Failed or stuck payments erode trust in instant payment systems such as India’s Unified Payments Interface (UPI), which processed more than 24 billion transactions in September 2026 alone. Most failure and fraud analytics act after a transaction has already been attempted. This paper presents DemoPay, a Flutter-based demonstration UPI application that estimates the likelihood of failure before the user confirms a payment. The estimate combines four transparent signals (bank server status, network latency, recent transaction load, and the bank’s recent decline history) into a weighted score between 0 and 1. The score is mapped to low, medium and high risk levels and accompanied by a plain-language list of reasons. A lightweight learning rule adjusts the factor weights from observed outcomes, with bounds and renormalisation so that every weight stays inspectable. To test the system without live banking access, the project separates the bank infrastructure into an independent Server app, which writes bank state to Cloud Firestore, and a Payment app, which reads it. We port the scoring and learning logic to Python and evaluate it on a synthetic payment stream with a known failure process, over 30 random seeds in two scenarios. When failures were congestion-driven, multi-factor scoring raised ROC AUC from 0.577 (server status alone) to 0.782. The adaptive weights changed ranking quality by less than 0.002 AUC but raised recall at the fixed medium-risk threshold from 0.19 to 0.25. We discuss why the learning effect is small, the limits of a synthetic evaluation, and how the design can be extended.

Ansari Abu Fahad Ashfaque Ahmed, Manoj Singh, "DemoPay: An Explainable, Adaptive Pre-Transaction Risk Scoring Framework for UPI Payment Failures, Built on a Decoupled BankServer Simulator" International Journal of Scientific Research and Engineering Development, V9(5): Page(1311-1319) September - October 2026. ISSN: 2581-7175. www.ijsred.com. Published by Scientific and Academic Research Publishing.