Data Cleansing Failures: Lessons for UK SMEs

Birmingham City Council’s Oracle ERP reimplementation has a detail that should make any small business owner sit up: phase one was budgeted at £19m, but the programme’s forecast cost through 2027/28 is now £144m. That’s not a “big organisation problem”. That’s what happens when data, governance, and resourcing don’t get treated as first-class workstreams.

This matters for the Governance, Regulation & Public Trust series because public-sector failures don’t just waste money—they erode confidence. And for UK SMEs, the lesson is even more practical: if your data is messy and your delivery capacity is thin, your systems will lie to you—about cash flow, stock, customer profitability, and compliance. The difference is that you won’t have an audit committee meeting to catch it before it hurts.

Here’s the reality I’ve seen across small firms: most “AI for business” conversations start with shiny features. They should start with data quality, ownership, and controls. The Birmingham Oracle story is a cautionary tale—and it also points to a better approach: using AI tools to keep data clean, work automated, and delivery predictable.

What Birmingham’s Oracle issues really warn us about

The headline risks raised around Birmingham’s project weren’t exotic technical problems. They were the basics—data cleansing, staffing capacity, change management, and governance under deadline pressure. Those basics are exactly what SMEs tend to skip because everyone’s busy.

Computer Weekly reported that auditors flagged insufficient data cleansing, especially around finance, and warned that documentation of data quality standards and readiness for migration “was not well-established”. The go-live was said to be July 2026, with risk of slipping to September 2026. On top of that: a stated 5% chance of abandonment—and the auditor suggested even that might be a low estimate.

If you strip away the enterprise branding, the underlying story is simple:

• If your finance data is inconsistent, your reports can’t be trusted.
• If your team is stretched, “project work” becomes “after-hours work”.
• If governance gets relaxed to hit a date, defects get baked in.
• If change management is weak, users create workarounds—and the system becomes fiction.

For SMEs, the stakes are different but the mechanics are identical. A small business doesn’t lose £144m. It loses invoices, misses VAT deadlines, over-orders stock, miscalculates margins, or can’t answer basic customer questions.

Data cleansing isn’t admin—it’s governance

Data cleansing is a governance decision. It answers: “What do we consider true?” and “Who is accountable when it isn’t?”

In public bodies, poor data quality can undermine statutory reporting and public trust. In SMEs, it undermines decisions that feel “obvious” right up until they aren’t—like whether you can afford to hire, which customers are profitable, or whether a marketing channel is actually working.

The SME version of “finance data isn’t ready”

You don’t need a complex ERP to get into trouble. Common SME data-quality failure modes include:

• Duplicate customers in CRM (one with a mobile number, one with an email)
• Supplier names spelled three ways in accounts (breaking spend analysis)
• Product codes reused “because it was quicker”
• Payments received but not matched to invoices (cash looks worse than it is)
• Staff exporting spreadsheets and re-uploading them (silent version drift)

These aren’t harmless quirks. They create governance risk: your business can’t produce consistent records when banks, auditors, insurers, or regulators ask.

Where AI genuinely helps with data prep (and where it doesn’t)

AI won’t magically fix chaotic processes. But it’s excellent at the repetitive, rules-driven work that humans hate and postpone.

Used properly, AI tools can:

• Detect duplicates (names, emails, addresses) with fuzzy matching
• Standardise formats (postcodes, phone numbers, dates)
• Classify transactions (categorise spend, flag anomalies)
• Monitor data drift (warn when “new messy patterns” emerge)
• Suggest mapping during migration (field-to-field matching, error detection)

AI won’t help if nobody agrees on the rules. If your team can’t answer, “What’s the authoritative customer record?” then automation just accelerates confusion.

A practical stance: clean data is a product, not a one-off task. Treat it like you treat your website—owned, maintained, and measured.

Resourcing problems: why projects break when people are “also” doing delivery

Birmingham’s audit committee discussion highlighted staffing pressure and the need to balance day-to-day responsibilities with programme demands. The programme director referenced over 100 full-time employees working directly on the programme, with backfill and agency support.

Most SMEs don’t have “backfill”. They have one person who knows how invoicing works, one person who can run payroll, and a managing director who is also the sales director.

The trap: “We’ll implement this between everything else”

This is where small business tech projects go to die:

• The project plan assumes stable capacity.
• Real life happens—sickness, peak trading, customer escalations.
• Data cleanup gets deferred because it’s not urgent today.
• Go-live arrives with messy data and half-trained users.

You end up paying twice: once to implement, and again to stabilise.

AI as a resourcing strategy (without pretending it replaces people)

AI’s best ROI in SMEs is often capacity creation—taking low-value work off your team so they can do the parts that require judgment.

High-impact automations include:

1. Invoice capture and coding: extract invoice data, propose ledger codes, route for approval.
2. Purchase order (PO) checking: flag mismatches between PO, delivery note, and invoice.
3. Customer support triage: summarise tickets, suggest replies, route to the right person.
4. Sales admin: auto-log calls/emails, draft follow-ups, update CRM fields.

This links directly to public trust and governance: when staff aren’t drowning in admin, they follow processes more consistently, records improve, and audit trails become real rather than performative.

Governance under pressure: the moment standards get “flexible”

One of the strongest warnings from the Birmingham auditor was about governance staying robust under pressure—because the bigger risk is compromised standards, not just schedule slippage.

SMEs face the same temptation, just quieter:

• “We’ll fix permissions later.”
• “Everyone can be an admin for now.”
• “Let’s import the spreadsheet and tidy it after.”

Then “later” never comes.

Minimum viable governance for small businesses

You don’t need a heavy framework. You need a handful of controls that stop preventable chaos.

Here’s a lightweight governance checklist I’d actually use in a 10–100 person business:

• Named data owners: one for customers, one for products/services, one for finance chart of accounts.
• Single source of truth: one system is authoritative for each dataset (not “whatever spreadsheet was last emailed”).
• Access control: least privilege, MFA, and no shared logins.
• Change control: any field/definition changes documented (even in a simple changelog).
• Audit trail: approvals recorded in-system (not in Slack, not verbal).

AI can support this by monitoring exceptions—e.g., flagging unusual journal entries, repeated manual overrides, or a sudden spike in duplicate records.

Predictive analytics for “risk of slip” (yes, even in SMEs)

Birmingham’s project discussed slippage risk (July to September 2026). SMEs rarely model this, but they should.

Even simple AI-assisted forecasting can help you spot:

• Tasks taking longer than baseline
• Backlogs building in approvals
• Support tickets rising ahead of go-live
• Data error rates not trending down

If you can measure it weekly, you can manage it. If you only notice at go-live, you’re already paying for rework.

Change management: where good systems go to die

The Birmingham discussion included monitoring operational signals like how many people raise purchase orders and how leave gets approved—basically, whether staff are practising the new world before it’s mandatory.

That’s a smart idea. Change management isn’t posters and training decks. It’s evidence that behaviour is changing.

What “change management” looks like in a small business

For SMEs, strong change management is usually three things:

1. Process clarity: “This is the new way. Here’s what good looks like.”
2. Short feedback loops: daily/weekly check-ins during transition.
3. Instrumentation: a small set of metrics that show adoption.

Adoption metrics you can track without making everyone miserable:

• % of invoices processed without manual edits
• % of sales records with required fields completed
• PO-to-invoice match rate
• Time-to-approve expenses
• Number of new duplicate customers created per week

AI can help by generating adoption summaries automatically and highlighting the teams or steps where errors cluster.

Practical next steps: a 30-day “clean data” sprint for SMEs

If the Birmingham Oracle story motivates you, use it. Don’t start with buying a bigger system. Start by making your existing data trustworthy.

Week 1: Decide what “clean” means

• Pick 3 datasets: customers, products/services, finance (suppliers + chart of accounts).
• Write a one-page standard for each (required fields, formats, naming rules).
• Assign a data owner for each.

Week 2: Measure your current mess

• Count duplicates (customers/suppliers).
• Count missing key fields (VAT numbers where applicable, addresses, contact details).
• Sample 30 transactions and check coding consistency.

Week 3: Fix the top 20% that causes 80% of pain

• Merge duplicates.
• Standardise names and codes.
• Lock down permissions so new mess isn’t created immediately.

Week 4: Add automation and monitoring

• Turn on AI-assisted categorisation or anomaly detection in your finance and CRM tools (many modern platforms include this).
• Set weekly alerts: duplicates created, approval backlog, high-risk anomalies.
• Publish a simple dashboard: “data quality score” + adoption metrics.

This is governance work, but it’s also commercial work. You’ll feel it in fewer invoice disputes, cleaner reporting, and faster month-end.

The bigger theme: governance builds trust—inside and outside the business

Birmingham’s Oracle issues are being scrutinised because public money and public trust are on the line. For SMEs, the trust relationship is different: customers, lenders, and regulators judge you based on whether your records are credible and your operations are controlled.

The primary lesson is blunt: technology doesn’t fix weak governance; it exposes it. If you want AI tools to help your UK small business, the fastest path to ROI is pairing them with clear data standards, realistic resourcing, and a few non-negotiable controls.

If you’re planning a system migration, a CRM overhaul, or even just “getting serious about reporting” this quarter, ask one forward-looking question: what would your business decisions look like if you trusted your data 99% of the time instead of 70%?