Birmingham's custom bank reconciliation and the clean core rule every CFO needs

Meta confirmed this week that chief global affairs officer Joel Kaplan plans to leave the company in 2027, ending a tenure of more than 15 years as a key adviser to CEO Mark Zuckerberg on political and regulatory matters. The length of the exit is the interesting part. Kaplan is expected to remain in his position through the U.S. midterm elections and into 2027 to facilitate a leadership transition. Companies plan a handover that carefully when the knowledge in question lives in one person's head and was never written down.

Your finance ERPERPA single integrated software backbone that runs core operations: finance, procurement, supply chain, HR and manufacturing on shared data.View full definition → holds the same kind of knowledge. It sits in custom code, modified tables and bespoke interfaces, written over fifteen years by developers and consultants who have mostly moved on, and none of them is staying for a twelve-month transition. The concept that deals with this is theclean core. Vendors and integrators treat it as an architecture term, so most CFOs hand it to IT. My argument in this piece is that a clean core is a finance governance decision about where the company's financial logic lives and who can explain it. Treat every modification inside the ERP core as a liability that keeps costing money, and most finance functions will find they carry far more of it than they can justify, just as the SAP maintenance deadline and AI agentsAI agentsAgentic AI refers to AI systems that pursue goals autonomously by planning, taking actions through tools, and adapting based on results, with minimal step-by-step human direction.View full definition → arrive to collect.

Why the clean ERP core belongs on the CFO's agenda

The calendar is the first reason. SAP has confirmed that mainstream maintenance for SAP ECC 6.0 ends on December 31, 2027. Adoption is behind. SAPinsider's 2026 benchmark, based on a survey of 296 community members conducted between December 2025 and March 2026, found that while 55% of respondents report having deployed a version of SAP S/4HANA or SAP S/4HANA Cloud, only 34% have fully switched their daily operations to it. In Germany, the DSAG user group's 2026 investment report found that only 37% of companies still running SAP ECC expect to be on its replacement, S/4HANA, by the end of 2027.

AI is the second reason, and it now carries more weight than the deadline. The same SAPinsider research found that SAP's AI announcements have overtaken the 2027 maintenance deadline as the top external factor shaping ERP strategy, cited by 43% of organizations versus 39% for the deadline. Gartner predicts that finance organizations using cloud ERP applications with embedded AI assistants will see a 30% faster financial close by 2028, and it also notes that most CFOs are still in the early stages of adoption, hindered by data qualitydata qualityThe degree to which data is fit for purpose: accurate, complete, consistent, timely, valid and unique. Poor quality data undermines analytics, reporting and AI.View full definition →, integration complexity, skills gaps, and inconsistent multientity support. Those embedded agents read the standard data model. A heavily modified core gives them a ledger they were not built to understand. This week's news that accounting firm advisor Stout acquired Flightline, an agentic AIagentic AIAgentic AI refers to AI systems that pursue goals autonomously by planning, taking actions through tools, and adapting based on results, with minimal step-by-step human direction.View full definition → company, building out its AI-enabled workbench is one more sign that agent tooling is reaching finance operations faster than most cores are ready for it.

The third reason belongs to the CFO alone: the audit opinion. If customized logic breaks reconciliation or period allocation, IT does not sign the accounts. You do.

How does a clean ERP core actually work?

A clean core means the vendor's standard code and data model stay unmodified, and anything specific to your business runs outside them through published interfaces. In SAP's version, as one implementation guide puts it, no SAP table is manually altered, no standard programme is overridden, and no unpublished internal object is called from client code. All extensions run exclusively through APIs officially published by SAP. The principle is the same for Oracle Fusion or any other cloud ERP: the vendor owns the core and keeps updating it, and you own the edge.

Why the old model stopped working

In the ECC era, customers altered SAP standard tables, inserted custom logic directly into standard programs, and built Z-objects that sat inches from core SAP code. SAP tolerated this because the upgrade cadence was slow. That tolerance is gone because cloud ERP updates arrive several times a year. Each modification inside the core has to be retested, and sometimes rebuilt, at every release, indefinitely. That is the carrying cost. In ECC the cost was spread thinly across occasional upgrade projects, so it never appeared as a line anyone owned. Under a continuous release model it shows up every quarter.

Gartner draws the strategic consequence: ERP project scope that is highly customized based on current or near-term requirements will fail to keep up with changing business goals after implementation. The firm also predicts that by 2027, more than 70% of recently implemented ERP initiatives will fail to fully meet their original business case goals. As many as 25% of these will fail catastrophically.

Birmingham and Zimmer Biomet show what a modified core costs

Birmingham City Council is the clearest public record of what happens when finance customizes the core. The council went live with the Oracle Fusion single platform for finance, payroll, HR, procurement, supply chain management, and customer relationship managementcustomer relationship managementCustomer Relationship Management: software and strategy to manage and analyse customer interactions throughout their lifecycle.View full definition →, replacing an SAP system first introduced in 1997, in April 2022. The plan was to run Oracle "out of the box". Instead, the council added customisations, including a bank reconciliation design that did not work. Manual workarounds have cost more than £5 million. Problems listed in press coverage include a decision to turn off audits for fraud detection for more than 18 months; allocating £2 billion in transactions to the wrong year. In January 2026 officials put the total forecast cost of the program up to the 2027/28 financial year at £144.4 million, up from £19 million, which is roughly 7.6 times the original figure. University of Sheffield researchers at the Audit Reform Lab estimate the full cost, including savings the council had banked and then lost, will reachreachThe number of unique people exposed to your message in a given period. Unlike impressions, reach counts each person once, no matter how often they see it.View full definition → £216.5m in April 2026. In September 2026 The Register reported that the council will go at least one more year without auditors vouching for its accounts. The fix it chose is telling: it is re-implementing the standard product.

Zimmer Biomet shows that the same risk applies in the private sector. These are allegations, and Deloitte disputes them. According to the medical device maker's complaint, Deloitte promised to deliver SAP S/4HANA within the agreed-to timeframe and budget, with minimal customization, and instead delivered an overly customized software system riddled with defects and functionality gaps for which Zimmer Biomet paid $94 million in fees. The company said it was barely operational through the third quarter of 2024, unable to ship or receive product, issue invoices, or generate basic sales reporting, and the problems led it to reduce its full-year 2024 guidance. Zimmer Biomet is seeking at least $172 million. Deloitte will defend ourselves vigorously against this meritless claim. Whatever the court decides, notice that "minimal customization" was a contract promise and nobody on the client side owned it as a measurable number.

A worked example: pricing your custom code before you migrate

Here is a hypothetical with arithmetic only, built on published benchmarks. ASUG survey data, as cited by ITChamps (a consultancy that sells code remediation, so check the figure independently), indicates the typical enterprise holds more than 2,000 custom objects requiring assessment before migration. The same firm cites SAP's own figure that up to 40% of custom code in legacy ECC systems is either unused or redundant.

• Start with 2,000 objects. At 40% redundancy, 800 can be retired before anyone writes a line of migration code. That leaves 1,200 to classify.
• Suppose a two-person team can classify 60 objects a week (a hypothetical rate). The 1,200 survivors take 20 weeks. Without the first cut, 2,000 objects would take about 33 weeks.
• On cost, VendorBenchmark, a firm that sells benchmarking services, reports that custom code rationalization projects of 3-4 months reduce migration consulting costs by $500K, $3M on mid-to-large projects. Use that range as a starting point for your business case and confirm it against your own integrator's quotes.

The point of the exercise is the order of work. Retirement is the cheapest decision you will make in the whole programme, and it is a finance decision, because only process owners can confirm that a 2011 report or a custom posting rule no longer matters. That makes the inventory as much aquestion of which numbers finance actually trusts as a technical audit.

When a clean core is the wrong call, and the rule for deciding

A clean core is a default. It is not an absolute rule, and pretending otherwise gets CFOs into trouble in three ways.

First, some logic really does differentiate you: a pricing model, a regulated revenue treatment, a settlement process your competitors cannot copy. Forcing it into standard either destroys value or pushes it into spreadsheets, which is worse than a well-documented modification. Second, moving code outside the core relocates complexity without removing it. Extensions need their own governance, testing and owners, and one SAP-focused consultancy warns that many are also creating new technical debt while they remediate the old debt. Third, the deadline can push you into a conversion you would not otherwise choose. ISG research reported in early 2026 found 34% of organizations choosing brownfield, 18% greenfield, and nearly half (48%) opting for a bluefield hybrid, and ISG warns that brownfield and hybrid approaches "may limit long-term benefits and make it more difficult to adopt newer technologies such as AI". Still, paying for extended support is a legitimate option. Gartner itself has advised that customers who struggle to quantify S/4HANA business value should evaluate whether staying on ECC for the near term is a viable option. A deliberate delay with a clean-up plan beats a rushed conversion that carries every modification forward.

The whiteboard rule for every ERP modification

Write this up and apply it to every custom object, interface and modified table:

If it does not change how we make money or how we meet a law, it goes back to standard. If it does, it lives outside the core, with a named owner who can explain it in one paragraph. Nothing new goes inside the core without CFO sign-off.

The named-owner test is where the Kaplan story applies. Meta can afford a long handover because it knows which person holds the knowledge. Most finance teams cannot name the owner of half their Z-programs. An object without an owner is already orphaned, and it should be retired or rebuilt while the team still has time to choose. The hardest part is human. ASUG found business-process change to be the single biggest barrier to migration, named by 49% of those surveyed, so thechange management work of getting controllers to give up processes they built deserves the same budget attention as the code.

What to do on Monday about your ERP custom code

• Ask IT for the count of custom objects, modified standard programs and custom interfaces in the finance modules, along with last-execution dates for each. One week is enough to get a first extract.
• Flag every object not run in the past 24 months as a retirement candidate, and send the list to the controllers who own each process for a yes or no.
• Pick the five modifications closest to cash and the close (bank reconciliation, intercompany, revenue postings) and require a named owner and a one-paragraph explanation for each by Friday. Birmingham's failure point was a customized bank reconciliation.
• Pull your integrator contract and find out whether "minimal customization" appears as a number with a remedy attached, or only as a phrase.
• Put one question on the next steering committee agenda: which modifications would block the vendor's embedded AI close tools, and what does each one earn to justify that cost?

The custom code inventory is the most underused document in finance transformation. If you can name the owner of every modification that touches cash, you have a clean core strategy. If you can't, that inventory is where to start this week.