Why AI agents stall in procurement, and how to get one into daily use
Key takeaways
- Building a procurement AI agent is now the easy part. Getting the team to trust it and use it every day is where pilots stall.
- Five common causes: no owner, no agreed limits, the wrong first workflow, no baseline, and no feedback loop after delivery.
- Treat the agent like a new team member: a role card, close supervision at first, and a review every two weeks against measures taken before it started.
- Start with frequent, rule-based work where mistakes can be caught early, such as supplier follow-up or first-pass bid checks.
Building an AI agent is no longer the hard part. A capable team can have one drafting supplier emails or checking bid documents within days. The hard part comes next: getting a procurement team to trust it, use it every day, and know when to overrule it.
Across the UAE and the wider GCC, procurement leaders are being asked what AI will do for their function. Many have run a demonstration or a proof of concept. Far fewer have an agent doing real work on live cases. When a pilot stalls, the technology is rarely the cause.
Five reasons procurement agents stall
1. Nobody owns it. The agent was built by IT or a vendor, and the procurement team treats it as someone else's tool. When it makes a mistake, no one is responsible for fixing it, so people quietly stop using it.
2. Its limits were never agreed. Can it send a chaser to a supplier without approval? Can it reject an incomplete bid? If nobody has written down what the agent may do alone and what needs a person, every output becomes a judgement call, and the team falls back to doing it themselves.
3. The workflow was the wrong one. Agents do best on work that happens often, follows rules that can be written down, and where a mistake can be caught before it costs money. Pilots that start with a rare, high-stakes decision struggle to show value and frighten the people who would use them.
4. No one measured the before. Without a baseline (time per case, rework, cycle time), the end-of-pilot review becomes a discussion of impressions. Sponsors cannot scale on impressions.
5. It was a project, not a way of working. The agent was delivered, handed over and left. Agents need the same short feedback loop as any new team member: watch the work, correct it, adjust the rules, review again.
What works instead
Treat the agent like a new member of the team, and run its introduction the way an Agile team runs any change: in short iterations, with clear roles and evidence at every review.
- Pick one workflow. Good first candidates in procurement are supplier follow-up, first-pass completeness checks on bids, requisition triage and contract renewal tracking. Each is frequent, rule-based and easy to check.
- Write a role card. One page: what the agent does, what it may do alone, what always needs approval, what it must never do, and who owns it.
- Supervise first, then loosen. Start with a person checking every output. Relax the checks only where the agent has earned it, workflow by workflow.
- Measure from day one. Capture the baseline before the agent starts, then compare at each review.
- Decide on evidence. After six weeks the sponsor has real numbers from their own cases: scale it, adjust it, or stop.
None of this is new. It is how good teams have always introduced new ways of working, and it mirrors the step-by-step approach in our guide to implementing agile procurement. What changes with AI agents is that the new team member works fast, never tires, and makes confident mistakes. That makes clear rules and close supervision more important, not less.
Where procurement teams should start
If you are planning a first agent, three questions help:
- Which task does your team repeat dozens of times a week that follows clear rules?
- Who in procurement would own the agent and answer for its work?
- What would you need to see after six weeks to decide whether to scale it?
If you can answer all three, you are ready for a pilot. If you cannot, that is the first thing to work on, and it costs nothing to start. For the wider picture of how AI is changing purchasing in the region, see Agile procurement and AI in the UAE.
Run a pilot with us
We are running a small number of six-week AI agent adoption pilots with procurement teams in the UAE and the wider GCC. Each pilot takes one real workflow, sets the agent's rules and review points with your team, and ends with a clear scale, adjust or stop decision based on your own measures.
It builds on our work helping procurement functions adopt Agile ways of working. If you would like to explore a pilot, talk to us.


