Microsoft Moved Sustainability Into the ERP Portfolio. No UAT Plan Has a Tester Who Can Sign Off a Carbon Number.

What Microsoft actually said

On 3 November, April Olson — Chief Product Officer for Dynamics 365 Finance, Human Resources and Sustainability — posted on the Dynamics 365 blog: "We are welcoming Microsoft Sustainability Manager into the Dynamics 365 ERP portfolio." The post calls the product Microsoft Dynamics 365 Sustainability throughout.

What follows is intent rather than a shipping date. Microsoft lists "greater interoperability with Dynamics 365 Finance" as a planned investment, and "a green ledger view designed to bring financial and sustainability insights together". It names the frameworks supported: CSRD, GRI, the IFRS Sustainability Disclosure Standards and SASB. And it sets a boundary worth quoting, because it is the most useful sentence in the post: "This does not mean every sustainability decision becomes an ERP transaction."

There are no dates in it, and nothing is marked public preview or generally available. ERP Today's write-up two days later called the Copilot features and autonomous agents preview; Microsoft's post labels nothing. Asked what a client gets and when, the honest answer is that it does not say.

What is new, and what was already there

Here is the part the coverage mostly missed. Sustainability Manager was already filed under the enterprise resource planning section of Microsoft's own release plans: the 2026 release wave 1 overview sits at a URL with enterprise-resource-planning in the path, and already lists "a green ledger view connecting Sustainability Manager with Dynamics 365 Finance" as one of four investment areas, alongside multiple calculation versions and reporting-period lock controls.

So the 3 November post consolidates branding and ownership around work already planned and already classified as ERP. That is the gap between announcement cadence and shipping cadence, and conflating them is how teams promise a date that never existed. The same trap caught the trade press when the release wave model was retired: much of the coverage implied Business Central's update cadence was going continuous, which it was not.

One real consequence follows: that same page notes release plans stop being published from September 2026, with new capabilities moving to the AI at Work roadmap. The reference you would check is winding down as the product arrives.

These numbers are calculated, not captured

An ERP test plan is built around transactions entered and then processed. An emissions figure is not one. Microsoft's documentation on calculation models describes a configured pipeline: an estimation factor step that "converts proxy data to activity data", a condition step that "determines when the calculation model is used", then a report step where you pick a calculation library (the documented example is EPA) and an emission factor, optionally mapped through reference data such as a facility.

Read that as a tester. Conditional logic choosing which model applies. Proxy data converted by factors. Custom logic in Power Fx, with documented gaps: the page states that "IsEmpty and IsBlank in Power Fx aren't currently supported" and that "multi-select option set fields aren't supported within calculations". A checkbox governs whether a calculation emits one record or one per report node, which the docs warn can produce "multiple records from one calculation".

That is configuration behaving like code, with the property that makes financial configuration dangerous: a wrong factor or a mis-ordered condition produces a plausible number rather than an error. Nothing fails. A figure appears, correct to four decimal places, and flows onward — the same failure mode as a mis-set posting group, and only caught by a test case with a pre-stated expected result.

Who signs off an emissions figure

On a finance workstream, sign-off has a settled answer: the controller accepts the trial balance, because the controller answers for it. Ask the same of a sustainability calculation and most project structures go quiet. The sustainability lead knows the frameworks but not the configuration. The consultant knows the configuration but cannot certify a disclosure. Finance has the accountability reflex but has never seen a calculation model.

That is an evidence problem, not a scheduling one, and it bites once a number is externally reported. A sign-off record exists to name who accepted each result on a date. A green cell and some initials cannot say which factor library version produced the figure or which condition fired — the structural reason spreadsheet-based UAT fails exactly when it is needed.

What belongs in the test plan

A worked example per calculation model, computed by hand first. Activity quantity, factor, expected output, written down before the system is asked. A model nobody can reproduce on paper cannot be verified, only watched.

Negative and boundary cases on the condition steps. Conditions select the model, so the interesting tests are activity records matching no model, or two. Silent omission is the characteristic defect, and never shows up in a happy-path script.

The factor library version recorded beside every signed-off result. Factor libraries are reference data, and reference data gets updated. Without the version, nobody can later say whether a number moved because the data changed or because the library did.

A named accepter who did not build the model. Treat it like a Business Central finance workstream: scripted, evidenced and signed, not demonstrated and nodded through.