Royal Mail fixed its claims automation by putting a human check before every AI claims decision. Its AI imaging tool confirmed that evidence had been uploaded, but not that it was genuine. Promise Akwaowo, Royal Mail’s AI Automation Lead, explained the fix at aiblLIVE London 2026.
A middleware layer now sends claims to a Salesforce queue, where customer advisors check the evidence against written rules. Only then does a claim go on to payment or back for resubmission. Here is how it works.
All figures as stated on stage by Promise Akwaowo.
The AI treated the presence of an image as proof. It confirmed that a claimant had uploaded a picture or PDF, but could not tell whether it showed the item they bought. Wrong evidence could pass and move towards a payout, causing rework, financial risk and customer friction when claims were later rejected.
The AI already assumes that that is the right evidence and it passes the check and it goes to make a decision.
Promise Akwaowo, AI Automation Lead, Royal Mail
Each claim lands in a Salesforce queue before any decision is made. Customer advisors check the evidence against rules about what counts as proof. They also merge or flag duplicate submissions. Valid claims go on to fulfilment, and weak or false evidence goes back to rework with a request to resubmit.
Errors are sorted into business exceptions, where the rules are wrong, and application exceptions, where the AI or automation fails. That tells the team what to fix.
It costed the process as it stood. Akwaowo set the annual claims value, the hours of effort and the staff involved against an estimated £500,000 exposure from doing nothing. He also spent three to four days in discovery with the claims advisors. The business case was approved.
Proportionately. Legal and compliance teams set the evidence rules and a value threshold, and only claims within it are investigated. Experienced advisors spot evidence that looks wrong, including images that may have been altered with AI. Asking for resubmission helps too, because claimants who know their evidence is false stop the process.
You can’t chase every case. You’d definitely just be on a goose chase for no reasons.
Promise Akwaowo, AI Automation Lead, Royal Mail
Promise Akwaowo is AI Automation Lead at Royal Mail. He works with business teams across the organisation, from customer experience to warehouses and finance, to find where simplification, automation and AI can help. He spoke on the Customer track at aiblLIVE London on 20 October 2026.
If your automation makes decisions on evidence customers supply, this is the control to borrow. The full PDF goes deeper than this page, in Akwaowo’s own words.
For heads of customer operations, claims and fraud leads, COOs and automation leads. Six pages, about an eight-minute read.
