Why Speed and Latency are Critical in Check Processing
- Balancing high throughput and low latency is essential for modern, AI-based check processing
- OrboAnywhere Turbo 6.0 boosts throughput while sharply reducing latency on existing hardware.
- Modern AI infrastructure with ONNX & NVIDIA GPUs enables real-time check recognition & fraud detection
In a recent Medium blog post, banking transformation leader Amit Batra draws a sharp distinction between two metrics that quietly determine whether AI-powered banking feels seamless or painfully slow: throughput and latency.
Think of a banking system as a busy airport.
Throughput is the number of flights that land and take off in a day — it’s all about volume. Latency, on the other hand, is the time it takes for a single passenger to board a plane from check-in to departure — it’s about speed.
In practical banking terms, high throughput means being able to process tens of thousands documents -- such as loan applications, transactions, or check images -- within hours, which is essential for overnight batch work and regulatory reporting.
Low latency, by contrast, measures the delay or time required to complete a single operation from start to finish. This is critical for check processing for both real-time and batch processing.
Speed and Latency in Check Processing
Mr. Batra emphasizes that these metrics are often in conflict: Batching transactions boosts throughput but forces individual items to wait, while always-on real-time processing improves responsiveness but can choke capacity at peak load. AI-driven banking stacks increasingly need to blend the two -- by running, for example, bulk fraud pattern analysis in the background while still handling high-risk events in real time.
This is exactly the focus for OrboGraph's OrboAnywhere Turbo 6.0 check processing. As noted in our Product Innovation Brief:
- Throughput improvements (Speed in DPM - Documents Per Minute): Most OrboAnywhere modules attain 2X to 4X overall speed improvement on the same hardware comparing V6.0 to V5.3. For example, Anywhere Recognition can run nearly 10,000 DPM on a single A100 GPU.
- Faster latency (Response time for real-time API calls): Reduction in recognition latency by up to 70%, with results as fast as 100ms (milliseconds). Real-time check fraud detection can be accomplished with low latency using Anywhere Fraud.
In addition, OrboAnywhere's Turbo 6.0 release modernizes the AI Infrastructure to ONNX (Open Neural Network Exchange) and new certifications of NVIDIA L-Series GPUs.
By upgrading to the latest OrboAnywhere Turbo 6.0, financial institutions receive a competitive edge. As explained by Medium.com:
Throughput and latency are like two sides of the same coin, each influencing the efficiency and effectiveness of banking operations in its unique way. While throughput enables banks to handle large volumes, latency ensures that time-critical processes are seamless and responsive.
Furthermore, "The future of banking AI lies in mastering this balance, leveraging cutting-edge technologies to redefine what’s possible."
If you would like to learn more, email marketing@orbograph.com or reach out to your account manager.