CRM data quality: how to keep your database reliable

CRM data quality is the degree to which the records in your database are complete, correct, free of duplicates, up to date and standardized, and it depends more on the behavior of the people who record than on any cleanup. The database becomes reliable when leadership sets a few required fields per stage, replaces free text with picklists, gives every import and integration an owner and measures quality every week. Almost every wrong report that reaches the board meeting was born from a bad record weeks earlier, made by someone who did not know that field fed a decision. That is why a cleanup sprint returns a clean database for a month or two and then loses its effect: the team keeps recording the same way, and the dirt comes back at the rate it comes in.

In the assessments I run, the director asks how much the manufacturing segment represented in the quarter’s pipeline, and the dashboard shows three bars, “Manufacturing”, “manufacturing” and “Mfg.”, because the field accepted free text. The problem is bigger than most managers imagine: in an exercise with 75 executives, each one marking errors in 100 records produced by their own department, the researchers found that on average 47% of newly created records had at least one critical error, and only 3% of the quality scores could be rated acceptable under the loosest standard (Harvard Business Review, 2017).

How does a bad record reach the board meeting as a wrong number?

The salesperson leaves the meeting with the customer, creates the deal with a value estimated off the top of their head and leaves the close date at the default the system suggests, thirty days ahead. The manager consolidates the month’s pipeline, and the board decides on hiring or media budget based on a number nobody checked. When the deal does not close on that date, the salesperson pushes it another thirty days, and the next quarter already starts with an inflated pipeline.

In marketing, a lead without a source erases the channel that generated it, and the cost per opportunity of each campaign becomes an estimate. In sales, a loss reason picked out of habit leads leadership to change the price when the customer walked away over implementation time. In post-sale, the contact who left the client company remains the decision maker in the record, and CS discovers at renewal that it had been talking to someone who no longer decides.

Why doesn’t a one-off cleanup fix the database?

Companies tend to treat a dirty database as inventory: they hire a cleanup sprint, fix everything in a spreadsheet and import it back. The sprint handles what has already come in and leaves untouched what will come in tomorrow. Every form without validation, every rushed import and every salesperson who creates a company without checking whether it already exists brings back the problem the sprint fixed. Within a few months the database returns to its previous state, and the board concludes the investment was wasted. Cleanup corrects the past once, and only entry rules, standardization and a weekly routine keep it from repeating.

What are the dimensions of CRM data quality?

I work with five dimensions, because each one has a different cause, owner and way of measuring. The table gives a commercial example of each one and the measure the CRM produces with a simple report.

Dimension	What it means	Commercial example	How to measure in the CRM
Complete	The stage’s required fields are filled in	Deal in proposal with no value or no close date	Records in the stage with all required fields
Correct	The recorded value matches reality	Director title for someone who is an analyst, tax ID with a wrong digit	Error-free records in the sample reviewed that month
Free of duplicates	Each company, contact and deal exists once	The same company created by the SDR and by the salesperson, with different owners	Records that repeat tax ID, domain or email, divided by the total
Up to date	The record reflects today’s situation	Open deal with a close date 40 days past	Open deals with past close dates and contacts with no interaction in 12 months
Standardized	The same data always appears in the same format	Segment written four different ways, phone with and without area code	Number of distinct values per field and share of “Other” in picklists

Correctness depends on a sample reviewed by someone who knows the customer, because the system does not know the job title is wrong. The other four come from reports starting in the first week.

What makes a CRM database dirty?

The causes I find repeat from company to company, almost none of them depend on the platform chosen, and the last one sustains the others.

Too many fields

Each team asks for its own field during implementation, and the deal form ends up with forty questions. A salesperson who has to fill in everything to move a stage fills in anything, and the requirement starts producing made-up data. A field no report uses only hides the ones that matter.

Free-text fields

Segment, source, loss reason and job title typed by hand produce dozens of variations of the same value, and no dashboard groups “Manufacturing” with “manufacturing”. The notes field serves for conversation context and supports no measurement.

Imports without rules

The event list, the purchased spreadsheet and a newly hired salesperson’s personal contacts go into the CRM in one afternoon, without a duplicate check, without a source and with columns mapped to the wrong field. A single import like that undoes months of the team’s care, and without someone to approve the file before the upload, each team imports the way it knows.

Integrations without an owner

The website form, marketing automation and the ERP write into the CRM all the time. When the integration creates a new contact instead of updating the existing one, or overwrites the phone number the salesperson corrected, the database gets worse without any person making a mistake. The failure lasts for months, because nobody looks at the integration’s error log or answers for it.

Nobody responsible for the database

The CRM administrator has no authority to demand records from the salesperson, and the manager pushes on the pipeline without looking at the quality of what is in it. In CRM adoption I deal with the team that does not use the system. Here the team uses it and records badly, and the fix requires one person responsible for the database and leaders who demand good records from their own teams.

Which fields should you require at each funnel stage?

A minimum field rule per stage solves the first two causes, because it asks for little and asks at the moment the person has the information. The salesperson does not know the deal value at first contact and does know it at the proposal, and requiring the value early produces a made-up number. The criterion behind the table, which I adjust with leadership, is to require only what feeds a report or a handoff.

Stage	Required fields	Who records	What it is for
Lead	Name, business email, company and source	Marketing or the form integration	Channel attribution and duplicate check
Marketing qualified lead (MQL)	Segment, company size and seniority, in picklists	Marketing or SDR	Qualification criterion and queue priority
Opportunity	Stated pain, decision maker and date of the next meeting	SDR or salesperson	Forecast and pipeline meeting
Proposal	Value, close date and products	Salesperson	Forecast and deal size by segment
Won	Signature date, final value and handoff fields for CS	Salesperson	Revenue, commission and onboarding
Lost	Loss reason from a picklist and competitor, when there is one	Salesperson	Loss analysis and offer adjustment

The qualified lead criterion and the handoff fields come from the SLA between marketing, sales and post-sale, and the field rule is how the CRM enforces that agreement without depending on anyone’s memory.

How do you standardize CRM data?

Standardizing means replacing typing with choosing. Segment, source, company size, seniority, loss reason and product become closed picklists, with an owner who approves each new value. A loss reason with thirty options goes back to working like free text, because the salesperson picks the first one that seems to fit.

Leadership needs to read every month what goes into “Other”, and if that option grows beyond a small share of the choices, the list no longer represents the operation. Phone, tax ID and website stay open with format validation on entry, such as phone with area code and tax ID with numbers only. For the company name, agree on a convention and write it in the team guide.

How do you handle duplicate data in the CRM?

Deduplication starts with the key, the piece of data that identifies a record without ambiguity. In B2B, the company key is the tax ID (CNPJ in Brazil) or the website domain, and the contact key is the business email. The company name does not work, because the same company shows up with and without “Ltd.”, as an acronym and under its trade name.

Merging needs a written rule before the first click: which record survives, which value prevails when the two disagree and who stays as owner. The owner is the sensitive point, because a duplicated company usually has two salespeople, and merging without a rule turns into a dispute over accounts and commission. Duplicates with different owners go to the manager to decide, and the database owner merges the rest during the routine.

Preventing costs less than merging. The salesperson searches for the company before creating it, integrations update the record by key instead of creating another, and every import goes through a duplicate check before the upload. With those barriers in place, the routine deals with exceptions.

Database hygiene routine: what to do every week and every month

The routine needs to be short, because one that takes a whole afternoon disappears in the first week of month-end close. The times are a reference and change with the size of the database.

Frequency	Task	Owner	Reference time
Weekly	Review open deals with past close dates	Each salesperson, followed up by the manager in the pipeline meeting	15 min
Weekly	Handle duplicates created during the week	Database owner	30 min
Weekly	Check leads without a source and contacts without an associated company	Marketing	20 min
Monthly	Read what went into “Other” and decide on new picklist values	Database owner, with the leaders	30 min
Monthly	Review integration errors and the month’s imports	Database owner, with IT	1 h
Monthly	Review the 100-record sample from each team	Each team’s leader	1 h
Quarterly	Retire unused fields and review the per-stage rule	Database owner and leaders	2 h
Every six months	List contacts with no interaction and no purpose to assess deletion	Database owner, with the data protection officer	2 h

How do you present database quality to leadership?

The board takes care of the database when quality appears on the same dashboard as the pipeline. I build an indicator with the four dimensions the CRM measures on its own and the monthly sample score, each one as a percentage of records within the rule, broken down by department and team, and I bring the trend of the last few months to the meeting.

In a hypothetical example, the database shows 92% completeness in proposal-stage deals, 94% of companies free of duplicates, 70% of open deals with current dates and a score of 72 on the monthly sample. The past close dates on three out of ten deals explain why the forecast misses. On the sales dashboard, the quality indicator goes on the first line, before the numbers that depend on it.

What does each leader need to ensure?

Data quality is the behavior of each role in the CRM, and each leader answers for their own team’s records. In sales, the manager asks for close date, value and loss reason in the pipeline meeting, deal by deal. In marketing, leadership answers for each lead’s source, for imports and for the forms that write into the database. In post-sale, the CS leader keeps the contacts of active accounts up to date and records the departure of anyone who left the client.

The board has a role no department can fill for it: appointing the database owner, giving that person authority to refuse an import or a new field, and refusing, itself, any report that comes from a parallel spreadsheet. When the board accepts the spreadsheet number, the team understands the CRM can stay dirty.

Data quality and LGPD: what the law asks of the commercial database

Brazil’s data protection law (LGPD) treats quality as a principle. Article 6, item V, of Law 13,709/2018 guarantees data subjects the “accuracy, clarity, relevance and updating of the data”, and Article 18 gives data subjects the right to request correction of incomplete, inaccurate or outdated data and deletion of unnecessary or excessive data. Since the CRM holds personal data of buyers and decision makers, a wrong or forgotten record is also a compliance matter.

The law provides for the end of processing when the data is no longer necessary for its purpose (Article 15) and for deletion after that, with exceptions such as compliance with a legal obligation (Article 16). The six-month review separates contacts with no interaction and no clear purpose, and the data protection officer, together with legal, decides what to delete, anonymize or keep, with the legal basis for each purpose on record. I write as an operations consultant, with no intention of giving a legal opinion; the use of CRM data in AI tools is covered in AI and LGPD.

Why does AI depend on a reliable database?

Lead scoring, the score that ranks leads by their likelihood to buy, AI sales forecasting and churn risk alerts all learn from the CRM’s history. A model trained on past close dates and loss reasons picked out of habit returns the same errors with a percentage and a chart, as I discuss in AI on top of an unstructured CRM. Before turning on a score or a forecast, run the five dimensions on the fields the model will read and fix the ones that fall below target.

Roadmap for leadership to improve data quality

The roadmap leaves the cleanup until after the rules, so the sprint does not have to be repeated.

1. Measure before cleaning. Run the five dimensions on the current database and take the first 100-record sample per team.
2. Appoint the database owner. Write down what this person approves, from imports to picklist values, and tell the team the board stands behind those decisions.
3. Cut fields and write the per-stage rule. Keep only the fields that feed a report or a handoff and define the required ones for each stage.
4. Replace free text with picklists and define the keys. Convert classification fields into picklists, choose the company and contact keys and set up format validation.
5. Clean the backlog once. Correct, standardize and merge the existing database using the merge rule, with duplicates owned by different people decided by the managers.
6. Close the entry points. Adjust integrations and forms to update by key and route every import through the database owner.
7. Prepare each role. Each person needs to know what to record at their stage and which report that field feeds.
8. Set up the routine and the indicator. Put the weekly routine in the pipeline meeting, the monthly one on the leaders’ calendar and the indicator on the board’s dashboard.

Data quality training: what HR and L&D need to assess

How to recognize the need

The signs appear when leadership presents numbers. The board questions the report before discussing the result, departments arrive with different totals for the same period and the manager redoes the pipeline in a spreadsheet the day before the meeting. Before treating the case as a tool failure, ask the leaders whether there is a per-stage field rule and who answers for the database.

Who takes part

The core group is the sales, marketing and CS leaders, who demand the records, together with the database owner and the CRM administrator. Then each role goes through its own class in the company’s CRM: salespeople and SDRs learn the fields for their stage and searching before creating, marketing learns the import and source rules, and CS learns to keep account contacts current.

How to measure

The database quality indicator is the measure of the program. Take the baseline beforehand, with the five dimensions and the 100-record sample per team, and repeat it after four weeks and after a quarter. New duplicates per week show whether behavior changed at the entry point, where dirty data is born.

If your board distrusts the report before discussing the result, that is the work of the in-company CRM adoption training. I run the program with the leaders and with each role, in the CRM you already use. The group leaves with the CRM best-practices guide, including the per-stage field rule and the standardization, and with the portfolio and database management manual, including the hygiene routine and the quality indicator.

Frequently asked questions

What is CRM data quality?

It is the degree to which CRM records are complete, correct, free of duplicates, up to date and standardized. With a good-quality database, leadership uses the report to decide without redoing the math in a spreadsheet.

How often should you clean the database?

Database hygiene works as a routine: short weekly tasks for past close dates, duplicates and leads without a source, a monthly review of picklists, integrations and a sample of records, and a quarterly review of fields. A large cleanup makes sense once, when the database has never been reviewed, and the routine starts right after.

Who should be responsible for CRM data quality?

A person appointed by the board, usually from revenue operations or CRM administration, with authority to approve imports, fields and picklist values. Each leader still answers for their own team’s records, and the database owner does not replace that follow-up.

How do you avoid duplicate data in the CRM?

Define a key for each object, such as tax ID or domain for companies and email for contacts, and set up integrations and forms to update the record by key. Teach the team to search before creating and route every import through a duplicate check.

Can I delete old contacts from the database?

The LGPD provides for the end of processing when the data is no longer necessary for its purpose and for deletion after that, with exceptions such as compliance with a legal obligation. The decision on what to delete, anonymize or keep belongs to the data protection officer and legal, and the six-month database review prepares the list for that decision.

How do you train the team to keep the database clean?

Train by role, in the company’s CRM, showing each person the fields for their stage and the report each field feeds. Leadership goes through the program first, because leaders are the ones who demand the records in the meeting, and the quality indicator measured before and after shows whether behavior changed.