Updated 2026-06-01
AI in spend management: signal vs noise
Every vendor says they use AI. Here's what's real, and what's marketing.
Quick answer
TL;DR
Every vendor says they use AI. Here's what's real, and what's marketing.
- •Real: receipt OCR, transaction categorisation, anomaly detection, policy enforcement.
- •Real-ish: AI savings insights (Ramp), AI invoice processing (Yokoy, Payhawk).
- •Marketing: 'AI-powered' on everything else.
Where AI actually works
Receipt OCR is solved — every modern platform does it well. Transaction categorisation is 90%+ accurate. Policy violation detection is reliable. Anomaly detection (duplicate spend, abnormal vendor behaviour) is good and getting better.
Where AI is still half-baked
AI savings insights are interesting but vendor-priced. AI invoice processing works for clean PDFs but breaks on weird formats. AI 'recommendations' on category spend are usually just averages with a chatbot bolted on.
What to ask vendors
Show me an unstructured invoice you can process. Show me a savings insight that's not just 'switch from monthly to annual'. Show me the model evaluation methodology.