Most procurement AI projects do not fail at deployment. They fail six months earlier, when someone decides to run a proof of concept on top of inconsistent data, with intelligence that nobody can rely on.
It often starts with a compelling GenAI demo. Real enthusiasm from the business. A pilot that produces output that looks useful. Then a slow stall, as the team realises they cannot access clean, connected data.
The key challenge with procurement automation
Procurement is one of the hardest functions to automate well, and I think the industry underestimates this. Supplier negotiation is a high-stakes, trust-based process. The outcome depends not just on price, but on relationship history, risk exposure, and commercial intent built up over years. The moment you cannot explain how a recommendation was reached, or who is accountable for acting on it, you have a commercial legal and compliance problem.
The data problem sits underneath all of it. Most large organisations use multiple enterprise resource planning tools (ERPs), with supplier, contract, and performance data scattered across different systems and owned by different teams that use different definitions and terms. Often data from one system gets ‘processed’ in another system also. Making traceability to its source, or defining the ‘authoritative datasource’ for each datapoint a cumbersome exercise. Yet AI cannot make reliable recommendations from inconsistent sources. And connecting the signals across supplier risk, contract terms, performance history, and commercial intent into something coherent is not a simple task either. It requires continuous investment in a single version of the truth. Without it, even the most sophisticated AI model won’t be able to deliver tangible improvements.
Intelligence nobody can act on
If the underlying data is inconsistent, procurement contracts will have gaps, and AI recommendations and automated actions will be unreliable. The result is wasted investment in intelligence no one can use.
The fix is not more AI. It is better data. This requires one version of the truth (the authoritative source) for supplier master data, contracts, performance, and risk as well as clear decision-making roles and responsibilities. It is also essential to establish strong governance policies that empower teams to act with confidence, rather than constantly going back and forth for approval. None of this is glamorous work. It does not demo well. But without it, even a well-built AI model is guessing without ‘understanding’ the context of the organisation and its suppliers.
Start small and move quickly
In 2026, Excel and email cannot be the backbone of procurement. This does not mean procurement should wait for a perfect technology program to materialise. It means being clear-eyed about what you are actually trying to do.
The pilots I have seen work well are not the most ambitious ones. They pick one specific, high-impact use case. They invest in the data foundations first. They define who owns each decision and what the guardrails are. And they run the ideate, test, and iterate cycles in days, not weeks or months. Traditional enterprise delivery, where you define requirements for six months and then run a long proof of concept, does not work in this environment. The conditions keep changing, so a program that arrives twelve months later, fully formed, is usually solving last year’s problem.
Autonomous negotiation is advancing, but only for the right use cases
Another area where we are seeing a lot of experimentation in is autonomous contract negotiation. AI could add value here, but only for the right use cases. Real automation is within reach only for transactional, commoditised spend. High-volume, rules-based negotiations can run end-to-end within clear guardrails, with humans stepping in for managing exceptions. We are testing this and it is delivering results. But the further you move toward complex, strategic relationships, the more decisions depend on judgment that no AI model handles well yet. Treating both categories as the same problem could result in misplaced investment and poor outcomes.
The bigger prize is speed, not autonomy
There is a more immediate role for AI that gets less attention than the autonomous negotiation conversation and it is where more of the value sits right now.
Teams are no longer just asking whether a product is provided at the optimal cost. They are asking whether a supplier is resilient, and what plan B looks like if they are no longer able to fulfil their obligations due to disruption. That due diligence takes time that procurement does not always have. When you can connect signals across suppliers, contracts, inventory, and logistics in real time, and turn them into clear recommended actions, you make decisions faster and with more confidence.
AI that identifies leverage you would not have spotted manually, or flags supplier stress early enough to act on it, is delivering the biggest value right now. It helps create more resilient supply chains.
Specialist AI tools can also play a key role in accelerating supply chain decarbonisation. AI, combined with digitisation, allows organisations to better track and manage carbon footprint across their supply chains, including scope 3 emissions. The ability to view the carbon footprint across your supplier network and make informed decisions about strategic partnership or how to support suppliers in advancing their own net zero strategies is key for enabling greener supply chains.
Getting the foundations right
But none of these benefits can be realised without getting the foundations of right. Gartner says only 29% of supply chain organisations have built the capabilities needed for future AI performance. It is a data, governance, and operating model gap. Procurement leaders who deploy AI onto fragmented foundations will keep running impressive pilots that go nowhere.
Readiness looks different from the outside than it does from the inside. From the outside, it looks like a technology programme. From the inside, it is mostly organisational work: getting teams to agree on what a supplier record should contain, who signs off on a recommended action, and how you know whether the tool changed anything.
Data quality. Process standardisation. Governance. None of it is optional. If procurement leaders want AI to move beyond the pilot phase, they need foundations strong enough for teams to trust the outputs. Get that right, and AI stops being an interesting experiment and becomes a practical advantage driving faster decisions, stronger resilience, and measurable impact at scale.