AI, in your work.

What changes in the code

On October 27, 2026, Microsoft merged BCApps PR #13052, covering Shopify tax matching. The published diff replaces `GetGPT41Latest()` with `GetGPT56CeresLatest()`, changing the managed deployment selector.

According to the description, Chat Completions, forced function calling, and existing validation and review mechanisms remain unchanged. The change introduces no schema modifications, permission changes, or data migrations.

Scope and availability

Microsoft’s application card places Shopify Tax Matching in the US version of Business Central online, as a production-ready preview subject to additional terms. It matches tax descriptions from imported orders to configured jurisdictions; it does not determine tax liability or file returns.

The PR belongs to the Version 30.0 milestone. That planning information and the repository merge do not establish a shipped version or an actual Ceres deployment in a particular environment. The feature documentation does not identify a specific foundation model either.

What retaining review means

The setup guide requires Shopify Connector NA, activation of the capability, and enablement for each shop. The default review mode is Always; Low Confidence Only and Never are also available.

Retaining human review therefore does not mean every suggestion always reaches a person. An unidentified jurisdiction or a difference between the Shopify rate and the configured rate holds the order for review regardless of the mode. The Use Shopify Rate action changes shared tax data and can affect other documents.

Nexo’s interpretation: validate the process and the suggestion

Our editorial interpretation is that retaining controls preserves part of the process, but does not establish stable matching results. An existing code can be a formally valid response while still referring to the wrong jurisdiction.

The author reports successful internal evaluations, although the detailed evidence is not public. That report does not establish a general accuracy rate, savings, or improvement attributable to the change. A comparison would require cases, criteria, and results that can be examined.

Synthetic examples for an evaluation

The following table proposes invented cases for designing a comparison. It does not describe real orders, observed results, or tests performed by Nexo.

How to read a future comparison

Nexo proposes keeping the same sample and configuration, identifying the model actually used, and separating three records: the model output, the application’s decision, and the reviewer’s decision. The example percentages are arbitrary and do not represent applicable tax rates.

A difference between suggestions is not automatically an error. The comparison needs a justified reference match and a predefined way to resolve disagreements. Any claim about time saved or quality would require additional measurements.