (performance-driven). Therefore, people are likely to post "agreeable content" rather than an independent opinion! Analyze any post and all you can comment on them is "100%", "Spot On", "Couldn't agree more"! 

High-quality content creation shares common factors: technical depth in Procurement content, specificity, honesty about failures and limitations, and, most importantly, independence from any commercial interest. (No hidden course or become a paid member of newsletter pitches). But not all is bad here, there are some really fantastic Procurement creators that I really enjoy, and I wish to see more content from them!

Without going into much detail, the following are some of the recent trends in Procurement content generation

1.) Generic AI content: 

High-level, big, and bold claims without any specific detail appear to be the trend! IBM claims "AI technology can bring operational insights into data that was not otherwise visible" without specifics. Oracle states "AI can help cut cycle times by 80%" citing a single KPMG study. The pattern is consistent - Generic transformation turnaround claims, big numbers without disclosing methodology, speculations presented as facts, and usually, such bold claims are based on one vendor's "study". 

2.) Echo chamber effects: 

This is the funniest. Every vendor uses identical language - "AI will transform/revolutionize procurement", "automate routine tasks", "400% ROI", "80% Cycle time reduction", "cost savings", "free up strategic time". The same claims are marketed on LinkedIn, at conferences, and even by presenters in their keynotes! Where are the case studies about lessons learnt, specific failure points? 

3.) The economics of Procurement conferences: 

I'll be straight to the point here! 90% of procurement conferences are mega sales events. CPOs are invited to be keynote speakers, and usually these are crowd pullers! These numbers are then sold to vendors to demand outrageous booth charges that fund the after parties! Startups are used to signify "innovation," but usually their pitches are crammed into 180-second elevator pitches! All the sponsors and main event coordinators are the legacy vendors! Where are the problem-solving aspects and tangible case study presentations? As one sales consultant aptly put it, "Every single person who attends your session is a high-quality lead". Conferences are designed as prospecting opportunities, and this model has been in place for many years!

4.) Pay-to-play awards:

These days, the Procurement awards and recognition industry has become a booming business. Every organizer charges a participation fee (upwards of USD 200-300 per application). Some categories are even vendor-sponsored (like SAP, Amazon, etc). Often, the selection process is non-transparent, and you will often notice sponsors are sometimes the participants in categories, too. No surprises, who will be the winner!

In this kind of setup, the financial relationship determines visibility and "award". The element of meritocracy is often sidelined!

Chapter 5: Reality Contradicts Vendor Narratives

Procurement professionals have been giving feedback to vendors, in private! But the vendors are not listening to or working on them! So, let me document some of the real procurement challenges.

1.) Data quality issues

Data quality issues dominate as the leading barrier to any meaningful implementation in Procurement. McKinsey found 51% of CPOs report their data quality as poor or at best average. Hence, only 20% of Procurement data is actually used! As one CPO aptly told MIT researchers, "The AI gives me an answer in seconds. Then I spend ten minutes verifying if it's right. How is that saving time?" MIT termed this "verification tax" - AI that's 80% accurate requires checking for 100% of outputs, eliminating efficiency gains. 

 2.) Pilot purgatory

Here is one fact - MIT's 2025 report found 95% of GenAI pilots deliver no measurable ROI, while IDC research shows only 4 out of 33 POC's graduate to production. McKinsey describes this as "Pilot purgatory," where teams prove effectiveness in test but cannot replicate or implement their use in core processes. 

3.) Integration complexity

This one is always kept hidden from CPOs. The integration complexity always exceeds estimated timelines, always! Legacy ERP platforms were never built to support modern AI tools. They need middleware, APIs, and customization. This takes time. Art of Procurement warns, "If you have technological debt caused by siloed procurement systems, the challenge is only going to get worse with AI. AI systems can not solve your integration challenges". The OCR implementation failure documented by Supply Chain Dive confirms this: an Oil & Gas company deployed OCR across accounts payable without standardized templates, proper pre-processing, and training, resulting in hiring more headcounts to manage exceptions and an eight-week payment backlog. 

Chapter 6: What CPOs should do differently

Here are some of my recommendations (you dont have to take these seriously!)

1.) Demand Independent Validation:

Customer verification (directly and not through vendors) is the key! Documented pilot to production timelines, including disclosed failure rates, are good metrics to request. I'd also rejected any vendor-sponsored study (like Forrestor TEI studies paid for by vendors). 

2.) 70-20-10 rule:

As per BCG, 70% of the value comes from people and process changes, 20% from technology infrastructure, and 10% from AI algorithm (In the current setup). If you are deploying AI solutions without changing processes, then the initiative is highly likely to fail.

3.) Build your Data Use Case journey:

This one is my favourite. I have been a strong, strong, strong advocate of the use case-driven Procurement Transformation journey. Last year, I published this, and I still think this template is highly relevant.

Each use case should then be mapped to business impact and complexity. I use high-low binary classification.

So, you may start with low-complexity, high-impact business use cases and slowly move towards high-complexity, high-impact use cases. Once the use cases are identified, you should move towards the data source mapping journey. This data source mapping journey will immediately reveal the gaps in your data infrastructure. You will then have to factor in integration efforts! Sometimes, you can change the process to eliminate the need for data integration altogether. Once a new process is agreed upon, only then should you begin the AI POC experiment.

I have repeated this template over and over again, and it has proven to be my secret hack in ensuring high success rates in all of my AI experiments/POCs in procurement.

Well, I hope you found this critical note on Digital Procurement a bit contrary to your conventional understanding. I love solving Procurement challenges through technology and will continue to experiment with AI and new use cases! Ofcourse, I'd love to share my success and failures with the community through my Supernegotiate community!

Until next time!

Thanks

Gaurav

Founder, Supernegotiate