Batch Processing
The execution of a large volume of transactions or computations as a single grouped job, typically on a scheduled basis rather than in real time. Overnight batch jobs for invoicing, reconciliation, and demand planning remain common in ERP-heavy operations, and AI inference must be designed to fit these windows.
What is Batch Processing?
Batch processing means collecting a set of tasks and running them all at once rather than handling each one as it arrives. The canonical example is the overnight payroll run: instead of calculating each employee's pay when their timesheet is submitted, the system waits until end of day, gathers all records, and processes them in one job. The same pattern applies to invoice posting, demand planning updates, inventory reconciliation, and month-end closing procedures.
Batch processing was the dominant paradigm in enterprise computing for decades and remains deeply embedded in ERP architectures. Many core financial processes still run on nightly or weekly batch cycles — not because real-time is impossible, but because batch processing is efficient, predictable, and easier to audit.
Batch vs. Real-time Inference
AI systems can be designed to operate in batch mode, real-time mode, or both. The choice depends on the process requirements:
Batch inference: Process all invoices from the past 24 hours in a single overnight job. Efficient use of compute, predictable runtime, easy to monitor. Suitable when same-day processing is sufficient.
Real-time inference: Process each document as it arrives. Required when downstream steps depend on immediate results — a shipment confirmation that must release a customer order, a stock anomaly that must trigger an immediate reorder.
Latency tolerance: For most back-office processes, 2–4 hour processing windows are acceptable. For customer-facing or supply-chain-critical triggers, minutes matter.
Cost: Batch inference is typically cheaper per item — larger context, better hardware utilization, no idle capacity cost.
Batch Processing in Operations
At Lleverage, most invoice and document processing runs in near-real-time — documents are processed within minutes of arrival. But batch processing remains relevant for reconciliation tasks: comparing all posted invoices against all goods receipts for the week, running a demand forecast update on the full sales history, or generating a consolidated supplier performance report across 300 purchase orders. These tasks benefit from batch architecture because they require full-dataset access, not just the most recent transaction. Knowing whether a process needs real-time or batch processing is one of the first design decisions in any operational AI deployment.