Data quality is an operations problem, not an IT problem

When reports disagree or a customer record turns out to be wrong, the instinct in most organisations is to call IT. But data quality problems are rarely created in the systems where they are discovered. They are created upstream, at the point where work happens — a field skipped on a form, a vendor set up twice under slightly different names, a workaround that stores the "real" information in a free-text notes field. Treating data quality as a technical clean-up exercise addresses the symptom and leaves the cause untouched.

Where bad data actually comes from

Across operational environments, the same root causes recur:

• Fields without owners. Nobody is accountable for the accuracy of the data, so nobody maintains it.
• Processes that reward speed over accuracy. If a team is measured purely on throughput, optional fields will be skipped and defaults will be accepted.
• Duplicate entry across systems. Every re-keying step is an error opportunity, and reconciliation becomes a permanent tax.
• Workarounds. When a system cannot capture what the business needs, people improvise — and the improvisation never makes it into any downstream report.

Fix the process, not just the record

One-off cleansing projects feel productive, but without process change the data degrades back to its prior state within months. Durable improvement pairs remediation with prevention: validate at the point of entry, remove duplicate keying through integration, redesign the steps that generate errors, and assign explicit ownership for critical data domains such as customer, vendor and product master data.

Make quality measurable

What gets measured gets maintained. A small set of data quality indicators — completeness, uniqueness, validity and timeliness for the fields that matter most — reported alongside operational metrics, changes behaviour quickly. Teams that see their error rates weekly fix root causes; teams that discover them at year-end audit do not. In many finance functions, a substantial share of month-end effort is reconciliation and correction work that exists only because of upstream data quality gaps.

Why this matters more now

Every analytics, automation and AI initiative inherits the quality of the data underneath it. An automation that processes bad data simply produces wrong outcomes faster, and an AI model trained on inconsistent records learns the inconsistency. Organisations that treat data quality as an operational discipline — owned by the business, measured continuously, designed into processes — build on solid ground. The rest keep paying the reconciliation tax.