How Poor Data Quality Affects Business Decisions

Most problems fall into four patterns. Missing data is a blank field, such as a phone number or product category, and it silently drops that record from any report that filters on the field. Duplicates are the same customer, product or transaction entered twice, usually because two systems or two staff members created it independently without searching for an existing record.

Stale data was once correct. A customer’s old address and a product marked in stock that sold out yesterday are the usual examples, and both look perfectly normal on screen. Inconsistent data is one fact written several ways: “Chicago”, “Chicago IL” and “CHI” count as three separate values for one city, which breaks any grouping that assumes one spelling.

Take an electrical-parts distributor whose warehouse updates stock counts once a day while sales staff keep taking orders against that ageing figure. If the system shows twelve units of a part when two remain, the reorder that should have gone out today does not, and a customer order slips. The reverse also happens. An understated count triggers an unnecessary reorder, and the cash tied up in parts that were not needed yet cannot pay a supplier who is genuinely owed.

Collections suffer from duplicates. When one customer’s balance is split across two records, neither entry looks overdue enough to trigger a follow-up call, and the true exposure goes unmanaged. Reporting suffers from labels. “Closed”, “closed” and “COMPLETE” can all mean the same thing, so a filtered count silently leaves out records that belong in it and understates a number managers rely on.

Quality can be measured. Completeness is the share of records missing a required field, and duplication is a count of records that share a phone number under different customer names. Staleness compares each record’s last-updated date with how often that kind of record should realistically change, such as quarterly for supplier terms and daily for stock counts.

None of this needs new software. A handful of spreadsheet formulas or simple queries against your existing systems can give each pattern a rough score, and running them monthly shows a worsening trend early.

Cleaning a batch once helps briefly. Someone must be accountable for both the cleanup and the entry point that let the bad data in, and these are different jobs. Correction is a project. Prevention is a standing responsibility, best assigned to whoever owns the process where the data is first typed in, such as the sales desk for customer records.

Prevention is simpler than it sounds. A required field, a status dropdown instead of free text and a duplicate warning when a new customer closely matches an existing one cost far less than repeated cleanups. They also stop the same error pattern from returning every few months, which is how a finished cleanup quietly undoes itself while nobody is watching the entry screens.

The cost is hard to calculate. The damage is spread across many small errors instead of arriving as one visible bill, which is partly why fixing it is so easy to postpone. Gartner’s survey put the average annual cost of poor data quality at roughly $12.9 million, a figure from large enterprises that will not transfer to small firms. The direction still holds.

Track fixes rather than estimating totals. Ask whether reorder-timing errors dropped after the stale stock data was corrected, and whether overdue collections improved after duplicate customer records were merged. Measured after-effects are more reliable than an estimated starting cost, and they give whoever approves the next data-quality project a result to read instead of a guess.