Philippines staffing research · Published · Updated
How batch size changes error discovery in outsourced document work
A larger batch can turn one wrong assumption into dozens of bad records. Compare batch sizes by correction cost, check cost, and the distance to discovery.

Research findings
Batch size is usually treated as a productivity setting. A larger batch looks efficient because the coordinator handles similar items together and avoids switching tasks. The same choice also changes when errors are discovered. In a small batch, a mistake appears close to the item that caused it. In a large batch, the mistake may repeat across dozens of items before anyone notices.
The effect is easiest to see in document processing. A coordinator opens a stack of invoices and keys the same fields for each one. If the field mapping is wrong, every invoice in the stack inherits the error. The first item could have revealed the problem, but no one checked it before the rest were processed. The batch turned one mistake into many.
Small batches create more opportunities to check. Each batch boundary is a natural pause where a reviewer can compare the output with the source. The pause costs time, but it limits the spread of an error. The trade-off is not between speed and quality in general. It is between fewer checks with a wider blast radius and more checks with a narrower one.
The right batch size depends on the stability of the task. A task with a stable template and a known error rate can tolerate a larger batch. A task with a new template, a new source, or a recent change should start small. The first batch after a change is a test, not a production run. Treat it that way and check before scaling.
Error discovery also depends on who checks. A coordinator checking their own batch may miss the same thing twice because they apply the same assumption. A reviewer checking a sample may catch a pattern the coordinator cannot see. The batch size and the review method should be designed together. A large batch with no independent check is a long feedback loop.
The cost of a late error is not only the correction. The team may have already sent outputs to customers, posted entries, or built reports on the wrong data. Repairing those downstream uses takes longer than fixing the original item. The later the discovery, the more places the error has traveled. Batch size is one of the few levers that changes discovery time without changing the task itself.
Discussion
A simple inspection model can help choose a batch size. Estimate the chance that a batch contains at least one error, the cost of correcting one item, and the cost of checking the batch. Compare a few batch sizes using the same error rate. The model will not give an exact answer, but it will show whether the current batch is far from a sensible range.
Do not assume that smaller is always better. Very small batches increase setup time, context switching, and the number of handoffs. They can also make the work feel fragmented and reduce the coordinator's ability to see patterns. The goal is a batch small enough to catch an error early and large enough to keep the work coherent. That range is usually wider than teams expect.
Record the batch size and the discovery point for each error found. The discovery point is where the error was caught, not where it was made. A pattern of late discoveries suggests the batch is too large or the check is too weak. A pattern of early discoveries suggests the current design is working. Keep the record simple so it is actually maintained.
Implications and limits
Review the batch size after any change in the source, the template, or the team. A new source may introduce a different format, and a new coordinator may need a smaller batch while learning. The right size is a property of the current process, not a permanent rule. Revisit it when the process changes.
Train the coordinator to stop the batch when something looks wrong. A batch that continues after a suspicious item turns a small question into a large correction. Give the coordinator a way to flag the item without completing the rest. The flag should be quick and should not require a manager's approval. A stopped batch is cheaper than a repeated error.
Measure the result in both directions. Track the error rate and the throughput after a batch change. A smaller batch that cuts errors but halves output may not be worth it. A larger batch that raises output but increases rework may look efficient on paper and fail in practice. The comparison should include the cost of correction, not just the count of items.
What batch size changes
Batch size changes the distance between an error and its discovery. It also changes the number of setup steps, the amount of work in progress, and the ease of reassigning a case. A large batch keeps a coordinator in one mode, which can improve speed. A small batch creates more checkpoints, which can improve detection. Neither effect is automatic.
The effect on morale is real but often ignored. A coordinator who discovers a large batch of errors may feel blamed for a design problem. A coordinator who works in small batches sees corrections as part of the routine. The design should make early detection normal rather than heroic.
Batch size also affects how work is assigned. A large batch can be handed to one person, while a small batch can be split across a team. Splitting can reduce the time to first check, but it also creates more handoffs and more chances for inconsistent interpretation. Choose the split only when the task can be divided cleanly. A shared batch with unclear ownership is worse than a larger batch with one owner.
A simple inspection model
Write the model with three inputs: error probability per item, correction cost per item, and check cost per batch. For each batch size, estimate the expected number of errors before a check and the expected correction cost. Add the check cost and compare the totals. Use ranges rather than single values because the inputs are uncertain.
The model is most useful for comparing two or three options, not for finding an optimum. It forces the team to state the cost of a late error, which is often the missing number. If the team cannot estimate that cost, start with the downstream repair time. The exercise is valuable even when the answer is approximate.
Choosing a batch size
Start small after a change and increase only after the error rate is stable. Set a maximum batch size for new tasks and a different maximum for mature tasks. Write both limits in the queue instructions so the choice does not depend on the coordinator's mood. Review the limits when the source or template changes.
Give the reviewer a sample rule that matches the batch. A small batch may be checked in full, while a large batch may need a sample plus a check of the first item. The rule should say what happens when the sample finds an error. The answer is usually to stop the batch and inspect the rest.
Set the batch limit in the tool when the tool allows it. A written limit that depends on memory will be ignored during a busy period. The tool can stop the coordinator at the limit and prompt a check. If the tool cannot enforce the limit, use a physical or visual marker at the batch boundary. The check should be easy to perform at that point.
Limits of the model
The model assumes that errors are independent, which is often false. A single wrong assumption can produce the same error across every item in a batch. That correlation makes large batches riskier than an independent-error model suggests. Treat the model as a lower bound on the risk.
It also ignores the value of a fast customer response. A small batch may be better for a customer waiting on one item even when the arithmetic favors a larger batch. The service promise belongs in the decision. Batch size is an operational choice, not only a quality choice.
Questions
Is a small batch always better for quality?
Usually it catches errors earlier, but very small batches add setup and handoff time. Choose a range that balances detection with a coherent workflow.
What should be recorded for each error?
Record the batch size and the point where the error was discovered. That pair shows whether the check is catching problems early or late.
When should the batch size change?
After a change in the source, template, or team. A new coordinator may also need a smaller batch during the learning period.
Should the coordinator be allowed to stop a batch?
Yes. A quick stop-and-flag rule prevents one suspicious item from becoming a large correction across the rest of the batch.