Transaction Activity Forecasting

How can we use AI to predict a client’s future transaction activity from their historical financial transaction patterns?

Problem

Clients generate rich streams of everyday banking activity: salary deposits, subscriptions, utility and other household bills, loan repayments, card purchases, transfers, and cash withdrawals. Some of these transactions are one‑off, while others follow recognizable patterns and form recurring payment streams that reflect a client's financial commitments and lifestyle.

Understanding and anticipating these recurring transactions can help clients stay in control of their finances: planning ahead for regular outgoings, avoiding missed payments, and spotting changes early. It also enables more tailored support, such as personalized insights, relevant alerts, and guidance aligned with each client's situation and goals, as well as earlier detection of unusual activity that may indicate fraud or misuse.

In this challenge, you will work with a synthetic transaction dataset designed to resemble real-world client behavior. Each record represents an individual transaction and includes information such as timing, transaction type, amount, merchant category, and description. Using historical transaction data, your task is to predict the next recurring transaction a client is expected to make.

The challenge requires participants to uncover recurring payment patterns hidden within noisy transaction histories. Successful solutions will recognize behavioral regularities, capture temporal relationships, and generalize across diverse client spending and payment habits, ultimately helping us anticipate client needs and support them more effectively.

Objective

Participants are invited to:

• Build an AI pipeline that uses historical transaction data to predict a client's next recurring transaction.
• Identify and model recurring transaction patterns from raw transactional data, including temporal, behavioral, and merchant-related signals.
• Explore different approaches to feature engineering, sequence representation, and recurrence detection.
• Compare multiple modeling techniques and evaluate their performance using appropriate forecasting metrics.
• Analyze the strengths, limitations, and interpretability of their solution, and communicate key insights derived from the data.