Working out what data to capture, in sales for example, and keeping it clean
Decide the questions you want answered, then capture only the fields that answer them. Clean data comes from fewer fields, filled at the moment the fact exists, with a check that runs every week instead of a clean-up every year.

Start from questions, not fields
Most CRMs have 60 fields because every system vendor and every previous sales manager added a few. The useful ones are the dozen or so that answer the questions your business actually asks. So write the questions down first:
| Question | Fields that answer it |
|---|---|
| Which accounts deserve time this month? | Fit score, tier, last contact date, open opportunity. |
| Who do we need to reach there? | Role, person, whether the contact details are verified, and when. |
| Why did we win or lose? | Outcome, one reason from a short list, competitor if known. |
| What did we promise? | Next action, owner, date. |
| Where is the pipeline really? | Stage, value, expected date, and a date the stage last changed. |
Anything not in the right-hand column is a candidate for deletion. In our experience, removing fields does more for data quality than any amount of training.
A rule for every field
Each field gets three things written down: what's allowed in it (a short list, a date, a number – never free text if a list will do), who fills it, and when. "When" is the one that matters. A field filled at the moment the fact is known, on the phone in the car park after the visit, is right. The same field filled on Friday afternoon from memory is a guess.
How data gets dirty
Four ways, and all of them predictable. Free-text fields, where "Melbourne", "Melb" and "VIC" are three different places. Duplicates, from imports and from people creating a record instead of searching first. Fields nobody uses, which fill up with rubbish because nobody looks. And time – people move jobs, companies move, and a record that was right in March is wrong by September with nothing in the CRM to say so. Keeping contacts current when people move roles is the slow version of this problem.
Where the tools help
They help with the filling and the checking. A call can be summarised into the fields rather than typed up. A reply can be read for a new phone number or a job change and the record updated. A company can be researched against your criteria and scored, with the sources kept. That's the work we do for sales teams, and it takes away the part of data quality that depends on a busy person's discipline.
They don't help with the deciding. Which questions matter, which fields answer them, what counts as a good fit – those are yours. A tool pointed at a CRM with 60 undefined fields will fill all 60 with confident nonsense.
The weekly check
A short report, once a week, by team and by rep: records missing a required field, possible duplicates, records untouched for 90 days with an open opportunity. One number per rep. It takes 10 minutes to read, and it means the annual clean-up never gets needed, because nothing gets to 90 days old without someone seeing it.