Data Quality Workbench

Data Quality Workbench

A register is only worth what its worst records are worth. The workbench puts a number on how good your equipment data is, sorts the problems into lists somebody can actually work through, and lets you fix hundreds of rows at once — while still asking a person to decide what the fix should be.

Why a score, rather than a list of complaints

Everyone already knows the register has gaps. What nobody can usually say is how bad, or whether last quarter's clean-up helped, or which fifty records to fix first. A score answers all three: it is comparable between two months, it sorts, and it can be filtered down to one category or one site so a branch manager sees their own number rather than the company's.

It is deliberately a measure of the record, not of the equipment. A perfectly maintained laptop with no serial number recorded scores badly, and it should: from the register's point of view that laptop cannot be told apart from any other of its model.

How the score is worked out

Every asset is marked against a table worth 100 points. What the record has, it earns; what it is missing, it loses. The result is a percentage between 0 and 100.

Five things that take the score to zero

Some problems are not incomplete data, they are data that is wrong. A record that is wrong must not outrank a record that is merely sparse, so these do not subtract points — they set the score to 0 outright.

A zeroed record still reports everything else that is wrong with it, because whoever resolves the duplicate will want to know the row is also missing a model.

Reading the top of the page

The strip across the top is the summary, and it obeys the same category and location filters as everything below it — so a site manager can narrow to their own site and read their own number.

• Average score — the headline, over whatever is currently filtered.
• Live assets and Needs attention — how many records exist, and how many are sitting in at least one queue.
• Score distribution — five bands, so you can see whether you have a few disasters or a broad shallow problem.
• What is actually wrong — the biggest queues, largest first. This is the list to read before deciding what to do this week.

The Assets tab — one queue at a time

Rather than one enormous list of everything that is wrong, the workbench offers a queue per defect. You pick a queue, and you get exactly the assets with that problem — which is what makes a bulk fix safe, because every row in front of you needs the same thing done to it.

Each row shows the tag, the asset, its holder, its score as a coloured chip, and when it was last marked. The chip carries the number and the colour only reinforces it — a chip that said nothing but “red” would be useless to anybody sorting by it, and unreadable to anybody who does not see colour the way you do.

There is an Export button on every queue. It builds the spreadsheet on the server from the same query the table is showing, so what you get is the whole queue and not just the page in front of you — up to the export cap your administrator has set (5,000 rows by default).

Fixing many rows at once

Select rows — or take the whole queue — choose an action, fill in whatever it needs, and the run begins. It is deliberately not instant: the work is done in small batches with a live console, a progress bar and four counters, so you can watch what is happening and stop if it is not what you expected.

What limits a run

• Batch size — how many assets are handled per step. Fifty by default; your administrator can set it between 1 and 200 on the Settings page.
• 5,000 rows is the ceiling on one selection. Beyond that, filter the queue down and run it in passes.
• A prepared selection expires after two hours. Leave the modal open over lunch and the run will tell you the selection has gone rather than acting on a stale list. Start it again.
• The action cannot change mid-run. Your permission was checked against the action you chose when the run was prepared; allowing a later step to do something different would mean acting on a decision nobody was checked for.

The Master data tab

Some problems are not in the asset records at all — they are in the catalogue those records point at. This tab finds two kinds.

Every row shows how many live assets currently point at it, which is the number that decides what you can do:

• Merge — pick which entry survives and everything pointing at the others is moved to it. Capped at 500 assets in one merge; past that, use the bulk Set model action on the assets themselves, which is chunked and resumable.
• Archive — removes the entry from use. It refuses while anything still points at it, because archiving a catalogue entry out from under live assets would leave them classified as nothing.

The Reconciliation tab

This is the exception queue: things the system noticed but must not guess about. When the custody paperwork was first built from the older assignment fields, most rows converted cleanly. A few did not, and each had several plausible readings:

Where to start, if the number is bad

1. Read “what is actually wrong”, not the average. The average tells you there is a problem; the queue list tells you which one.
2. Clear the zeroing problems first. Duplicates and inactive holders are few, and each one is dragging a record from whatever it deserves down to nothing.
3. Then take the biggest single queue and run one bulk action against it. Missing tags and missing models are usually the two largest, and both have a one-step fix.
4. Clean the catalogue on the Master data tab before assigning models in bulk — otherwise you will carefully point five hundred assets at a placeholder.
5. Answer the reconciliation questions last. They are few, they need judgement, and they are the ones an auditor will ask about.
6. Re-score what you touched, and check the number moved.

Who can do what

Related