
Data Hygiene Best Practices: 12 Questions
On this page
- How fast does CRM data go stale?
- How do I find and merge duplicates in HubSpot?
- How do Salesforce duplicate rules and matching rules work?
- How do I find duplicates in Google Sheets?
- What should be required, and what can be free text?
- How do I stop Facebook, facebook, and FB becoming separate rows?
- Who owns data quality in a 10-person company?
- How do I validate emails and phones at entry?
- How do I clean old CRM data without breaking reports?
- How do I handle a GDPR erasure request across the CRM and GA4?
- How often should I audit, and what exactly do I check?
- What do I do when GA4 and the CRM disagree?
Analytics & Data
Data hygiene best practices for a small team are to name one owner, use picklists over free text, deduplicate on a normalized key, validate at entry, and audit monthly. The reason to start now: in January 2026, 20.6% of US wage and salary workers had been with their current employer for 12 months or less, according to the Bureau of Labor Statistics Employee Tenure release. The figure comes from a Current Population Survey supplement covering about 60,000 eligible households.
Our inference is that a business contact list built more than a year ago carries a large share of changed jobs, although tenure is a rough proxy, not a measured contact decay rate. Vendor blogs attribute a 2.1% monthly and 22.5% annual decay figure to MarketingSherpa, but we could not trace the original study, sample, or method. Our arithmetic shows that the annual figure follows from compounding the monthly figure: 1 - (1 - 0.021)^12 = about 0.225. Sources were read on 29 October 2026.
How fast does CRM data go stale?
The BLS tenure share is a rough proxy for job changes, not a B2B data decay rate. Our inference is that older contact lists deserve regular checks because some people will have changed employer, title, or work email.
Say a list contains 2,000 contacts and, as a hypothetical assumption, 20.6% changed employer over a year. That would put about 412 records at risk of being out of date. The BLS figure does not establish that those 412 contacts actually moved: it also includes people entering employment and does not measure your list.
Start with bounced work emails, contacts with no activity in 12 months, and job titles that need confirmation. A bounce is a reason to investigate, not proof that a person left. Record the check date so the next person knows whether a blank or old title has already been reviewed.
How do I find and merge duplicates in HubSpot?
HubSpot automatically deduplicates contacts by the Email property: form submissions and imports with a matching email update the existing contact. Companies are matched on their primary Company domain name, except companies created through the API. HubSpot documents the automatic rules.
Our inference is that one person submitted with a work email and a personal email can become two contacts because the email values differ. HubSpot's duplicates manager offers a separate review: its default contact criteria are First Name, Last Name, Email address, IP country, Phone number, Zip Code, and Company Name.
Managing individual duplicate pairs requires a Professional or Enterprise subscription and displays up to 10,000 pairs. Bulk management requires Data Hub Professional or Enterprise, with limits of 30,000 and 100,000 pairs respectively. Rejected pairs can be restored within 14 days; the duplicates manager documentation also describes the required permissions.
Merged records cannot be reverted. Salesforce merges also cannot be undone, according to secondary sources. Export the records you intend to merge and decide which record should survive before you start.
How do Salesforce duplicate rules and matching rules work?
Salesforce provides three standard matching rules for business accounts, contacts, and leads. Standard duplicate rules are also available for orgs created since the Spring '15 release, and a rule can block creation or allow it with an alert, according to Salesforce Trailhead.
Matching rules decide whether records look alike; duplicate rules decide what happens when a match is found. Matching rules can use fuzzy matching, for example on first names, so Margaret Chan and Maggie Chan are flagged as possible duplicates.
Secondary sources report a limit of three records per merge and say a Salesforce merge cannot be undone. Treat a suggested match as a review queue, not an instruction to merge: two people at the same company can share a surname, and a shared inbox can belong to more than one person.
How do I find duplicates in Google Sheets?
Create a normalized email key, then flag matching keys with COUNTIF. Keep the original rows for review before deleting or merging anything.
Assume email addresses are in column B and lead sources are in column C, starting on row 2. Put these formulas in the indicated cells and fill the first three down. The duplicate flag returns TRUE on every copy, so review the group and choose the record to keep.
E2: =LOWER(TRIM(B2))
F2: =COUNTIF($E$2:$E, E2)>1
G2: =REGEXMATCH(B2, "^[^@\s]+@[^@\s]+\.[^@\s]+$")
=UNIQUE(C2:C)
The shape check spots some malformed addresses; it cannot tell whether a mailbox exists. The UNIQUE list exposes source spellings for a mapping exercise. Empty email keys need separate attention because multiple blank cells can also be flagged; a blank is not evidence that the rows belong to the same person.
What should be required, and what can be free text?
Require a field when a report or decision depends on it and the person entering the record can know the answer at that moment. Lead source, owner, lifecycle stage, and country can meet that test, but only where the team has a usable answer during entry.
A required field that nobody can answer invites n/a or x. Those values are worse than a blank because they appear complete and can contaminate a report. If source is genuinely unknown, give the picklist an explicit unknown value and review how often it is chosen.
Our opinion is to reserve free text for notes: picklists make the values used in counts consistent, while notes preserve context that does not fit a fixed set. Review a proposed required field against a real decision before adding it. If nobody can name the decision, leave the field optional.
How do I stop Facebook, facebook, and FB becoming separate rows?
Use a lead source picklist that mirrors the source and medium words in your UTM vocabulary in readable channel names, such as Paid Social and Email, with Unknown for genuinely unknown sources. UTM values are case sensitive, so Facebook and facebook can appear as separate source rows even though channel definitions are not case sensitive.

Three spellings produce three rows until the lead source dropdown uses one value. Match that value to the lowercase UTM vocabulary used in shared links.
Use lowercase values and hyphens within values. Keep utm_medium to email, paid_social, social, referral, and cpc only outside Google Ads; retain the underscore in paid_social. Google says to always use utm_source, utm_medium, and utm_campaign in its URL builder guidance. For Google Ads, use auto-tagging with no manual UTMs.
Give one person ownership of the shared link sheet so new values get checked before they spread. The measurement plan has the full UTM convention and link sheet approach. In HubSpot, keep the system-set Original Traffic Source intact and place a separate “how did you hear about us” picklist beside it.
Who owns data quality in a 10-person company?
One named person should own the CRM schema: fields, picklists, required rules, and merge decisions. Everyone who creates records owns the accuracy of those records.
That division is our opinion for a small company, where the operations or sales lead can usually make the schema decisions without forming a committee. Give the owner a 30-minute monthly slot and the authority to question a new field before it becomes another column nobody trusts.
The owner should maintain a short field list that states what each picklist value means and who can change it. For site events, the business tracking plan provides an owner column per event, so a CRM field owner and an event owner can coordinate when a form changes.
How do I validate emails and phones at entry?
An HTML email input checks the shape of an address, not whether the mailbox exists. The WHATWG input specification calls its valid-email definition a willful violation of RFC 5322.
Use input type="email" to catch malformed entries, then use a confirmation email or verification service to establish whether the mailbox receives mail. Do not label an address “verified” merely because it passed the browser check.
Store phone numbers in E.164 form: a leading +, country code, and no more than 15 digits. A library such as libphonenumber can parse and format numbers when the country context is available. Iron Goo's Foundation service includes a Next.js site rebuild where forms can be set up to pass clean fields.
How do I clean old CRM data without breaking reports?
Snapshot the monthly figures you report before changing records, and log the cleanup date. Map old free-text values to a new picklist in a lookup table, backfill a new field, and keep the old field read-only until reports have moved.
For example, a hypothetical lookup might map Facebook, facebook, and FB to facebook, while leaving genuinely ambiguous entries for review. Export the current dashboard figures before applying that mapping, including the period and filters used for each number.
Our inference is that merges can move activity to the surviving record and deletes can remove records from past reports, changing counts that someone already presented. Keep a dated record of the mapping, field changes, and report migration; the measurement change log gives that history a place to live. Compare the snapshot with the revised report before retiring the old field.
How do I handle a GDPR erasure request across the CRM and GA4?
Route the request to every system that holds the person, because a CRM deletion does not automatically delete GA4 or spreadsheet data. GDPR Article 12(3) generally requires a response within one month; complex or numerous requests can take two further months if you tell the person the reasons within the first month.
Article 17(3) has exceptions for legal obligations and for establishing, exercising, or defending legal claims. An invoice that must be retained for tax can remain under the applicable obligation. Check Article 12 and Article 17; this is not legal advice.

One request branches to each system holding the person's data. Each system has its own deletion tool and timing.
| System | Tool | Scope | Timing |
|---|---|---|---|
| HubSpot | Permanent delete on each contact | One contact per action | Purge can take up to 30 days |
| GA4 person | Admin API properties.submitUserDeletion | One supplied identifier | No processing time stated on the API page |
| GA4 parameters | Data deletion requests | Selected parameters or user properties | Seven-day cancellation period; processing takes 7 to 63 days |
In HubSpot, permanent deletion cannot run in bulk or through a workflow. The email cannot later be re-added through the UI or import, anonymized analytics remain, and blog comments need manual deletion. See HubSpot's permanent deletion instructions.
For a person in GA4, the Admin API deletion request takes exactly one of userId, clientId, appInstanceId, or userProvidedData containing one email or phone number. Normalize email to lowercase without spaces, removing periods before @ for gmail.com and googlemail.com; normalize phones to digits with a leading +. The response's deletionRequestTime marks the point before which the visitor's data should be deleted.
Our inference is that you need a way to map the requester to an identifier GA4 holds. Storing a client ID or user ID on the CRM record can provide that connection, subject to consent handling; collected user-provided data is another possible identifier.
GA4 Data deletion requests address parameters, such as an email leaked into a URL, rather than a person. They require the Editor role, apply to data more than 12 days old, can be canceled within seven days, and cannot be reversed after deletion.
Use a short routing checklist for each request:
- Check the CRM and record the permanent deletion action or applicable exception.
- Check GA4 for a usable identifier and submit the appropriate request.
- Check the email tool and any spreadsheets holding the person's data.
- Log the receipt date, systems checked, and completion dates against a ticket ID, with no other personal data.
How often should I audit, and what exactly do I check?
Our rule is a 30-minute monthly audit, a deeper quarterly check, and another check after every form, import, or integration change. Each check should produce a short list of records or values for an owner to resolve.
Monthly, inspect duplicate normalized emails, picklist variants, blank required fields, and newly created properties. Quarterly, review bounces, contacts with no activity in 12 months, and unused fields. After a change, submit or inspect a test record and confirm that its values land in the intended fields.
Run these queries against a contacts table with email and lead_source columns:
SELECT LOWER(TRIM(email)) AS email_key, COUNT(*) AS n
FROM contacts
WHERE email IS NOT NULL AND TRIM(email) <> ''
GROUP BY LOWER(TRIM(email))
HAVING COUNT(*) > 1
ORDER BY n DESC;
SELECT lead_source, COUNT(*) AS n
FROM contacts
GROUP BY lead_source
ORDER BY n DESC;
The first output lists normalized email keys occurring more than once, with their counts. The second lists each stored lead source and its count, so Facebook, facebook, FB, a blank, and NULL appear as separate rows.
What do I do when GA4 and the CRM disagree?
The CRM is the count of record for leads and customers; GA4 describes site behavior. A GA4 key event count and the CRM count need not match.
Start by checking that each number answers the question being asked: a CRM lead count is a count of records, while a GA4 key event count is a count of recorded events. The guide to reading data without fooling yourself describes the failure of treating two tools' different answers as though they measure the same thing.
Put your next steps in order
Build on what you have just read with a free ebook on semantic SEO and AI visibility. Get worked examples, decision rules, checklists and a do-this-week action in every chapter, plus a six-week plan for your site.


