The Search Box Is the Catalogue: A Two-Minute Test for UK Data Teams

A data catalogue can be complete, well-governed and almost useless. The tell is simple: an analyst cannot find the right dataset unless they already know its name. They know the business question. They know the report they need to build. But the catalogue search box returns a list of tables with technical names, stewardship fields and a description that repeats the table name.

This is not a failure of governance. It is a failure of discovery. Many UK organisations have treated the catalogue as an artefact to be populated for audit, rather than a tool to be used every day. The result is a library with no index, or an index written in a language that only the librarian understands.

The Filing Cabinet Problem

Catalogues often grow from the inside out. A data governance team defines a maturity model, selects a tool and begins collecting metadata. Business glossaries are drafted. Ownership is assigned. Sensitivity classifications are applied. These are worthwhile activities. But they can produce a catalogue that answers the question what do we have? rather than the question what should I use?

The search box is where this becomes visible. If search only works when the user types a table name, the catalogue is functioning as a filing cabinet. A filing cabinet is useful if you know the file you want. It is poor at helping you decide which file matters.

A Two-Minute Test

Choose an analyst who is not a data steward and who does not know the target dataset. Give them a business question, not a dataset name. For example: which customers are likely to renew late? Or which suppliers have repeated quality failures? Ask them to use the catalogue search box to find a dataset they would trust enough to start an analysis.

Set a two-minute limit. Observe four things:

• Did the first search use business language rather than technical language?
• Did the top three results include a plausible dataset?
• Was the owner or contact visible without opening a separate governance page?
• Could the analyst judge whether the dataset was fit for the question?

If the analyst needs a steward to translate, the catalogue is not yet a discovery tool. If they find a dataset but cannot tell whether it is current, complete or certified, the search result is only half a result. If they find nothing and give up, that is the most useful finding of the day.

Run the test with three people and three questions. Do not coach them. Do not accept I would ask someone as a pass. Asking someone is a valid fallback in a small team, but it is not a scalable discovery model.

What the Test Reveals

The test usually exposes one of three problems. The first is search relevance. The catalogue may index technical table names and column names but not business terms, synonyms or common abbreviations. A user searching for churn may not find a table called customer_attrition_score. A user searching for arrears may not find payment_delinquency.

The second is weak ownership signals. A catalogue can list an owner who is a team rather than a person, or a person who left the organisation. Search results need a current contact, a review date and a simple status such as certified, deprecated or experimental. Without those signals, the user cannot tell whether the dataset is safe to use.

The third is metadata written for compliance rather than comprehension. A description that says this table contains customer records tells the user nothing about fitness. A plain-English description says what the data covers, what it excludes, how often it is refreshed and which decisions it has supported. That is not dumbing down. It is making the metadata useful.

Fix Search Before You Add Fields

When a catalogue is failing at the search box, the temptation is to add more fields. More fields can make the problem worse. Start with search relevance and the first three results. Add synonyms for the business terms people actually use. Boost results that have a current owner and a recent review. Demote deprecated datasets. Make the description visible in the result card, not buried three clicks away.

Then fix ownership. A named person, a backup and a review date are more valuable than a long stewardship form. If no one will accept ownership, the dataset should be marked as unowned and excluded from trusted search results.

Finally, rewrite the top twenty descriptions in plain English. Use the language of the business question, not the language of the schema. Include limitations. A good description helps an analyst decide not to use a dataset as often as it helps them decide to use it.

The Catalogue Is a Service, Not a Shelf

A catalogue earns its place when it helps someone find, judge and use data without a guided tour. That is a service design problem as much as a metadata problem. The search box is the front door. If the front door opens onto a wall of technical names, the rest of the governance work is hidden behind it.

The two-minute test is deliberately modest. It will not solve metadata quality across an estate. But it will show whether the catalogue is being used as a finding aid or a filing cabinet. For UK data governance teams in 2026, that distinction is worth more than another round of field completion.