A 3-role governance model is enough to get started
In 80% of SMBs and mid-market companies, data is managed "on the fly": each department has its own files, its own definitions, its own habits. The result: the same data exists in 3 different versions, no one knows which is correct, and AI projects fail for lack of reliable data. The solution isn't software, it's a simple organizational model: a Data Owner who decides, a Data Steward who operates, and clear quality rules. This framework is set up in 6 to 8 weeks and halves data-related incidents.
The problem
Take an industrial mid-market company of 400 employees. The sales department uses a CRM with 12,000 customer records. Accounting manages the same customers in the ERP. Marketing has its own database in an Excel file for email campaigns. Three systems, three different truths.
When the sales director asks "how many active customers do we have?", they get three answers: 8,200 in the CRM, 9,100 in the ERP, 7,500 in the marketing file. No one can settle it because no one is responsible for the definition of "active customer".
This scenario repeats at every level of the company:
- Inconsistent product data — Product references differ between the catalog, the ERP and the e-commerce site. Based on our field observations, it's not uncommon for a significant share of a product catalog to consist of duplicates.
- Fragmented HR data — Skills are in one Excel file, annual reviews in another, training in the HRIS. It's impossible to answer "who in the company speaks German and knows SAP?" in less than a week.
- Distorted decisions — A dashboard that aggregates inconsistent data gives an illusion of control. Based on our observations, it sometimes happens that a major investment decision rests on a revenue figure overestimated because of a duplicate in the billing data.
The underlying problem is always the same: there's no framework defining who is responsible for what. That's exactly what data governance provides. And for an SMB or mid-market company, you don't need a 200-page framework — a simple three-role model is enough.
The AI solution
The governance model we deploy at our clients rests on three pillars. AI steps in to automate monitoring and anomaly detection, but the organizational framework remains human. Here's how to structure the approach with the support of artificial intelligence.
Data Owner: the business decision-maker
Each data domain (customers, products, suppliers, HR) has a single business owner. They define the rules, validate the definitions and arbitrate conflicts. AI assists them by automatically generating quality reports and detecting cross-system inconsistencies via fuzzy matching algorithms.
Data Steward: the quality operator
The Steward applies the rules day to day. They monitor the indicators, fix anomalies and document the datasets. Thanks to AI, they receive automatic alerts when a quality threshold is crossed (e.g. duplicate rate above 5%) and correction suggestions via NLP.
Automated quality rules
Validation rules — completeness, uniqueness, format compliance, freshness — are defined by the Data Owner and executed automatically by a rules engine. AI suggests new rules by analyzing historical data patterns and identifies at-risk columns before an incident occurs.
This model fits into your existing organization. The Data Owners are your current department heads. The Data Stewards are operational staff who devote 20% of their time to data quality — not a dedicated role. To go deeper, see our data quality workshop.
Implementation
Rollout happens in three phases over 6 to 8 weeks. The goal: an operational framework by the end of the second month, with first measurable results.
Mapping and appointment (weeks 1-2)
Identify your 5 to 8 critical data domains (customers, products, suppliers, employees, finance, etc.). For each, appoint a Data Owner (the relevant business director) and a Data Steward (an operational person who knows the data day to day). Document it all in a simple table: domain, Owner, Steward, source systems, volume. A Google Sheet is enough at this stage.
Definitions and rules (weeks 3-5)
Bring each Data Owner together with their Steward to define the key business terms (e.g. "active customer = at least one order in the last 12 months"). Document 3 to 5 quality rules per domain: mandatory completeness on critical fields, valid email format, uniqueness of product codes, maximum freshness of 30 days. These rules become your control baseline.
Automation and monitoring (weeks 6-8)
Deploy a script or tool (Great Expectations, Soda, dbt tests) that runs your quality rules automatically. Set up alerts by email or Slack when a threshold is exceeded. The Data Steward receives a weekly report. The Data Owner receives a monthly summary. Iterate: add new rules each month based on the incidents observed.
Results
Here are the results measured at our SMB and mid-market clients after 3 to 6 months of structured governance.
Frequently asked questions
What is the difference between a Data Owner and a Data Steward?
The Data Owner is the business lead for a data domain: they decide on the rules, validate the definitions and arbitrate conflicts. This is often a director or a department head. The Data Steward is the operational role that applies these rules day to day: they monitor quality, fix anomalies and document the data. In an SMB, a Data Steward can cover 2 to 3 domains.
Do you need a dedicated tool for data governance?
Not to start. A shared spreadsheet (Google Sheets or Excel Online) is enough to document the roles, the definitions and the quality rules. A dedicated tool (Atlan, Alation, DataHub) becomes relevant when you exceed 50 datasets or when you have more than 3 Data Stewards. Start simple, add tooling later.
How long does it take to set up data governance?
For an SMB of 50 to 300 employees, count on 6 to 8 weeks to lay the foundations: identifying the Data Owners, appointing the Stewards, documenting the 10 to 15 critical datasets and defining the first quality rules. The model is then refined continuously. The classic mistake is trying to cover everything at once.
How do you convince management to launch a data governance project?
Start from a concrete, quantified problem: "We double-billed 45,000 euros last month because the customer data was inconsistent between the CRM and the ERP." Show the cost of poor quality (errors, time lost, wrong decisions) and propose a pilot on a specific domain with a measurable ROI in 3 months.
For tech profiles
Tooled governance architecture
DataHub / Atlan / OpenMetadata
Centralized catalog that automatically indexes the metadata of your sources (SQL databases, files, APIs). Lets you document business definitions, tag sensitive data and visualize lineage. DataHub is open source and deploys in 2 hours via Docker.
Great Expectations / Soda / dbt tests
Automatic execution of validation rules on your data pipelines. Great Expectations generates detailed HTML reports. Soda offers a no-code interface for Data Stewards. dbt tests integrates directly into your SQL transformation pipeline.
Recommended stack
Comparison of approaches
| Criterion | Structured governance + AI | Manual governance | No governance |
|---|---|---|---|
| Anomaly detection time | Minutes (auto alerts) | Days | Weeks or never |
| Monthly cost (SMB) | 500-2,000 € | 0 € (human time) | High hidden cost |
| Scalability | Excellent | Limited | None |
| Business adoption | Strong (readable reports) | Average | Nonexistent |