Before AI touches your spend data, fix the supplier master: a seven-step cleanup for procurement and finance
AI spend tools categorize and flag faster than any analyst, but duplicate suppliers, misread invoices and loose coding still produce wrong totals. The fix is shared ownership and monthly checks, not more software.
Article summary
- Why it matters: Spend visibility is the starting point for every savings case, and AI tools are now the default way teams build it. But a model that reads “Acme Ltd” and “ACME Limited” as two suppliers will understate your leverage with that supplier, and a misread invoice date or total will flow straight into your category reports.
- What to do first: Agree ownership before buying tools. Write down who owns the supplier master, the category map and the fallback code, and how often each is reviewed.
- Run a duplicate-supplier sweep. Group suppliers by tax ID or registration number and merge records before your next spend cube refresh.
- 7
- data checks to run before trusting AI spend analysis
- April 2029
- UK e-invoicing mandate for VAT invoices
- 6 years
- typical UK VAT record retention
On this page
The details
In a CPOstrategy column, Harvey Dhillon, founder of UK accountancy firm Zmartly and a former investment banker, argues that AI spend tools only reflect the finance records beneath them. If the supplier list, invoices and codes are wrong, the dashboard is wrong too, he writes, just “faster and in better colours.” Clean records come from ownership and routine checks agreed between procurement and finance, not from the software.
His checklist translates into seven controls procurement teams can own or co-own:
- Supplier master data. One owner. Match suppliers on company and tax registration numbers, not names, to remove duplicates that split spend.
- Invoice capture. OCR tools misread dates and totals. Sample-check monthly, including credit notes, part payments and foreign-currency invoices.
- Tax evidence. In the UK, valid VAT invoices are the evidence for reclaiming VAT and generally must be kept at least six years. Making a valid invoice a condition of payment keeps the data complete.
- Overseas suppliers. Foreign software vendors may not charge VAT, but the buyer can still owe it under the reverse charge. Keep a list for finance.
- Category mapping. One agreed map from spend categories to general ledger codes, with the fallback “miscellaneous” code cleared every month.
- Duplicate payments. Check supplier, amount and invoice number before each payment run.
- Monthly reconciliation. Match every bank and card line to an invoice and a code. This is where card-bought software below approval limits shows up.
The UK will require e-invoicing for all VAT invoices from April 2029, through the Peppol network. Structured invoices will improve capture accuracy, but as the column notes, they will not merge duplicate supplier records.
The most common duplicate-payment causes
Dhillon lists the patterns finance teams see most often:
- Paying both an invoice and the supplier’s statement
- Paying the same item by card and by invoice
- Processing a payment reminder as a new bill
- Having the same supplier on file twice, such as “Acme Ltd” and “ACME Limited”
Each also risks claiming the same VAT twice.
How duplicate records hide leverage: an example
Suppose one packaging supplier appears in your system three times, as “Acme Ltd,” “ACME Limited” and “Acme Packaging,” with $1.2 million, $800,000 and $500,000 of spend. If your policy sends suppliers above $1.5 million to a strategic review, none of the three records qualifies, and the dashboard shows a mid-sized supplier at most. Merged, it is a $2.5 million supplier, your negotiating position changes, and it may belong in a sourcing event. These figures are illustrative.
The same problem works in reverse for risk. A supplier split across records looks less important than it is, so it may not get the financial monitoring or business continuity review it needs.
How to test an AI classification tool before you buy
- Build a test set. Take a few thousand invoice lines your team has already classified correctly, covering direct, indirect and tail spend.
- Run it blind. Have the vendor classify the set without seeing your answers.
- Measure by spend, not lines. A tool that is right on 95% of lines but wrong on your largest invoices is worse than it looks.
- Check the low-confidence queue. Ask how the tool flags uncertain classifications and who reviews them.
- Re-run after cleanup. Merge duplicate suppliers and re-run the test. The improvement tells you how much of the accuracy depends on your data rather than the model.
This is our recommended approach.
A 30-day cleanup plan
| Week | Focus |
|---|---|
| 1 | Agree owners for the supplier master, category map and fallback code with finance |
| 2 | Group suppliers by tax or registration number; merge duplicates |
| 3 | Clear the fallback “miscellaneous” code and fix the category-to-ledger map |
| 4 | Sample-check captured invoices and set the monthly reconciliation routine |
Who owns what
| Control | Procurement | Finance |
|---|---|---|
| Supplier master and onboarding checks | Owns | Reviews |
| Valid invoice as a condition of payment | Sets in contracts | Enforces at payment |
| Category-to-ledger mapping | Co-owns | Co-owns |
| Duplicate checks before payment runs | Supports | Owns |
| Monthly bank and card reconciliation | Reviews tail spend | Owns |
| Overseas supplier list for reverse charge | Maintains | Uses |
This split is our framework based on the checklist; adapt it to your organization.
What procurement teams should do
- Agree ownership before buying tools. Write down who owns the supplier master, the category map and the fallback code, and how often each is reviewed.
- Run a duplicate-supplier sweep. Group suppliers by tax ID or registration number and merge records before your next spend cube refresh.
- Measure the “unclassified” line. Track the share of spend in fallback codes each month. If it isn’t shrinking, the AI is classifying around a data problem.
- Ask vendors the right question. In demos, ask how the tool handles duplicate suppliers and low-confidence classifications, not just how fast it categorizes.
What to watch next
- E-invoicing mandates. More countries are requiring structured e-invoices, which will improve data quality if suppliers comply.
- AI classification claims. Ask vendors how their accuracy changes when the supplier master is clean versus messy.
- Tariff exposure analysis. Expect spend tools to add country of origin, which only works if supplier records include it.
The bigger picture
AI is making spend analysis faster and cheaper, which raises the cost of bad inputs: wrong answers now arrive with more confidence. The teams getting value treat data quality as an operating rhythm shared with finance. For how leading platforms approach classification, see our spend analytics software ranking.
Questions procurement teams are asking
Can AI spend analytics fix bad supplier data on its own?
No. AI tools help categorize, match and flag anomalies, but they analyze the records underneath them. Duplicate suppliers, misread invoices and inconsistent coding produce wrong totals unless someone owns and checks the data.
What is the first data fix procurement should make?
Clean the supplier master: give it one owner and match suppliers on registration and tax numbers rather than names, so 'Acme Ltd' and 'ACME Limited' count as one supplier.
Why does spend analytics need clean supplier data?
Duplicate supplier records split spend across two entries, which understates your volume with a supplier and weakens negotiations. AI tools categorize faster, but they can't decide what a clean record is or who maintains it.
How do duplicate supplier records affect procurement?
They split one supplier's spend across several records, which understates your leverage, can keep the supplier out of strategic reviews and risk monitoring, and raises the chance of paying the same invoice twice.
How should I test an AI spend classification tool?
Give the vendor a set of invoice lines your team has already classified, measure accuracy weighted by spend rather than by line, check how low-confidence results are flagged, and re-run the test after cleaning up duplicate suppliers.
How we reported this
This analysis draws on 3 sources, listed below, and reflects figures as reported on October 14, 2026. The sections on why it matters and what procurement teams should do are Procurement Signal analysis. It is not legal or financial advice. See our editorial standards.
