HubSpot Data Quality: Detection, Automated Fixes, Tier Gates and Accuracy Sampling

A HubSpot administrator can work the data quality overview down to zero open issues and still leave a sales team calling numbers that reach the wrong person. The formatting suggestions are settled, the duplicate pairs are merged or dismissed, and a leader paying for Data Hub reads the falling counts as better data. A job title that was true two years ago, a direct line that now rings a former colleague and an email address for someone who has left are all well-formed values. None of them appears in the overview's counts, and each one is found only when somebody uses it.

This article sets out what each HubSpot data quality tool compares a value with, which fixes each subscription tier automates, and the check that no tier performs. It starts with the difference between a well-formed value and a correct one, then covers detection, automated fixes in tier order, enrichment and a reference table by tier. A sample portal then sets the overview's zero open issues beside an accuracy sample, and the sampling procedure follows. It closes with costs and returns, four symptoms and their mechanisms, and the limits of the evidence.

The data quality overview is the page at Data Management > Data Quality that reports duplicates, formatting issues, enrichment coverage and property insights. Data Hub is the product HubSpot sold as Operations Hub: a footnote to HubSpot's Fall 2025 Spotlight states that "Data Hub replaces Operations Hub". Its editions are set out in HubSpot Data Hub. A value is well-formed when it matches the pattern its property expects, and correct when it describes the person or company it belongs to.

Well-Formed Values and Correct Values

Every HubSpot data quality tool works against other data, and none compares a value with the person or company it describes. The question of whether CRM data is clean runs five properties of a value together: well-formed; unique, with no second record for the same entity; complete, with the field filled; current, still true today; and correct. HubSpot's tools test the first three and neither of the last two.

Formatting checks and validation rules compare a value with a pattern. Duplicate management compares it with other records, and property insights compare a field with its own fill rate and usage. Enrichment is a source rather than a check: it fills a value from HubSpot's commercial dataset. The overview's Data Enrichment tab reports a match rate and per-property coverage, not whether stored values agree with that dataset. The nearest thing to a check against the contact is the hard bounce, which arises only from sending. It covers only the Email field, and it reports what a mail server said rather than what the contact said.

The distinction predicts which errors software can find. An ontological analysis of data quality by Wand and Wang separates a value mapped to a meaningless state, which nobody can map back to the world, from one mapped to a meaningful state that is the wrong one. It names erroneous entry and a failure to record a change in the world as typical origins of both. A pattern can reject jOhN, n/a or the area code 123, which HubSpot's phone validation example marks as invalid in the United States. Nothing in a job title the contact held until last year distinguishes it from a current title.

HubSpot's own guide to maintaining a trusted source of customer data lists five metrics for data quality and team alignment, from record fill rate to SLA miss rate. None measures whether a value is true, although the same guide states that "data decay is inevitable". The Custom code and Send a webhook workflow actions, on Data Hub Professional and Enterprise, can pass a value to an outside verification service. That comparison still reaches a proxy, such as whether a mail server accepts an address, rather than the contact.

Detection in the Data Quality Overview

The overview detects malformed values on five contact properties and likely duplicates among contacts and companies. An overview that flags nothing after a cleanup is not evidence that a fix held. HubSpot's formatting documentation names Email, First Name, Last Name, State/Region and Country, and flags issues such as unexpected capitalisation, placeholder text and swapped or combined names. Each fix is a proposal about form from Breeze, HubSpot's AI, and the documentation allows that a user may disagree with it. One issue listed for State/Region and Country, "the value is wrong", reads like a correctness check. Its example is an incorrect time zone, and the documentation names nothing the value is compared with.

The check is included in Professional and Enterprise editions and open to Starter accounts through a public beta. The product catalog sets the scan at every two weeks for Starter, daily for Professional and real time with Data Hub Professional or Enterprise.

A fix also changes what the overview reports afterwards. The data quality documentation states that a value later rewritten by a manual update, the API, an integration, an import or a data migration "will typically not be flagged again", because the tool "interprets the overwritten format as the correct".

Duplicate management compares seven contact properties and five company properties by default. It scores each candidate pair for similarity and recalculates as records are created, or once a day when none are. Professional and Enterprise editions review pairs one at a time, up to 10,000. Data Hub Professional adds bulk merging, 30,000 pairs (100,000 at Data Hub Enterprise), up to two custom rules per object on up to nine properties, a winning-record beta and an audit log of automated merges. The two possible errors carry unequal costs: a wrongly rejected pair can be restored for 14 days, whereas a merged record cannot be reverted. Merge behaviour field by field is covered in HubSpot duplicate management.

Property insights judge the field rather than the value. The Property Insights tab flags properties as duplicated, empty or unused, refreshed daily, with anomaly alerts at Data Hub Professional. The data model health check scans for 16 issue types on all plans. HubSpot Smart CRM shows how per-property fill rates overstate the share of records carrying every field a feature needs.

Automated Fixes in Tier Order

Every plan can refuse a malformed value at entry, and Data Hub Professional rewrites values at volume. No tier can tell a right value from a wrong one that fits the pattern. The Professional Validate and format phone number action shows where a correct value has to come from: for numbers it finds invalid, the documentation states that "HubSpot cannot automatically correct these issues". It recommends clearing the value or creating a task to ask the contact.

Validation Rules and Deduplication on All Plans

Validation rules are available on all plans for text, number, date and phone number properties, including up to ten unique-value properties per object, and regular expression rules need Professional. The rules apply in the CRM interface, on import and in forms built in the updated editor. They are not enforced in workflows, chatflows, meeting scheduling pages or legacy forms, and existing values are not reformatted retroactively. The property validations API guide states that API writes are validated from the 2026-09 API version, released on 8 September 2026. Its changelog leaves property validation rules out of the three behaviours it enforces, so one test write settles whether a rule binds a given integration.

Phone validation requires the E.164 format: a plus sign, a valid country and area code, and between 4 and 15 digits. Automatic deduplication matches contacts on email address and companies on primary domain. Companies created through the API, including by third-party sync apps, are not deduplicated by domain.

Formatting Automation in Data Hub

The data quality documentation allows bulk acceptance of formatting suggestions at Data Hub Starter. The formatting documentation reserves bulk management for Data Hub Professional, a conflict that step 2 of the procedure resolves. Data Hub Professional adds automation rules that accept suggestions for current and new records, with a Changes to records tab listing every value rewritten. It also adds the Format data workflow action, which transforms values through options such as title case or a custom formula.

The two tools apply different rules to the same surname. The formatting documentation gives "macdonald would be updated to MacDonald" as an automated fix and lists "Desantos instead of DeSantos" as an issue. The Format data option Change to title case lowercases every letter after the first in each word, so it writes Macdonald and Desantos. Whichever spelling a contact uses, one of the two tools writes a well-formed wrong value, and only the contact can settle which is correct.

Enrichment and Its Overwrite Rules

Enrichment is the one data quality tool that draws on data from outside the portal, and in automatic mode, by its own documentation, it leaves a wrong entered value in place. Enrichment, from Starter, fills properties such as Job Title and Industry from HubSpot's commercial dataset, for contacts with a business email address and companies with a domain. Automatic enrichment fills empty values and does not overwrite a value set by a user or another system, whereas manual enrichment can overwrite existing values. Continuous enrichment refreshes values that enrichment itself filled, monthly according to the enrichment settings, and stops for a property once anyone edits it. The same settings article states that the Enrich record workflow action can overwrite properties regardless of the configured overwrite rules. The catalog's credits rate sheet lists no credit charge for enrichment as of October 2026.

The settings article also offers a per-property rule to overwrite existing values, suggested for Job Title. It names only conversational enrichment and the workflow action as ignoring the rule, which implies, against the enrichment article, that the rule also governs automatic enrichment; one test record settles which applies. Any overwrite that replaces a wrong title also replaces correct titles a person typed, and HubSpot's enrichment documentation states no accuracy rate for the dataset.

HubSpot Data Quality Tools by Tier

Of the twelve capabilities HubSpot documents, eight need no more than a Professional edition and four need Data Hub. No tier checks a value against the person or company it describes.

Overview Counts Against an Accuracy Sample

On a sample portal whose overview shows no open issues, an outside check of 200 contacts estimates that about 3,000 of its 19,460 contacts hold a wrong value in one of three fields. Every figure in this section is sample data, chosen to show the arithmetic; none comes from a client or a measured portal. The portal runs Sales Hub Professional with 20,000 contacts. Its overview opens with 1,400 contacts carrying formatting issues and 600 duplicate pairs. An administrator accepts 1,330 suggestions, rejects 70 that misread names, merges 540 pairs and rejects 60, which leaves 19,460 contacts and no open issues.

The administrator then draws 200 contacts at random from an export and checks Email, Phone number and Job title on each against an outside reference: a reply from the contact, a call, or the employer's website. Twelve contacts cannot be verified. Of the 188 that can, 29 carry one wrong value each: 15 out-of-date job titles, 8 phone numbers that reach a switchboard or another person, and 6 email addresses for people who have left. All 29 values are well-formed, so the overview shows none of them.

The free-of-error rate in Pipino, Lee and Wang's paper on data quality assessment is one minus the proportion of data units in error. They note that deciding what counts as an error requires clearly defined criteria; in this sample, a value is in error when the outside reference contradicts it. On the verified records the rate is 1 − 29/188 = 0.846, an error rate of 15.4 per cent. With a standard error of 2.6 percentage points, a 95 per cent interval by the normal approximation runs from 10.3 to 20.6 per cent. Projected onto 19,460 contacts, about 3,002 hold a wrong value in one of the three fields checked, with an interval of 1,997 to 4,007, against an overview count of zero.

The twelve unverifiable contacts bound the estimate rather than drop out of it. Counting none of them as wrong gives 29 of 200, or 14.5 per cent, and counting all of them gives 41 of 200, or 20.5 per cent. If the contacts who could not be reached are disproportionately those who have left, the rate from the 188 is biased low, so the unverified count belongs beside the rate.

Accuracy Sampling Procedure

1. Confirm access. According to the data quality documentation, a Super Admin, the permission set with access to all account tools and settings, or a user with Data quality tools access and View all permission for contacts and companies can fix records in the overview. The export in step 4 requires Export permission. Record the overview's counts under Data Management > Data Quality as the baseline for form.
2. Test one formatting fix on one record before any bulk acceptance or automation rule. Note whether bulk acceptance appears at the account's tier.
3. Under Settings > Properties, create two contact properties: Accuracy check result, a dropdown select with the options No wrong value found, Wrong value found and Could not verify, and Accuracy check date, a date picker.
4. Export the contacts with All properties on records selected, which places Record ID in the first column. Sort on a column of random numbers and keep the first 200 rows, enough for an interval of about plus or minus 5 points near a 15 per cent error rate, since 1.96² × 0.15 × 0.85 ÷ 0.05² = 196.
5. Verify Email, Phone number and Job title on each sampled record against an outside reference, and record the result and date. For each wrong value, read the property history to find the source that wrote it. Correct the value in that source as well as in HubSpot, since a sync restores a value fixed only downstream.
6. Import the results with Record ID as the identifier. The deduplication documentation states that rows carrying a Record ID update the existing records, while any row without one creates a new record.
7. Build a report on Accuracy check result in the custom report builder and add it to a dashboard. HubSpot documents no report of the overview's counts, so add a dashboard note through Add content > Image or text. In the note, record the rate from verified records only, its 95 per cent interval, the unverified count and the step 1 baseline.
8. Check the report against the spreadsheet. It must return the spreadsheet's counts exactly, which on the sample portal are 159 No wrong value found, 29 Wrong value found and 12 Could not verify. Then open the property history of three corrected records and confirm that each shows the new value and the expected source.

Costs and Returns of Data Hub and Accuracy Sampling

Every tier buys checks of form, Data Hub Professional buys them at volume, and an accuracy figure costs staff time in any tier. The checks in every Professional hub carry no added licence cost. They cost an administrator's review time, and every merge carries a risk that cannot be undone.

Data Hub Professional is listed in the product catalog at $800 a month, or $9,600 a year. It buys the same checks in real time and at volume, with bulk merges and rules that rewrite new records as they arrive. At a loaded cost of $60 an hour, an assumption to replace with the portal's own figure, the licence equals 160 hours of administrator time a year, about 13 hours a month. The case is strongest where imports and integrations produce issues faster than that time can clear them. It is weakest where one person can review the volume by hand.

Automation also raises the cost of a wrong rule, which rewrites records nobody reviews. The Changes to records tab and a sample of rewritten values therefore belong in each rule's first week, and the duplicates documentation asks that pairs be verified before a bulk merge.

Accuracy sampling buys the one figure none of the tools report. At five minutes a record, a second assumption to replace, 200 records take about 17 hours a quarter. The case is strongest where a wrong value is expensive when used, as with direct lines for a sales team, segments built on job title and routing on location fields. It is weakest for records used once and discarded. HubSpot stores each value's age, a signal of whether it is still current: the segment filter (not) updated in last selects records whose property has not been updated within a stated number of days, so verification beyond the sample can start with the oldest job titles. The filter dates the last update in HubSpot, so an out-of-date title imported last month counts as recent.

Data Quality Symptoms and Their Causes

The overview shows no open issues, and the sales team still reports wrong numbers and bounced email. The values are well-formed and wrong, and a hard bounce marks an address only after an email to it has failed. The remedy is a recurring accuracy sample, with each correction made at the source.

Surnames such as MacDonald appear as Macdonald after a workflow runs, because a Format data action applied Change to title case. The restore tool, from Starter for Super Admins, rolls back changes made by a named workflow within 14 days. The lasting fix is to keep title case off surname properties.

A value fixed in the overview returns to its old form. A sync, an import or an API write restored the source value, which the overview does not flag again. The fix belongs in the source system or in a two-way mapping.

A rejected duplicate pair reappears in the queue. The duplicates documentation notes that a record from a rejected suggestion can return once it is merged into another record. A custom rule keyed on an integration's unique ID, on Data Hub Professional, gives the reviewer a stronger basis than similarity on names.

Documentation Conflicts and Sampling Limits

Every tier gate, limit and setting name here is as documented in October 2026, read on 6 October 2026 from HubSpot's knowledge base, developer documentation and product catalog. The data quality overview and formatting recommendations are in beta. The three documentation conflicts set out above were not tested in a live account, and how the overview treats a value written by a workflow is not documented. The article does not cover smart properties, data studio datasets, data sync settings or the Data agent. The product catalog describes the Data agent as updating CRM data from information in the CRM and on the web; like enrichment, it reaches a proxy for the entity rather than the entity.

Two conclusions go beyond any single document. The clash between the formatting rule and title case is read from two articles. The statement that no HubSpot tool compares a value with the entity rests on the documented behaviour of each tool, and a release after October 2026 could change it. The sample portal's figures illustrate the arithmetic and are not evidence about any real portal, and the interval assumes a simple random sample. The method has limits of its own: the outside reference can itself be wrong, and two reviewers applying different criteria for an error will report different rates from the same records.

In Summary

HubSpot's data quality tools compare a value with a pattern, with other records, or with its field's fill rate and usage, and enrichment fills empty values from HubSpot's commercial dataset. They detect malformed values from Professional, or from Starter through a beta, likely duplicates from Professional and empty or unused fields on all plans. Data Hub Professional automates the fixes at volume for $9,600 a year at list.

None of them compares a value with the person or company it describes, so none can show that a value is correct. Automatic enrichment, by its own documentation, defers to any value a person or another system wrote. A pattern can reject a value mapped to a meaningless state; for a well-formed value it can only guess, as the Desantos and macdonald examples show.

The practical form of the difference is a second number: an accuracy rate from about 200 verified records, reported with its interval and the unverified count beside the overview's counts. A count of zero from the overview describes the format of 19,460 contacts; only the sample estimates how many of them carry a wrong value.