In 2018, Birmingham City Council approved £19.965 million to replace its legacy SAP system with Oracle Fusion. The current forecast for the program is £144.4 million through 2027/28, more than seven times the original figure. Along the way the system allocated £2 billion of transactions to the wrong year, the council could not produce auditable accounts, and the whole thing had to be rebuilt.
The rebuild went live in August 2026. The external auditor called the switch largely successful. It still expects to issue disclaimer opinions on the council's accounts for 2025/26, 2026/27 and 2027/28, and does not expect a clean opinion before 2028/29. That is roughly a decade after the program began.
It is tempting to read this as a story about Oracle, or about one council's management culture. Both played a part. But the decision that did the most damage was made on the buyer's side of the table, and it is one every ERP buyer faces: whether to customize a system they bought specifically so they would not have to.
The plan said adopt. The delivery said adapt.
Birmingham's stated intent was to implement Oracle out of the box. As Computer Weekly's account puts it, the original plan was for the council to change its processes to fit the software. "But it was changed, and Oracle was customised to support Birmingham's existing business processes."
The Register reports that the council created several customizations, including a banking reconciliation system that "failed to function properly." The council "struggled to understand its cash position and was unable to produce auditable accounts." Audit trails used for fraud detection were switched off for more than 18 months. The failed implementation, together with poor management of equal pay claims, left the council effectively bankrupt in 2023.
Cash and bank reconciliation are about the worst places to put a custom build. They sit at the centre of period close and audit. When they break, the damage lands in the statutory accounts, where it is slow and expensive to unwind.
No reporting we found names who approved each deviation from standard, or why. That absence is the point. At a March 2025 audit committee, Grant Thornton's Mark Stocks said "all the suppliers said go live", but with caveats that should have been examined. His summary: "Nobody took ownership." Accounts payable and receivable customizations were not switched on at go-live, and change requests were still arriving.
"Out of the box" was a sentence in a business case. Nobody had the job of enforcing it.
How the fix was built
The repair is more instructive than the failure. Birmingham did not write a better custom reconciliation module. According to Computing, after the service failures the council dropped its customizations and moved to an out-of-the-box version of Oracle Fusion, using third-party software to fill the functional gaps. It bought a separate product for banking reconciliation, and its replacement income management system went live in December 2025, ahead of the main rebuild.
That is the pattern: keep the ERP core standard, and put anything genuinely different alongside it, connected through supported integration points. The council reached that design only after the first system failed. It could have been chosen at the start, when the cost of choosing it was a design workshop rather than a rebuild.
The rebuild was also run differently. Officials described a staged approach with independent assurance at agreed checkpoints, and moved go-live from April to the summer, partly to protect payroll accuracy, insisting the delay was "not a reset or a scope change." The auditor said the reimplementation "appears to have been better managed than the original 2022 implementation," with no outstanding business-critical or priority 1 or 2 issues after cutover. The old system had 311 open tickets in September 2025.
The cost nobody prices in
Two costs of customization rarely appear in a business case.
The first is time. Birmingham's new system is live and, by the auditor's account, working. The accounts are still expected to carry disclaimers for three more years because of problems with historic data. Damage done to financial records during the failed years outlasts the system that caused it.
The second is recurring. Oracle Fusion Cloud updates arrive every quarter, and Oracle's own guidance is blunt: they are "mandatory and cannot be skipped." Test environments update two weeks before production, and Oracle tells customers to "[t]est customizations (including integrations, extensions, reports, and workflows)" in that window, every cycle. Each deviation from standard is a regression-testing obligation four times a year, for as long as you run the system.
Dorset Council's experience shows the long tail from the other direction. Its own cabinet report notes "various bespoke extensions to the core system over the years," and its SAP version reaches end of life at the end of 2027. An outline business case found the platform "no longer fit for purpose", blaming heavy customization for higher workloads and manual "offline" processes. Dorset now says it will adopt "standard Oracle best practice, rather than try to bend the solution to fit," and cites Birmingham by name.
Birmingham set out in 2018 with the same intent. Intent did not hold then. It needs something stronger behind it.
A scenario worth rehearsing
Picture a mid-sized organization six months into an ERP rollout. The finance team says the standard bank reconciliation screen does not match how they match receipts across three banks and a payments processor. The integrator says it can build a custom matching module in eight weeks. It is the easiest path, and it keeps the finance team happy.
A governed process asks three questions before anyone writes code. Can the business change its process to fit the standard? If not, is there a product that does this job and connects through the vendor's supported interfaces? If not, who on the buyer's side owns the custom build, its quarterly regression tests and its audit exposure for the next ten years? Asked plainly, the custom module rarely survives the third question for a finance-critical function.
A rule you can adopt on Monday
If your organization has bought, or is about to buy, a platform on the promise of running it standard, write that promise down as a rule with an owner.
- Name a design authority on the buyer side. One person or small group who must approve every deviation from standard in writing. Not the integrator, whose incentives point toward building.
- Rank the options, in order. Change the process. Configure the platform. Add a product or service alongside it through supported integration points. Build inside the core only as a last resort.
- Price the full life of every exception. Include quarterly update testing, documentation and the cost of removing it later. If nobody will fund that, the exception is not approved.
- Set a higher bar for finance-critical functions. Cash, bank reconciliation, period close and payroll touch the audited accounts. A defect there costs years, not weeks.
- Keep a live register of deviations. Review it before every go-live gate and every quarterly update. Anything without a current owner gets challenged.
- Stay the intelligent customer. Suppliers will say go live. Someone on your side has to read the caveats and decide.
None of this is exotic. Much of it mirrors how Birmingham ran its second attempt: standard processes, staged checkpoints and independent assurance, with a forecast now more than seven times its first budget. The cheaper time to adopt it is before the first custom build is approved.
If you are weighing configure against build on an ERP or integration program, we are happy to talk it through at midrasolutions.com.