Leak 39 · Data & Product Information - Duplicate customer and product records

Acme Mfg, ACME Manufacturing, and Acme Manufacturing Inc. Three records, three price files, three credit limits, and a sales report that shows none of them as your biggest customer.

Diagnostic note. Symptoms, required records, an initial check, and possible fixes. This shorter note does not include a full worked example.

What is it?

Duplicates are created at speed: a CSR cannot find the customer and creates a new one; a buyer adds an item that already exists under another supplier’s number. Every analysis in this library is then wrong by the amount of the duplication. Pricing is inconsistent because the customer exists three times. Slow stock is overstated because the item exists twice.

Family
Data & Product Information
Primary owner
IT / Controller
Secondary owners
Customer Service Manager, Purchasing Manager
Primary impact
System-wide
Typical data source
ERP customer master, item master
Detection difficulty
30-day measurability

Ask yourself

How many customer records do you have for your largest ten customers, and how many item records for the same physical product?

Yes, partially, no, or don’t know. “Don’t know” is the most useful answer, because it points at the test below.

What does it look like?

Warning signs. None of these proves the leak exists. They tell you where to look.

  • The same customer under several names or ship-tos set up as separate accounts.
  • The same product under several item numbers.
  • Reports that require manual consolidation before they make sense.

What data do I need?

The minimum viable set. Most of it is already in your ERP.

Field
Customer master: name, address, tax ID, phone
Item master: description, manufacturer part number, UPC, supplier part number

The initial check

Start with a small sample. Gathering the exports, agreements, or observations is separate from running the check; agree that work with the person who owns the records.

  1. 1Export the customer master and sort by the first eight characters of the name. Count near-matches for your top twenty customers.
  2. 2Export the item master and sort by manufacturer part number. Count items that share one.
  3. 3For one duplicated customer, compare the price files and credit limits across records.
Duplication rate = duplicate records ÷ total records, for customers and items separately

Then ask one question: Which analyses in your business are currently wrong because of these duplicates?

How much could it be costing us?

A conservative range, not a headline. The goal is a number management can trust enough to investigate.

Not directly quantified. Duplicates inflate or hide every other leak. Estimate via the leaks they distort, such as inconsistent pricing and slow stock.

Common root causes

Fixes fall into three layers. Not every problem needs software, and almost none needs AI first.

Process
No ownership of master data and no creation standard.
Data
No unique keys such as tax ID or manufacturer part number enforced on creation.
Technology
The ERP allows creation without a duplicate check.

What should we do?

Start with the simplest intervention that could solve it. Move down the list only if the one above is not enough.

  1. Level 1

    Measurement

    A duplicate report for customers and items, ranked by activity.

  2. Level 2

    Ownership

    One person approves new customers and items, against a checklist.

  3. Level 3

    Cleanup and controls

    Merge the active duplicates, then add duplicate checks at creation.

Where AI helps

  • Fuzzy matching of names, addresses, and descriptions to find duplicates that exact matching misses, and proposing merges for review.

Where AI probably doesn’t

Stopping new duplicates is a process control. Do that before the cleanup or the cleanup will not last.

Before you call it a leak

  • Some apparent duplicates are legitimate: separate legal entities, or the same product in different packaging. Review before merging.

Think this might be happening in your business?

Turn the finding into a next step.

If the numbers say there is something there, send us what you found and we will help you decide whether it is worth a full investigation. No transaction files needed for that conversation.

Let’s start with one thing.

What would better performance look like?

Bring a result you want to improve, a symptom, or a workflow you already understand. You do not need to know the bottleneck yet. We’ll help choose what to investigate first.

No transaction files needed for the first conversation.