Ask a sales leader how clean their CRM is and you’ll get a feeling. Ask for a number and you’ll usually get a completeness score, the share of fields with something in them. That number is easy to produce and easy to defend in a meeting. It says nothing about whether the something is right.
A contact can have a job title, a phone number and an email address, all filled in, all from two jobs ago. The dashboard scores that record as healthy. Your sequence bounces off it.
The case for why CRM data goes bad is made in our look at the reality of dirty CRM data. This piece is the measurement method: how to put a number on it that you can defend.
What is CRM data accuracy, and how is it different from completeness?
CRM data accuracy is the share of checked fields that match the real world. Completeness is the share that are filled in. The two get reported together, which is how they get confused. Validity’s State of CRM Data Management in 2025, a survey of 602 CRM users and stakeholders published in July 2025, found 76% said less than half of their organization’s CRM data is accurate and complete.
Native tooling is built around the questions a database can answer about itself. HubSpot’s data quality overview, per its Knowledge Base article on data quality tools (last updated July 20, 2026), summarizes data issues and recommended actions, manages contact and company duplicates, resolves formatting issues and charts property issues by type: Duplicates, No data and Unused. Those are useful signals about whether data exists, repeats and gets used. Whether a filled field is true is a separate question, and you answer it by checking the record against the world.
Completeness and accuracy, side by side
| Completeness | Accuracy | |
|---|---|---|
| What it asks | Is the field filled in? | Does the field match the real world? |
| How you measure it | Count blanks across the database | Check a random sample against an outside source |
| Example failure | Job title is blank | Job title is filled in with a role the person left |
| Where the number comes from | Fill-rate reports inside the CRM | A sample audit like the one below |
Step 1: pick 100 records at random
Export your active contacts into a spreadsheet, and your open deals as a separate file if you forecast from the CRM. Add a column with =RAND(), paste it as values so it stops recalculating, sort by it and take the top 100 rows. Skip the shortcut of grabbing the first page of a saved view. Saved views usually sort by created date or last activity, which favors recent or recently edited records, and that sample flatters the database.
The 100 has precedent. A 2017 Harvard Business Review article by Tadhg Nagle, Thomas C. Redman and David Sammon, headlined “Only 3% of Companies’ Data Meets Basic Quality Standards,” suggests the same size of exercise: assemble 100 data records completed by your department and work through each.
Run contacts and deals as separate samples. They fail in different ways, and blending them hides which one is creating the drag.
Step 2: which fields should you check?
Check four or five fields, chosen by what they feed. Routing runs on owner, territory and company. Outreach runs on email, phone and job title. The forecast runs on deal amount, close date and stage. Pick the fields your team leans on hardest this quarter and leave the rest of the record alone. Checking the whole record turns an afternoon into a week.
If your CRM carries several versions of the same field, check the one your reports read from, and see our piece on CRM field bloat for why there are several. If an enrichment tool writes your company field, put it in the sample, because enrichment can write the wrong data with total confidence, as covered in when AI enrichment writes the wrong company.
Two of these fields go stale while sitting untouched.
Email. ZeroBounce’s Email List Decay Report for 2026, based on the email data it verified from January to December 2025, found only 62% of emails submitted to it in 2025 were valid, and at least 23% of an email list decays within one year. ZeroBounce sells email verification, and those are email lists sent to a verification vendor, not CRM records in general. Read it as a reason to put email in the sample, not as your rate.
Job title. Lusha measured 140,284 US sales leaders at VP and C-suite level and found 12.25% changed roles over twelve months and 25.67% over twenty-four (published September 12, 2026). That covers senior US sales leaders and role changes only, measured by a data vendor. Your contacts may move faster or slower. Your sample will tell you which.
Lusha found 12.25% of senior US sales leaders changed roles over twelve months.
Source: Lusha, B2B Data Decay Rate: We Measured, 2026 · US VP and C-suite sales leaders, cohort of 140,284, role changes only
| Period | Changed roles |
|---|---|
| 12 mo | 12.25% |
| 24 mo | 25.67% |
Step 3: verify each field against an outside source
A field passes only when a source outside the CRM confirms it. Check title and company against the person’s current LinkedIn profile or the company’s own site. Check email with an email verification check or the record’s bounce history. Check deal amount against the signed contract or the invoice. A blank field fails for accuracy, and you log it separately as incomplete, so you finish with both numbers instead of one blurred one.
Score each record pass or fail. A record passes only if each field you checked on it passes. Keep a per-field failure count alongside the record score, because Step 5 runs on those counts.
Example scoring sheet (layout only, not real results)
| Record | Title | Company | Phone | Owner | Record result | |
|---|---|---|---|---|---|---|
| Row 14 | Pass | Pass | Fail (bounced) | Pass | Pass | Fail |
| Row 37 | Pass | Pass | Pass | Pass | Pass | Pass |
| Row 52 | Blank (log as incomplete) | Pass | Pass | Pass | Pass | Fail |
| Row 88 | Fail (left the role) | Fail (new employer) | Pass | Pass | Pass | Fail |
Note row 88. One job change took out two fields, and the sequence keeps writing to someone at the old company.
Step 4: how accurate is your CRM, give or take?
Divide passing records by records checked, then report a 95% confidence interval with the result, so the room doesn’t read it as exact. NIST’s Engineering Statistics Handbook, section 7.2.4.1, describes the Wilson method for confidence intervals on proportions, notes that its lower limit can’t go negative, and warns that normal-approximation limits may not be accurate enough when the sample is very small or the failures are very few.
Worked example, with made-up numbers: say 62 of your 100 records pass. Your sample accuracy is 62%, and the 95% Wilson interval runs from about 52% to about 71%. You’d report it as “62% (95% interval roughly 52% to 71%)”. Those inputs are hypothetical, chosen to show the arithmetic, and they aren’t a finding from any company or study.
For your own count, run your passing records and your sample size through a Wilson score interval calculator, or work it by hand from the method on the NIST page. Either way, write the range down next to the rate before anyone sees the number.
The interval is the honest part. A 62 that could sit anywhere from the low 50s to the low 70s tells you there’s a problem worth working on. It doesn’t tell you whether last quarter’s cleanup moved anything by a few points. If you need a tighter range, check more records.
Step 5: rank the failures by what they break
A wrong email costs you a bounced sequence. A wrong deal amount moves the forecast. Sort the per-field failure counts by the process each field feeds, and start with the forecast and routing fields, because those errors reach the pipeline review and the territory map before anyone checks them. Outreach errors cost a send. Forecast errors cost credibility in the room where the number gets committed.
If the same person shows up twice in your sample, count it as a finding, then read why HubSpot duplicates keep coming back. If two reports still disagree after the cleanup, the cause may sit in how they’re built, which we cover in why HubSpot reports don’t match.
Across the 150+ B2B companies we’ve worked inside, roughly 7 in 10 had material CRM data accuracy problems. The audit tells you whether yours is one of them, and which fields carry the problem.
You can keep reading the completeness dashboard and trusting the green. Or you can spend an afternoon on 100 records and walk into the next pipeline review with an accuracy number and the interval around it.
Start this week: export 100 contacts at random, put the four or five fields you will check across the top of a spreadsheet, and verify the first ten records against an outside source before Friday. The first ten show you how long the other records will take.
If you want the wider view of the database, the free, read-only AeolusGTM CRM Diagnostic runs 40+ automated checks across data quality, pipeline hygiene, freshness, reachability and ICP fit, on HubSpot, Salesforce or Pipedrive.