CRM data quality: how to measure it on five dimensions

Short answer

CRM data quality is measured on five dimensions: completeness (are the key fields filled in), accuracy (does what is written match reality), uniqueness (is one customer one record), timeliness (do status and stage show the current situation) and consistency (is the same thing written the same way everywhere). The practical method: every month, open 50 random records, check them on the five dimensions and keep the result in a one-page table. When a number worsens, look for the cause — usually not a careless manager but a badly designed field, free text or an unchecked import.

What bad data costs

Data quality sounds abstract, but its consequences are very concrete. A lead with a mistyped number cannot be reached. A lead whose status is out of date inflates the pipeline and throws the forecast off. A duplicate record sends the same person two messages in a campaign. Leads closed with "other" as the reason leave "why are we losing?" unanswered.

In the worst case, leadership stops trusting the CRM and goes back to deciding by feel. From then on, the motivation to enter data disappears too, and the cycle gets worse.

Five dimensions

1. 1CompletenessThe share of key fields filled in: phone, source, status, next step, close reason.
2. 2AccuracyWhether the value matches reality: the number works, the name is right, the need matches what was said in the conversation.
3. 3UniquenessHow many records one person has: the share of records sharing a normalised number.
4. 4TimelinessWhether status and stage reflect the present: leads in "working" with no activity for 30 days.
5. 5ConsistencyWhether the same thing is written the same way: campaign names, city names, the reason list.

How to measure each dimension

• Completeness — automatic: export from the CRM and count records with key fields empty.
• Uniqueness — automatic: normalise numbers and count records sharing a number.
• Timeliness — automatic: open leads whose last activity is older than 30 days.
• Consistency — semi-automatic: look at the list of distinct values in a field. Twenty-three spellings instead of five cities means a problem.
• Accuracy — by hand only: open 50 records and compare them with the conversation or call.

The monthly 50-record check

Each month, pick 50 random leads created in the last 30 days. Ask five questions of each record: are the key fields filled in, are the number and name correct, does this customer have another record, do status and stage reflect the current situation, were campaign and reason picked from the list? Write the answers in one table. It takes one person an hour or two, and it tests the reliability of every report you have.

A one-page quality table

Keep the result in the same form every month so the trend shows:

• Completeness: share of key fields filled — target and actual.
• Accuracy: in how many of the 50 records at least one error was found.
• Uniqueness: share of duplicate records.
• Timeliness: number of "forgotten" open leads.
• Consistency: number of values written outside the list.
• This month's main problem and one fix — with an owner and a date.

Look for causes, not culprits

Bad data usually comes from how the system is designed. Typical causes and fixes:

• A free-text field → a pick list.
• A key field that is not required → required on transition.
• Numbers copied by hand → numbers captured automatically from the conversation.
• An unchecked import → normalise and check for duplicates before importing.
• An interface where changing status is awkward → a one-tap change.

Cleaning up existing duplicates is covered in CRM duplicate records.

Checking data written by AI

In an AI CRM, some fields are filled in by AI from the conversation. That improves completeness but does not remove the need to check accuracy. In the monthly check, mark AI-filled fields separately and look at three things: was a value written that is not in the conversation, was a relative date ("tomorrow") converted correctly, does the AI summary match what the customer actually said? In a well-built system, AI leaves a field empty when it cannot find the value — an empty field is better than a wrong one.

Who is responsible

Data quality needs an owner — usually the CRM admin or whoever owns sales operations. They run the monthly check, present the table to the head of sales and track the fixes. Managers are responsible for the quality of their own records, but they need the right tools: required fields, pick lists, one-tap changes.

An illustrative example

This is an illustrative example. A car service chain's first monthly check shows: 40% of records have no source, the "city" field holds 19 different spellings instead of 5 cities, and in 11 of the 50 records the number is either wrong or belongs to someone else. Half of the close reasons are "other".

Fixes: the source and reason fields become required pick lists, the city field is chosen from a list, and numbers are no longer copied by hand. Three months later the table shows clear improvement in completeness and consistency, and the accuracy check continues every month.

Common mistakes

• Treating data quality as a one-off "clean-up project".
• Measuring only completeness — fields that are filled but wrong stay invisible.
• Blaming managers for bad data without looking at system design.
• Making every field required — managers fill them with meaningless values.
• Never checking the fields AI fills in.

Limitations

A 50-record sample can miss small problems — it shows direction, not an exact percentage. Some errors only surface when you contact the customer: you find out a number belongs to someone else when you call it. Data quality is also tied to personal-data processing: not collecting unnecessary data and deleting old data on time are part of quality too — check retention rules against local law.

Data quality in Vexvon

In Vexvon, part of data quality is protected at intake: numbers are recognised in chats in five formats, the company's own numbers are not counted as leads, no new record is created if an open lead exists for the same number, and CSV imports are checked for repeated numbers. Company-defined fields are extracted from the conversation conservatively — a value is written only if it is in the text, and left empty otherwise. The close reason is chosen from a list, and every change goes into the activity log. More on Vexvon CRM.

Next step

Run your first 50-record check this month and write the five-dimension result on one page. The weakest dimension will point to next month's single fix. Which fields a chatbot should write to the CRM is covered in chatbot and the CRM customer record. More articles are in the AI CRM section, and we can set up the check together during a demo.