ERP Roundup: The BC30 Preview Environment Deletes Itself, and Microsoft Stops Selling AI Builder Capacity in Nine Days
A preview environment you cannot reschedule and cannot recover, a deployment that quietly became reversible, an add-on that stops selling in nine days, and a backlog number that is really a staffing forecast. Four stories from the past seven days. No named ERP failure, lawsuit or testing study surfaced in that window, so this is four rather than five.
1. The BC30 preview opens, and the rehearsal environment has a self-destruct
Yun Zhu posted on 21 October that the early access preview for Business Central 2027 release wave 1 — BC30, the April version — is available, relaying an announcement in the BC Viva Engage group; his own sandbox reported platform 30.0.56466.0 and application 30.0.56564.0. He is an MVP relaying a community announcement rather than a Microsoft release note, so treat the build numbers as his and the rules as Microsoft's.
The rules are the story, and they are on Microsoft Learn. Early access is sandbox only, DE, GB and US localisations only, and needs the Partner Sandbox licence — a partner rehearsal facility, not something a client spins up. The environment updates to the latest preview build every Sunday, and Microsoft is explicit that "it isn't possible to manage or reschedule this update". Environments that fail it "for any reason are automatically deleted and can't be recovered". All previews are deleted 30 days after GA, and registered notification recipients get no email about any of it — the Operations page is the only record.
One point of accuracy, because the trade press keeps blurring it: retiring the release wave announcement model did not change Business Central's shipping cadence. BC30 is still the April 2027 major version, opening early access when the October update period begins, exactly as documented. What has gone is the release plan you used to scope from.
What it means for testing
Read the deletion clause as a design constraint, not a footnote. Anything you build inside a BC30 preview — a configured company, a test data set, a half-finished compatibility matrix — is on a weekly coin toss and a 30-day post-GA expiry. The evidence has to live outside the environment: scripts, expected results and sign-off records in your test management system, with the sandbox as a disposable execution surface. The deliverable this month is one run answering one question per extension — does it install and upgrade on 30.0 — recorded somewhere that survives Sunday.
2. You can now cancel a scheduled PTE install
Zhu also reported on 20 October that a per-tenant extension install scheduled through the Business Central admin centre can now be cancelled there: Manage apps, then Manage updates, then Cancel scheduled install. Previously the only way to unpick one was to upload the same version again and overwrite it. The confirmation warns that cancelling "will remove this version of the PTE from our services", the Operations page records who cancelled it, and the action is not available in the moments immediately after scheduling. This is a community finding — Zhu cites no Microsoft documentation for it, and nor could we find any — so verify it in your own tenant before writing it into a runbook.
What it means for testing
A deployment that can be withdrawn changes what sign-off is for. Scheduling a PTE against the next minor update used to be a commitment, which pushed teams to sign off first. A cancel action lets you schedule early and keep testing — useful, and also the exact condition under which a half-tested build drifts into production because nobody revisited the schedule. The control is a named owner and a dated decision point before the update window closes, not the existence of an undo button. Same discipline sign-off is supposed to encode: someone says yes, in writing, on a date.
3. AI Builder capacity add-ons stop selling on 1 November
A Partner Center announcement dated 22 October tells CSP distributors, direct bill partners and indirect resellers that Microsoft stops selling AI Builder capacity add-ons on 1 November 2026. Customers keep existing credits to the end of their current subscription term. The action Microsoft names is specific: turn off auto-renew before each customer's next renewal date so the retired add-on does not renew — in Partner Center for direct bill partners and indirect providers, via the distributor for indirect resellers. Once credits expire, the recommended path is Copilot Credits.
What it means for testing
Nothing breaks on 1 November. It breaks at a renewal date months out, on a tenant nobody flagged, when a document processing step stops having capacity behind it — the same class of problem as last Friday's Copilot renaming: a commercial change with no code change and no screen to click. No acceptance script catches it, because the script passed when it was written. What catches it is an inventory — which tenants consume AI Builder capacity, against which term, renewing when — and a regression assertion that fails loudly when the capacity is gone rather than degrading into an empty field. Entitlements are a dependency. Test them like one.
4. SAP books another quarter of backlog: that is your 2027 delivery queue
SAP published Q3 2026 results on 21 October: current cloud backlog of €24.1 billion, up 28% (26% at constant currencies), cloud revenue up 26%, Cloud ERP Suite revenue up 29%, total revenue up 13% in what the CFO called the first quarter above €10 billion. SAP also says its Autonomous Enterprise strategy "is not just resonating, it is being adopted" — the vendor's characterisation of its own demand, with no adoption metric behind it in the release, so read that part as positioning rather than evidence.
The backlog figure is harder to argue with. Contracted cloud revenue not yet recognised is, functionally, work somebody has to stand up. A 28% larger number is a 28% larger queue of go-lives arriving into a delivery market that did not grow 28% in qualified people.
What it means for testing
Capacity pressure lands on acceptance first, because it is the last phase and the only one whose stakeholders are part-time. When a programme runs out of calendar, scope and date do not move; the test cycle compresses. Monday's post made the point from the other end of the pipe: detection capacity now scales with compute and disposition capacity does not. A growing backlog is the macro version. Fix the acceptance scope to the contracted requirements before the squeeze arrives, so what gets cut is a decision someone signed rather than coverage that quietly did not happen.
The through-line: four calendars, none of them yours
A preview that updates on Sunday whether you are mid-cycle or not. A scheduled install that runs at the next minor update unless someone stops it. An end of sale on 1 November that detonates at a renewal date you have not looked up. A backlog that becomes go-live dates eighteen months out. None is a defect, none has a screen, and all four are clocks running in someone else's building. The work is the dull part: know which dates bind you, and keep the evidence somewhere that outlives the environment that produced it — most of what structured UAT on Business Central actually buys you.
Free ERP UAT checklist (Excel)
A structured workbook covering the full UAT cycle: pre-UAT preparation tasks, test execution tracking, an issue log with severity guide, and post-UAT wrap-up. Built from real ERP implementations.
No spam. We will occasionally send UAT and ERP implementation resources. Unsubscribe any time.
Sources and further reading
- Yun Zhu (Microsoft MVP), Dynamics 365 Lab: The Early Access Preview for Dynamics 365 Business Central 2027 release wave 1 (BC30) is available, relaying the BC Viva Engage announcement (21 October 2026)
- Microsoft Learn: Preview environments, covering the Partner Sandbox licence requirement, DE/GB/US localisations, the weekly update that cannot be rescheduled, automatic deletion on a failed update, and deletion 30 days after GA
- Yun Zhu (Microsoft MVP), Dynamics 365 Lab: Business Central 2026 release wave 2 (BC29) — cancelling a scheduled PTE update install from the admin centre, a community finding with no Microsoft documentation cited (20 October 2026)
- Microsoft Partner Center: October 2026 announcements, including the AI Builder capacity add-on end of sale on 1 November 2026 and the auto-renew action for CSP partners (announced 22 October 2026)
- SAP: SAP Announces Q3 2026 Results, with current cloud backlog of €24.1bn up 28%, Cloud ERP Suite revenue up 29% and total revenue up 13% (21 October 2026)
- Microsoft Learn: Update cycles, confirming Business Central still ships two major versions a year in April and October
Your Test Evidence Should Outlive The Sandbox That Produced It
LogicHive keeps the acceptance layer where a disposable preview environment cannot take it with them: structured test cases tied to requirements, named business owners, linked defect tracking, and sign-off built from real execution data rather than a screenshot of a passing build. When a preview is deleted, an add-on is retired or a delivery squeeze arrives, the record of who accepted what, and when, is already there. The free plan runs a real project end to end.
Written by Sharif George
LogicHive Frontman
Sharif is the frontman for LogicHive. He writes the ERP news roundups and the Business Central release coverage, following each Microsoft, SAP and Oracle change through to what it means for regression scope and acceptance testing on live implementation projects.
View all articles by Sharif GeorgeRelated reading
ERP & Enterprise
AI Found the Defects Faster Than Anyone Could Fix Them. PostgreSQL Cut the Features. ERP Projects Cut the Testing.
PostgreSQL 19 reached RC1 on 15 October with about 8 to 10 significant features reverted, because LLM-assisted review produced defect reports faster than committers could close them. The widely repeated "forty bugs" figure was an illustration of workload, not a count. The transferable point for ERP delivery: detection capacity now scales with compute, disposition capacity does not, and when a cycle runs out of calendar it is usually the testing that gets cut rather than the scope or the date.
19 October 2026 · 6 min read
ERP & Enterprise
ERP Roundup: UCLA Settles With Oracle Over a System That Allegedly Did Not Meet Its Requirements
UCLA pays $350,000, Oracle waives $1.31m and credits $3.1m, and the allegation recorded in the settlement is that the system lacked the functionality to meet the university's requirements — six years after kick-off. Alongside it: three SAP HotNews notes where the workaround switches off a business process, a Copilot catalogue that renames itself today, and BC29 update notices with a March 2027 backstop. Four stories, none of them a defect a test cycle would have found.
16 October 2026 · 7 min read
ERP & Enterprise
AI Writes the Test Cases Now. The Survey Behind This Week's Headline Says Less Than the Headline Did.
Applause's functional testing survey reached the UK trade press this week under the headline that AI use has surged and defects are rising. Its own data table is messier: more respondents reported defects falling on both number and severity (26.4%) than reported both rising (14.7%), and 19.8% do not track it at all. Meanwhile the single most common use of AI in testing is writing the test cases. On an ERP project the scripts are the acceptance scope, so that is a change in who authored it.
12 October 2026 · 6 min read