Pay.UK unlocks £100m in preventable APP scams losses
Industry collaboration proves a smarter way to detect scams earlier, reduce losses at scale, and help PSPs meet reimbursement rules without delaying genuine payments.
Overview
1. The Ambition
To prove whether AI, combined with syndicated and onboarding data, could detect APP fraud earlier and reduce losses without adding friction.
2. The Solution
Three models were tested on over 1.7 billion transactions. The best results came from blending AI with syndicated transaction data and client-specific onboarding signals.
3. The Results
We pioneered an approach that can prevent over £112m in annual losses. Our groundbreaking project delivered a 5:1 false positive rate and 68% efficiency gain with 98.8% safe approvals.
Company
Pay.UK
Industry
Banking
Location
United Kingdom
The Ambition
APP fraud is rarely simple. It is entangled in complex economic crime networks, often involving first-time offenders or unwitting participants. Most scams show little sign of traditional fraud risk until it's too late.
With mandatory 50/50 victim reimbursement, Payment Service Providers face growing pressure to detect and block scams before authorisation. Synectics, in collaboration with Pay.UK, Visa and Featurespace, launched a pilot to test whether combining AI with syndicated data could materially improve scam prevention.
The project aimed to explore the value of a new fraud overlay service that uses transactional intelligence, onboarding signals and predictive modelling to uncover risk earlier in the payment journey.
“This pilot demonstrates the importance of innovation and collaboration in staying ahead of fraudsters and protecting people in the ever-changing payments landscape.”
- Pay.UK
The Solution
Under a pioneering data-sharing agreement, Synectics was able to analyse one year of syndicated Faster Payments data from participating banks and PSPs. This included over 1.7 billion rows of live transactional records.
Precision AI was used to test three distinct approaches. First, a model using only core transaction data. Second, a version enhanced with syndicated account-level features. Third, a bespoke model incorporating each organisation’s own onboarding and application data.
The combined approach delivered the most accurate and efficient detection outcomes. It showed that when AI is powered by both shared and local intelligence, scam detection is faster and more accurate.
This is because models can draw on behaviour data from across multiple organisations and the entire customer lifecycle - enabling prompt, precise identification of scams that would otherwise evade detection.
The Results
The combined model is projected to prevent over £112 million in scam losses every year. It achieved a better than 5:1 false positive rate - meaning at least one in five flagged payments was a confirmed scam, with minimal friction for genuine customers.
Detection efficiency also increased by 68 percent compared to using transaction data alone. In fast-track scenarios, 98.8 percent of transactions were safely approved. For high-value payments above £1,000, the model blocked the equivalent to £41 million in annualised savings.
These results show how combining Precision AI with syndicated and onboarding intelligence delivers sharper decisions and stronger fraud control at scale.
About our partners
This project was a joint initiative between Synectics, Pay.UK, Visa and Featurespace. Together, we combined syndicated Faster Payments data, predictive AI and fraud expertise to test a new approach to scam prevention to support the entire payments industry tackle fraud with confidence.
Got a challenge or a question?
Get in touch to see how we can work together to prevent fraud by mitigating risk and staying ahead of emerging threats.
We’d love to hear from you.