How to start data governance: a 90-day plan

For many people, "data governance" brings to mind thick policy documents, committees and expensive platforms. That is why data governance programs often either never start or start too broad and quietly fade out six months later.

In practice, data governance answers simple questions: who is responsible for the data, how it is defined, how its quality is checked and what happens when an error is found. You can start small and get visible results within three months. Here is a plan for doing exactly that.

Why the business needs data governance

Without agreements on data, every department lives with its own numbers. Sales count revenue by order date, finance by payment date. Marketing exports a customer base full of duplicates, logistics corrects addresses in its own system, and the corrections never reach anyone else. As a result:

• meetings turn into arguments about whose numbers are right;
• analysts spend their time reconciling instead of analyzing;
• errors are found by customers or management, not by the people responsible for the data.

Data governance removes this uncertainty: every important dataset gets an owner, every metric gets one definition and every error gets a clear path to resolution.

The main mistake: starting with the whole company

The most common failure scenario is trying to describe all the data at once, assign owners in every department and write a company-wide policy. Such a project takes a long time to deliver anything visible, and the business loses interest.

The opposite approach works: pick one area where data problems hurt the most, bring order there and show the result. Then repeat in the next area, with templates already in hand.

The 90-day plan

Weeks 1–2: choose the area and record the baseline

Look for an area where bad data costs money or time and management can see it. Typical candidates:

• a customer reference with duplicates and incomplete contact details;
• a product reference with different names and units of measure;
• management sales reporting where numbers differ between departments.

Record the baseline right away: how many duplicates, what share of fields is empty, how many hours a month go into reconciliation. Without these numbers, you will have nothing to prove the result with in three months.

Weeks 3–4: assign roles

The chosen area needs two roles:

• Data owner — a business leader. They approve rules, resolve disputes and answer to management for quality. This is not IT: the owner of the customer reference is more likely the commercial director than the CRM administrator.
• Data steward — a specialist who looks after the data day to day: investigates errors, maintains the reference and answers colleagues' questions.

The roles do not require new hires. They are usually part of the duties of people who already work with this data, now made explicit.

Weeks 5–8: agree on definitions and rules

Bring together the owner, the steward and the main users of the data and write down:

1. Definitions. What counts as a customer, an active customer, a duplicate. The definitions go into a business glossary — a simple document or spreadsheet available to everyone.
2. Quality rules. Which fields are mandatory, what format phone numbers are stored in, how to tell that two customers are the same person.
3. Source of truth. Which system data is created and edited in, and which systems only receive a copy.

Do not try to cover everything. Ten to fifteen rules for the key fields are enough for the first step.

Weeks 9–12: automate the checks and show the result

Rules on paper do not work if nobody checks them. Turn them into automated checks that run on a schedule and notify the steward of violations. You do not need a dedicated platform for this: at the first stage, database queries and a simple report showing the trend are enough.

In parallel, describe the resolution process: who gets the error alert, by when and in which system they fix it, and what to do if the error keeps coming back.

At the end of the third month, compare the metrics with the baseline and show management: there were this many duplicates and this many hours spent on reconciliation, and now there are this many.

Which documents you need at the start

The minimum set that data governance cannot do without:

• a list of the area's key datasets with their owners and stewards;
• a business glossary with definitions of metrics and entities;
• a list of quality rules and check results;
• a description of the error resolution process.

All of this fits on a few pages. A fuller data catalog becomes worthwhile once there are several areas and people need to search for data across the company.

How to measure the result

Choose two or three metrics the business understands and track them monthly:

• the share of records that pass all quality checks;
• the number of duplicates or empty mandatory fields;
• the time spent on manual reconciliation and corrections;
• the number of incidents where report users discovered a data error.

A growing share of error-free records and less manual work are the best arguments for extending the program to the next areas.

Do you need to buy a platform?

Not at the start. Definitions, rules and roles matter more than the tool, and the first checks can be built on your existing database and open-source tools. Dedicated platforms for catalogs, data quality and MDM pay off when there are many areas, several teams and growing data volumes. By then you will know exactly which features you need and will not overpay for the rest.

Summary

Data governance does not start with a hundred-page policy but with one painful area, two assigned roles, a dozen rules and automated checks. Within three months, this approach delivers a measurable result and ready-made templates you can carry over to the next set of data.

If you need help getting started, see our Data Quality & Data Governance service: we start with an audit, set up roles and rules, and leave you with working checks.