The cost of bad data: how to put a number on quality problems

Industry estimates of what bad data costs are enormous. Gartner has put the average cost to an organization in the millions of dollars a year, and an often-quoted IBM estimate put the cost to the US economy in the trillions. Those figures describe large companies in aggregate, and they are too abstract to act on. The useful question for a 20-person business is smaller: what does bad data cost us, and which fix pays back first?

This guide covers where the costs hide, a simple way to estimate them, and a fix order that works for teams without a data engineer.

Where the cost hides

Bad data almost never appears as its own line in the budget. It is folded into other costs.

1. Rework. Someone exports from the CRM and the billing system every month and reconciles them by hand. Someone deletes duplicate contacts before every email send. Someone fixes dates that came in three different formats. This is the most visible cost, and the easiest to measure.

2. Wrong decisions. This is usually the biggest cost and the hardest to see. A few typical cases:

• Ad attribution credits the wrong channel, so budget moves toward the one that is not working.
• Duplicate customer records inflate the customer count and deflate revenue per customer.
• Cost data with a wrong allocation makes a product look profitable when it is not.
• A churn analysis misses at-risk accounts because usage data from one system never joined to the account list.

3. Customer-facing mistakes. Wrong names in emails, duplicate mailings, invoices with the wrong amount, parcels sent to an old address. Each is small. Together they create support tickets and erode trust.

4. Slow or abandoned projects. A dashboard project stalls for weeks because two systems disagree on how many customers you have. The project did not fail because of the tool; it failed because nobody knew which number was right.

5. Compliance exposure. Privacy laws such as the GDPR give people the right to have inaccurate personal data corrected. Records you cannot keep accurate are a risk, not only an annoyance.

6. Wasted tools. A BI tool on top of unreliable data produces unreliable dashboards faster. People stop trusting it and go back to spreadsheets. That looks like a tool problem, but it is a data problem.

Estimating the cost in your business

You do not need precision. A defensible estimate is enough to decide what to fix.

Step 1: Count the rework hours

Ask everyone who touches data (sales ops, marketing, finance, support, the founder) two questions:

1. Roughly how many hours a week do you spend finding, fixing or working around wrong or missing data?
2. What is one concrete example from the last month?

Then multiply by a loaded hourly cost.

Worked example. Five people each spend 3 hours a week on reconciliation and cleanup. At a loaded cost of 40 dollars an hour over 48 working weeks:

5 people x 3 hours x 48 weeks x 40 dollars = 28,800 dollars a year

That is before counting a single wrong decision.

Step 2: List the known problems

For each known problem, write down:

• the source it lives in
• the type of problem: duplicates, missing values, wrong values, inconsistent formats, stale data
• roughly how many records it affects
• which reports or decisions use that data
• the current workaround

A list of fifteen known issues with owners and workarounds is often enough to get a budget approved.

Step 3: Estimate decision impact

Go through three or four significant decisions from the past year. For each, ask: which data did we use, how sure are we it was right, and what would we have done differently with correct data?

Worked example. Paid search looked like it returned 3 dollars for every dollar spent, so you moved 10,000 dollars a month into it. A later check shows a third of those conversions were duplicates from a tracking tag that fired twice. The real return was 2 dollars, not 3. If another channel returned 2.50 dollars, moving that 10,000 a month cost roughly (2.50 minus 2.00) x 10,000 = 5,000 dollars a month in lost return.

Rough ranges are fine. "Somewhere between 3,000 and 6,000 a month" is enough to rank the problem.

What to fix first

Sort problems by two things: how much they cost and how hard they are to fix.

First: high cost, easy fix

• Required fields in the CRM for the things you report on (deal value, close date, source).
• Dropdowns instead of free text for country, industry, product and lead source.
• Validation that rejects impossible values: negative order amounts, emails without an @, dates in the future.
• Deduplication rules in the CRM or email tool. Most have them built in.
• More frequent syncs where stale data is the problem. This is usually a setting, not a project.

Second: high cost, more effort

• Cross-system checks. The number of active customers in billing should match the CRM. Compare them every month and investigate the gap.
• A one-time cleanup of the worst table, such as merging duplicate customers.
• One agreed definition per key metric. The single source of truth playbook covers how to set that up.

Later: large projects

Master data management and full data lineage are worth it for larger companies. For a small team they usually cost more than the problems they solve.

Preventing new problems

• Fix at entry. The cheapest moment to correct data is when it is typed in.
• Stop copying by hand. Every manual export and paste introduces errors. Connect systems directly where you can; migrating from Excel to a BI tool covers how to make the switch.
• Give each source an owner. Someone who notices when it breaks and is expected to fix it.
• Show people where their data goes. Sales reps fill in close dates more carefully once they see the forecast that depends on them.
• Watch for problems. Look at freshness, completeness and sudden jumps regularly. Spotting data quality issues before they cost you has a checklist.

Where clariBI helps, and where it does not

clariBI cannot clean your CRM. It does make problems easier to see:

• Every connected source gets an automatic insights dashboard. It is a quick way to see whether dates were read as dates, whether amounts are numbers and whether a column is mostly empty.
• Settings > Integrations shows each source's sync status, and a dashboard warns when its data may need a refresh or when the source's columns changed. The help center's data source errors guide lists what each message means, and the data source troubleshooting docs cover sync problems.
• Reports include a data confidence note and data quality warnings: the time window covered, how complete the sources are and any field substitutions. See report sections.
• Every number narrated in an answer or report is checked against the chart data, so the text cannot quote a figure the data does not contain. That does not fix a wrong input; it only stops the write-up from adding new errors.

Making the case

Keep it to one page:

1. The estimate: "Bad data costs us roughly X a year in rework and Y in decisions."
2. Two or three concrete stories with numbers.
3. The top three fixes, with effort and cost for each.
4. The expected return in plain terms: hours saved per month, and the decision error you avoid.

A good first step this week: send the two rework questions to everyone who handles data, add up the hours, and pick the one problem that appears in more than one answer.