KTS-[SB10]Branch Traffic Prediction & Smart Queue Management

## Inspiration

Bank branches still handle many important customer interactions, but branch traffic is highly uneven throughout the day. Customers often arrive during peak hours without knowing expected waiting time, while branch managers have limited tools to anticipate surges and prepare the right staffing mix. This creates a poor customer experience, longer queues, staff overload during peak periods, and underutilization during quieter windows.

We were inspired by a simple question: what if customers could know the best time to visit a branch before leaving home, and what if branches could predict traffic early enough to act on it? For Shinhan Bank, this problem is especially meaningful because branch service quality directly affects customer satisfaction, while smarter traffic prediction can also encourage more customers to use digital channels for simple transactions. We saw an opportunity to combine historical branch traffic, transaction patterns, and live queue/check-in signals into an AI system that improves both customer experience and branch operations.

## What it does

KTS-[SB10] Branch Traffic Prediction & Smart Queue Management is an AI-powered system that predicts branch traffic and estimated waiting time, then turns those predictions into practical recommendations for both customers and branch staff.

For customers, the solution shows the best time slots to visit a branch, estimated waiting time by branch and time window, and suggestions to use digital channels for simple transactions that do not require in-person service. This helps customers avoid unnecessary waiting and choose the most convenient branch experience.

For branch operations, the platform forecasts customer volume, identifies expected peak periods, estimates service demand by transaction type, and recommends queue-management actions such as opening more counters, reallocating staff, or redirecting low-complexity requests to self-service or digital channels. It gives branch teams a forward-looking view instead of forcing them to react only after queues have already formed.

In short, the product acts as an intelligent traffic control layer for branches: predict demand early, manage queues smarter, and improve customer flow across the network.

## How we built it

We designed the solution around three main intelligence layers.

First, we built a prediction layer using historical branch traffic patterns such as branch visit volume, day-of- week behavior, hour-of-day peaks, and transaction mix. This layer estimates expected branch demand for upcoming time windows and identifies likely congestion periods.

Second, we added a real-time queue intelligence layer that incorporates live check-in and queue signals. Instead of relying only on historical averages, the system continuously adjusts waiting time estimates based on current branch conditions. This makes the output more useful for live decision-making.

Third, we created an action layer for both customers and branch teams. On the customer side, the system recommends the best visit windows and suggests digital alternatives for simple requests. On the operations side, it provides branch-level recommendations such as preparing more tellers for high-demand periods, shifting staff to match transaction demand, and flagging unusual spikes that need intervention.

To make insights easy to consume, we framed the output in a simple dashboard experience: predicted traffic level, expected wait time, branch congestion status, and recommended actions. We also explored how an AI assistant can explain why traffic is increasing and help teams interpret demand patterns more quickly.

## Challenges we ran into

One of the biggest challenges was data realism. Queue prediction is only as good as the quality and granularity of the input data, and branch traffic depends on multiple moving factors: day, time, branch location, transaction complexity, staffing level, and sudden real-world events. In a hackathon setting, turning this into a reliable predictive model required us to simplify some assumptions while still keeping the prototype useful and believable.

Another challenge was converting raw traffic prediction into operational value. Predicting “more customers will come at 11 AM” is not enough. The harder problem is translating that into estimated waiting time and recommended actions that branch teams can actually use. We had to think beyond analytics and focus on decision support.

We also faced the challenge of balancing customer convenience with branch efficiency. Recommending alternative times is helpful, but the system becomes much more valuable when it can also redirect simple transactions to digital channels and reserve branch capacity for more complex services. Designing that logic in a practical, user-friendly way was an important part of the solution.

Finally, as with many hackathon projects, time was a constraint. We had to choose which parts to prototype deeply: forecasting logic, queue estimation, dashboard experience, and actionable recommendations. The challenge was not only building AI, but building AI that solves a real banking workflow.

## Accomplishments that we're proud of

We are proud that our project goes beyond a generic forecasting demo and addresses a real operational pain point inside the bank. Instead of building AI for the sake of AI, we focused on a concrete business problem with clear value: reducing waiting time, improving service efficiency, and creating a better branch experience.

We are also proud of how we connected multiple stakeholders in one solution. Customers benefit from better visit planning, branch staff gain better visibility into expected demand, and the bank benefits from smarter staffing and stronger digital-channel adoption. That end-to-end thinking makes the concept more practical and scalable.

Another accomplishment is that we framed the system as both predictive and actionable. The prototype does not stop at showing traffic charts. It recommends when to visit, when to prepare for queue surges, and when to shift simple requests to digital journeys. That action-oriented design makes the solution more aligned with real branch operations.

## What we learned

We learned that branch traffic is not just a forecasting problem. It is an orchestration problem involving customer behavior, service capacity, transaction complexity, and operational decision-making. Good AI in this space must connect prediction with real-world branch actions.

We also learned that explainability matters. For internal users, a forecast is much more useful when they can understand what is driving the prediction and what action they should take next. In banking operations, trust in AI output is just as important as model accuracy.

Another key learning was that customer experience and operational efficiency can reinforce each other. By helping customers choose the right time or use digital alternatives for simple tasks, the bank can reduce congestion and free branch teams to focus on higher-value interactions. That creates a better experience on both sides of the counter.

## What's next for KTS-[SB10] Branch Traffic Prediction & Smart Queue Management

The next step is to move from prototype to pilot with real branch and queue data. We want to validate prediction accuracy across different branch types, traffic patterns, and transaction categories, then fine-tune the model using actual operational feedback.

We also want to expand the solution into a more complete branch intelligence platform. That includes branch-level heatmaps, live congestion alerts, staffing recommendations by service type, and personalized customer notifications inside the SOL app suggesting the best time to visit or the best digital alternative.

In the longer term, this solution could support branch network optimization at a strategic level. Beyond daily queue management, the same intelligence could help identify chronic congestion patterns, underused branch capacity, and opportunities to redesign service journeys across physical and digital channels. Our vision is to make branch service more predictive, more efficient, and more customer-centric.

Built With

• c++
• dart
• flutter
• javascript
• qwen