Improve CRM Data Quality
A rep pulls up a renewal call list on Monday and finds the same buyer three times. One record says Maria Gonzalez at Isla Verde Dental, another says M. Gonzalez at Isla Verde Dental Group, and the third has no company at all, only an email address with a typo in the domain. Two reps called her last quarter and neither knew about the other. She was annoyed, and she was right to be.
Nobody planned that mess. It arrived one reasonable shortcut at a time. If you work in sales operations, you already know that "bad CRM data" is not one problem. It is five or six smaller problems that look identical from the outside, and each one has a different fix. This article separates them, shows where to spend effort first, and gives you a plan you can start without buying anything.
Start by naming what "bad" means in your CRM
Before you clean anything, write down the kinds of damage you actually have. Most CRMs suffer from the same handful.
| Kind of damage | What it looks like | Where it usually comes from |
|---|---|---|
| Missing values | Blank phone, no industry, no owner | Optional fields, quick lead capture |
| Invalid values | "n/a" in an email field, a phone with 7 digits | Free-text fields, imports |
| Duplicates | Same person or company in several records | Web forms, imports, rep-created contacts |
| Inconsistent values | "PR", "P.R.", "Puerto Rico", "Puerto rico" | Free-text fields, no controlled list |
| Stale values | A contact who left the company a year ago | Time, job changes, mergers |
| Misused stages | Deals stuck in "Negotiation" for 400 days | Vague definitions, no exit rules |
Pull a sample of 200 records, chosen at random rather than the ones everyone already knows are bad, and tag each one with the damage you see. Ten minutes of tagging tells you more than any vendor scorecard. If 40 percent of your problems are duplicates, you do not need an enrichment tool yet. If most are stale values, deduplication will not move the needle.
Stop the leak first, validate at entry
Cleaning a CRM while records keep arriving dirty is like mopping while the tap runs. The cheapest correction is the one made at the moment someone types the value, because the person still remembers the answer.
Required fields that mean something
Make a field required only when a downstream process breaks without it. A deal without an amount cannot be forecast, so amount is required at the stage where forecasting begins, not at creation. Requiring everything at creation just teaches reps to type junk to get past the form. Stage-gated requirements work better. Each stage asks for the few things that stage needs.
Controlled lists instead of free text
Replace free-text fields with picklists wherever the answer set is small and stable. Country, state, industry, lead source, and lost reason are the usual candidates. Free text is where "PR" turns into four spellings, and reports that group by those values quietly lie.
Format and shape checks
Most CRMs support validation rules. Useful ones include a phone number that normalizes to a single format such as E.164 (a standard way of writing numbers with the country code, for example +1 787 555 0123), an email that matches a basic pattern and does not end in a known typo like gmial.com, and a postal code that matches the country. Keep the error messages human. "Phone must include area code" works. "Validation error 14" does not.
Validate at the door, not only inside the CRM
Web forms, imports and integrations bypass the screens your reps use. If a form posts directly to the CRM through an API, the rules need to run there too, or in the layer that sits in front of it. This is also where a lot of duplicates are born, so check for an existing match before creating a new record. We come back to that below.
Deduplication without destroying good records
Duplicates are the failure people notice first, and the one most often handled badly. A hasty bulk merge can destroy data you cannot get back.
Pick matching keys in order of trust
A matching key is the value you use to decide that two records describe the same thing. Rank yours.
- A system identifier, such as a customer number shared with billing.
- A normalized email address, lowercased and trimmed.
- A normalized phone number.
- A fuzzy match on name plus company or postal code.
Exact matches on the first three can merge automatically in most cases. Fuzzy matches, where "Gonzalez" and "Gonzales" score as similar, should go to a human review queue. Fuzzy matching is useful, and it is also how two different people named Jose Rivera at different companies end up merged by mistake.
Decide survivorship before you merge
When two records merge, one value has to win for each field. This is called a survivorship rule. Write the rules down.
- The most recently verified value wins for email and phone.
- The oldest created date is kept, since it records when the relationship began.
- Owner stays with whoever has the open opportunity.
- Notes and activities from both records are preserved, never discarded.
Without survivorship rules, the outcome depends on whichever record the person clicking happened to open first.
Account for relationships and history
Merging two contacts also means moving their deals, tasks and email history. Merging two companies means checking whether their contacts overlap. Test your CRM's merge on a copy or a sandbox first, and export the records you are about to merge so you can reverse a mistake. Merges in most CRMs cannot be undone.
Prevent the next round
Turn on the CRM's built-in duplicate detection at creation if it has one. For forms and imports, look up by email or customer number first and update the existing record instead of creating a new one. That update-or-create step is sometimes called an upsert, and it is the single most effective deduplication control you can add.
Enrichment, and when it helps
Enrichment means filling in fields from an outside source, such as company size, industry or a verified work email. It is helpful when your reps spend time researching things a service could supply. It is wasteful when the underlying records are duplicates or fakes, because you will pay to enrich garbage.
Some practical rules.
- Clean and deduplicate first, then enrich.
- Enrich only the fields someone uses. If no report or routing rule reads employee count, do not buy it.
- Never let enrichment overwrite a value a human verified. Mark the source and date of each enriched field, and give human entry higher priority in your survivorship rules.
- Check what the vendor's data terms say about storing and reusing the data after you cancel.
- Privacy and marketing consent rules differ by jurisdiction and by whom you contact. This is not legal advice, so have counsel review how you collect and use contact data.
Ask any enrichment vendor how they source the data, how often they refresh it, what the match rate was on a sample of your own records, and whether you can export what you paid for. A trial on 500 of your own records is more honest than any brochure.
Stage definitions are data quality too
Pipeline reports are only as good as the stage a deal sits in, and stage is a judgment call. If "Proposal Sent" means different things to different reps, the forecast will drift no matter how clean the email addresses are.
Write each stage as an observable event rather than a feeling.
| Stage | Entry condition | Exit evidence |
|---|---|---|
| Qualified | Budget owner identified, need confirmed in writing | Discovery call logged |
| Proposal sent | Quote document attached and delivered | Buyer acknowledged receipt |
| Negotiation | Buyer has asked for changes to price or terms | Revised quote or signature |
| Closed lost | Buyer declined or no response after 3 follow-ups over 30 days | Lost reason selected |
Add an aging rule. A deal that sits in one stage for twice its normal duration gets flagged to its owner. Add a required lost reason from a picklist, so that your "why we lose" report is something other than a pile of free text.
Data decay is a schedule, not a surprise
Contact data decays because people change jobs, companies close, and phone numbers get recycled. I will not quote a decay percentage, because it varies by industry and by how the data was collected, and you should measure your own. A simple way is to send a re-engagement email to a sample of 300 contacts and count the bounces and the "I no longer work there" replies. That is your decay rate for the segment.
Then build a refresh rhythm.
- Contacts tied to open deals get checked each time the deal moves a stage.
- Customers get a light verification at each renewal.
- Cold contacts get an annual pass, or an archive flag if they have had no activity for a set period.
- Hard bounces get flagged on the record automatically so no one emails the address again.
Archiving is better than deleting. Keep the record, mark it inactive, and exclude it from active reports.
A worked example, with illustrative numbers
This example is hypothetical. A distributor with five reps has 8,000 contact records. A random sample of 200 shows the following.
- 22 records are duplicates of another record, which is 11 percent.
- 30 have no valid phone, which is 15 percent.
- 18 have a bounced or obviously wrong email, which is 9 percent.
- 40 have an inconsistent state or industry value, which is 20 percent.
Scaled to the full database, roughly 880 contacts are duplicates and 1,600 have inconsistent values. The operations lead decides on this order.
- Turn the state and industry fields into picklists and normalize existing values with one scripted update. This fixes the 20 percent in a day and prevents recurrence.
- Add an email lookup to the web form so new leads update existing records.
- Run a deduplication pass with exact email matches merged automatically and fuzzy matches sent to a review queue handled for one hour a week.
- Add phone formatting validation.
- Only then consider enrichment.
The numbers are made up, but the sequence is the point. Cheap and preventive fixes come first, and spend on outside data comes last.
Stewardship, so it does not drift back
Data quality is a job with an owner. Without one, every cleaning project lasts about a quarter.
Name a data steward
This can be a part-time role held by someone in sales operations. The steward owns the field dictionary, approves new fields, reviews the merge queue, and publishes a short quality report monthly. If every rep can add a custom field, the CRM will have 300 fields in a year and nobody will know what half of them mean.
Keep a field dictionary
A shared document or sheet lists each field, what it means, who may edit it, how it is filled, and which reports depend on it. It sounds dull, and it saves hours of arguments about what "Last Contact Date" is supposed to mean.
Measure a handful of numbers
Track a short set of quality indicators over time.
- Percent of active records with a valid email and phone.
- Percent of open deals with an owner, amount and close date.
- Duplicate rate on a monthly sample.
- Deals past their stage aging threshold.
- Bounce rate on outbound email.
Publish them where reps can see them. People fix what they can see.
Make quality part of how reps work
Reps will keep records clean when it costs them less than ignoring them. Give them one-click actions, pre-filled forms, and a call list that is accurate, and they will notice the difference. Punishing messy records without removing the friction that causes them rarely works.
Failure modes to expect
- Over-strict validation drives reps to enter fake values to get past the screen. Test rules with real reps before turning them on.
- A one-time cleanup with no prevention returns to the original mess within months.
- Automatic merges on weak keys combine different people and quietly damage relationships.
- Enrichment from a second source overwrites verified data, and nobody notices until a deal goes to the wrong contact.
- Integrations create records the CRM would reject through the screen, so check each connected tool.
What to ask before you pick a tool
If you are looking at data quality software or a CRM add-on, ask these questions.
- Does it show me what it will change before it changes it, and can I reverse it?
- Can I define my own survivorship rules?
- How does it handle matches it is unsure about?
- Does it work on records created through the API and through forms, or only through the user interface?
- What happens to my data if I leave?
What to do on Monday morning
- Pull a random sample of 200 records and tag the damage.
- Pick the single most common problem and write down the fix, either a rule, a picklist or a merge pass.
- Export a backup of the records you plan to touch.
- Write your stage definitions in a table like the one above and share them with the team.
- Name a steward and book 30 minutes weekly for the review queue.
You do not need a platform to begin. You need a sample, a few rules, and one person who owns them. If your CRM is feeding other systems such as billing or reporting, and you want a second pair of eyes on how records flow between them, Ozcorp InfraOps does this kind of work for small teams in Puerto Rico and the United States, and I am glad to talk it through with you. If you would rather just compare notes, send me your messiest field and I will tell you what I would try first.
Discuss your infrastructure requirements
Review the current service offer or schedule a conversation about the scope you need.
Stay in touch
Follow Ozcorp
Leave your details for occasional notes about new articles and Ozcorp updates.