Why your automation might miss part of a spreadsheet

A green check does not prove every record arrived

Imagine a team moving a supplier list into a new system. The workflow finishes without an error, but several suppliers are absent. Before asking an AI assistant to repair the list, check how much data the connection actually retrieved. Some actions return a limited set of records unless configured to fetch further pages. A summary built from an incomplete input can sound perfectly reasonable while leaving out real work.

Check the specific connector, not a general promise

Microsoft documents that the Excel Online (Business) List rows present in a table action returns up to 256 rows by default and points to pagination for retrieving more. It also documents that filters already visible in Excel are ignored when the action does not apply its own filter. These details apply to this connector and action; verify the equivalent settings for your actual tools rather than assuming every spreadsheet integration behaves the same way.

Documents the default row count, pagination and filter behaviour. Supplier reconciliation steps are proposed implementation guidance. Microsoft: Excel Online (Business) connector limits

Define the expected set before moving anything

Write down which table, rows and columns belong in the handover. For a supplier list, perhaps only active suppliers are wanted, but records missing a phone number must still appear for review. Save the selection rule with the run. Have the source owner confirm the expected record count and compare using the same cutoff time. A total from yesterday is not a reliable match for a list that colleagues are still editing today.

Reconcile identities as well as totals

For a hypothetical test, use 320 fictional suppliers with stable, unique references. Check that each expected reference appears once in the output and that no unexpected reference appears. Matching totals alone is insufficient: one duplicated supplier and one missing supplier could cancel each other out. Keep rejected records in an exception list with reasons. In this example, an output of 256 records would leave 64 missing; investigate the retrieval limit before treating the output as complete.

Test beyond the first page

Place recognizable test records near the beginning, around the retrieval boundary and at the end of the source. Include one excluded record and one record needing review. Run the workflow into a test destination, then inspect the actual output. Increasing a limit should be followed by the same completeness check. Ask the implementer how further pages are requested, what ends the retrieval and what happens if a later page cannot be read.

Make incomplete input visible

Agree a result such as complete, incomplete or needs review, with retrieved and expected counts where available. If completeness cannot be established, do not label the handover complete. For this supplier workflow, pause downstream updates until staff have resolved the difference. Preserve the references already processed so a retry can be reconciled instead of blindly adding the same suppliers again. An AI-generated explanation can help staff investigate; it should not invent records to make totals match.

Keep the solution proportionate

Sometimes the fix is a connector setting and a clearer selection rule. Other cases need a different retrieval method because of data volume or tool limits. Measure manual checking and correction time before and after the change, including maintenance effort and recurring software costs. Exovara can review one handover with your team, document how completeness is checked and confirm the scope before extending it to other business records.

Discuss an implementation

This guide applies across Canada. We provide remote AI consulting for businesses in Coquitlam; it does not describe a local client or a staffed office.

AI consulting in Coquitlam · Explore the service · Compare setup options