What Is Happening in the ERP World: August 2026

1. France switches on e-invoicing this week

On 1 September 2026, France starts its mandatory e-invoicing regime. The obligation that catches people out is not the sending one, it is the receiving one: from that date every business established in France has to be able to receive structured electronic invoices, whatever its size. Large and mid-sized businesses must also issue them. Smaller businesses get until September 2027 to issue, but not to receive.

Invoices flow through the public portal or a certified partner platform, and the certified-platform layer has been renamed from PDP to PA during the run-up, which has not helped anyone reading last year's project documentation.

The testing point is straightforward and gets missed anyway. This is not a configuration change, it is a new integration sitting in the middle of order-to-cash and purchase-to-pay. The invoice leaves your ERP, gets transformed, gets transmitted, gets acknowledged, and comes back with lifecycle statuses that have to land somewhere sensible. Every one of those hops is a place where a document can be rejected for a reason that never appears on a screen your finance team looks at.

If your UAT scripts for this stop at "post the sales invoice and check it looks right", you have tested about a third of it. The scenarios that bite are the ordinary ones: a credit note against an invoice from the previous regime, a customer whose registration details are subtly wrong, an invoice that gets rejected downstream and has to be corrected and resubmitted, a self-billing arrangement, an intercompany invoice between two entities on different platforms. None of that is exotic. All of it is untested on most projects.

2. The SAP ECC endgame is now a capacity problem

Mainstream maintenance for SAP ECC ends on 31 December 2027. That has been true for years and has been treated as a distant fact for most of them. It is now sixteen months away, and the earlier milestone has already passed: the transition window for Compatibility Packs closed on 31 May 2026, which for a lot of on-premise customers quietly removed the option of running certain S/4HANA functionality on the old licensing footing.

The numbers explain the mood. Analyst estimates put roughly half of the ECC base as having moved or actively moving by the end of 2025, which means a very large group is starting a programme that typically runs 18 to 36 months into a deadline that is sixteen months out. Consulting rates are widely expected to rise through the final year, and demand for experienced S/4HANA people is forecast to outstrip supply by some margin as 2027 approaches.

The practical defence is to make the shrinkage visible rather than silent. If you know your coverage, you can say precisely what a five-week window buys you and what it leaves untested, and someone can make an informed decision about accepting that risk. If your test evidence lives in a spreadsheet nobody has reconciled since week two, the conversation is just optimism against optimism. We wrote about how that decay happens in why UAT spreadsheets fail.

3. Agents stop being a demo

Business Central 29 arrives in October as the start of 2026 release wave 2, running through to spring 2027. The direction of travel is clear from what Microsoft has published: the agents introduced in wave 1 get extended rather than redesigned, with the Payables and Sales Order agents handling more of the process end to end and needing less manual review at each step. Oracle has been shipping the same shape of thing across Fusion, and SAP continues to push Joule deeper into the core processes.

The framing has changed in a way worth noticing. Last year the pitch was that an agent would draft something for a person to approve. This year the pitch is that the person steps in less often. That is a meaningful difference in where the risk sits.

It is also happening against a fairly sober backdrop. Gartner has forecast that more than 40% of agentic AI projects will be cancelled by the end of 2027, with unclear business value and inadequate risk controls among the reasons. Both things are true at once: the capability is landing in mainstream ERP releases, and a large share of the projects built on it will not survive contact with a business case.

The scenarios that need writing are the awkward ones: the near-duplicate invoice, the partial receipt, the vendor whose name changed, the document that arrives outside the pattern the agent has seen. And the acceptance criteria has to include the escalation path, because an agent that quietly does the wrong thing is a worse outcome than one that stops and asks. We went into this in more depth in autonomous ERP: who tests the machine? and, for BC specifically, in Business Central 2026 wave 1 and UAT.

4. A $172m lawsuit is reframing partner accountability

The Zimmer Biomet claim against Deloitte, filed in the New York Supreme Court in September 2025 over an S/4HANA implementation, is still working its way through the courts and is still the case the industry keeps coming back to. The medical device manufacturer is seeking $172 million, having paid around $94 million in fees, and its complaint describes a business that was barely operational through the third quarter of 2024, unable to ship product, issue invoices, or produce basic sales reporting. Deloitte disputes the claims, and the contract terms look likely to do a lot of the work in deciding it.

What makes it relevant here is the language in the complaint. The allegation is not that the software was wrong. It is that the delivered system was over-customised and, in the plaintiff's words, riddled with defects and functionality gaps. That is a description of something that went live without anyone having established what worked.

It sits alongside the other case we covered recently, where Tennant's ERP disruption turned into a securities investigation. Different mechanism, same underlying pattern: the operational failure is bad, and then the question of what the organisation knew, and when, and what it told people, turns out to be the expensive part.

5. The UK picks Peppol for 2029

Closer to home, the shape of the UK mandate firmed up considerably this summer. Following the 2025 consultation, e-invoicing becomes mandatory for VAT invoices from April 2029, and in June 2026 the government confirmed Peppol as the interoperability framework: a four-corner model, without the real-time e-reporting obligation that several European regimes have bolted on. A fuller technical roadmap is expected at the Budget in November.

April 2029 sounds comfortably far away. It is not, if you are choosing an ERP now. A system selected in 2026 and implemented through 2027 will be the system that has to handle this, which makes it a question for your current requirements list rather than a problem for a future project team. The useful thing about the Peppol decision is that it removes the main excuse for waiting: the destination is known, even if the detail is not.

Anyone with French operations gets a preview this week, and it is worth treating that as free reconnaissance rather than an isolated compliance job.

The through-line: more change, same test window

Read those five together and a single pattern falls out. The volume of change arriving in ERP systems is going up, from several independent directions at once: statutory compliance deadlines with fixed dates, a migration wave concentrated into a shrinking window, twice-yearly release waves that now move functional behaviour rather than just features, and agents taking decisions that used to require a person.

The amount of time organisations give to validating that change has not gone up at all. In most programmes it has gone down, because the deadline is fixed and testing is the last flexible thing before it.

That gap is where the failures in this article came from. Not from anyone being careless, and not from bad software. From change arriving faster than anyone could establish what it had broken.