Your engineers need to ship fast. Your stakeholders need governance they can trust. Your compliance demands keep multiplying. AI was supposed to help—instead, it’s exposing every gap in your data foundations.

Organizations are racing to deploy AI applications, but they’re discovering a brutal truth: AI amplifies whatever data quality and governance problems you already have.

Feed an AI model inconsistent data definitions? It will confidently produce inconsistent outputs. Train it on data with unclear lineage? You won’t be able to explain how it reached its conclusions when regulators come asking. Deploy it without proper access controls? You’ve just automated your compliance violations at scale.

Yet most governance programs share the same fate: they launch with enthusiasm, mandate a series of new processes, and within months, quietly fade into irrelevance as teams find workarounds and adoption plateaus.

The solution? Design a data governance experience that makes doing the right thing the easiest thing.

At CONTEXT: The Metadata & AI Summit, Björn Barrefors, Metadata Management Lead at ICA Gruppen shared how he transformed governance from a checkbox exercise into infrastructure that data teams rely on daily.

The approach consists of two principles:

1. Shift-left governance that meets engineers where they work 
2. Psychology-driven design that makes participation feel like help rather than homework

Watch the session on demand:

Or, read on for the best practices you can use as a blueprint for building data governance that scales without slowing down your teams.

Why is data governance important for AI?

According to research from Gartner, 60% of organizations may not realize the benefits of AI without a solid data governance framework. And a report from KPMG found that nearly two thirds (62%) of data leaders cite a lack of data governance as the main challenge to their AI initiatives. 

• Unclear ownership paralyzes projects when no one can authorize data use. 
• Inconsistent policies create legal exposure when different teams apply different standards. 
• Privacy concerns block deployments when you can’t demonstrate data access controls.

The irony? Most organizations respond to this AI governance crisis by doubling down on the exact approaches that created the problem in the first place. They establish centralized committees that become bottlenecks. They mandate documentation that engineers ignore because it’s divorced from their actual work. They chase 100% coverage instead of focusing on the 20% of data assets that matter most.

Why do traditional approaches to data governance fail?

Traditional data governance programs fail because they create friction rather than eliminate it. Every governance checkpoint becomes another reason projects slow down, another meeting to schedule, another ticket to file. Engineers learn to route around governance because following it would mean missing their deadlines.

The result is predictable: governance theater that looks good in audits but doesn’t actually protect the organization or enable better AI outcomes.

Data governance succeeds when it blends into the way teams already work. Two principles make this possible: shift-left governance that meets engineers in their native tools, and psychology-driven design that makes participation feel valuable rather than burdensome.

Principle 1: Shift-left data governance

What is shift-left data governance?

Shift-left data governance puts governance controls at the source, where data engineers and developers already work. Instead of forcing teams to use a separate data governance software, this approach captures information automatically from tools they’re already using.

Shift-left data governance makes compliance easier simply because it eliminates the friction of context-switching and reduces the cognitive load on engineering teams.

4 principles for shift-left data governance success

The key to success with shift-left data governance is understanding the core principles that make this approach work across organizations:

1) Keep the platform democratic

Trust teams to govern their domains. For example, put the responsibility of governing assets on individual teams because those teams are the subject matter experts. A central committee wouldn’t be able to understand every schema across billions of records.

2) Intercept metadata at the source

Build processes that capture information from where data engineers naturally work. Avoid asking them to duplicate effort in a separate system.

3) Start with the essentials

Begin with the assets and governance requirements that actually matter for downstream consumption. Don’t attempt full coverage from day one.

4) Use discovery as the carrot, not compliance as the stick

Lead with the productivity benefits: faster research, better discovery, automated deprecation workflows. The compliance benefits follow naturally.

Principle 2: Psychology and design principles for data governance adoption

Understanding the human side of governance

While shift-left governance solves the technical problem, it doesn’t address the human challenge: most engineers don’t wake up excited to do data governance tasks.

This is where product design and psychology become critical.

Treat your internal products like they’re competing in the open market. Product design matters for internal tools just as much as customer-facing products.

Product thinking is about treating your data catalog as an ecosystem with multiple stakeholders, each with different motivations. Think of it like YouTube’s model: creators, consumers, and advertisers. If any group disengages, the whole system suffers.

For data governance, the typical groups are:

• Data consumers: Analysts, data scientists, and business users who need well-documented, discoverable, trustworthy data
• Data owners: Data engineers and product managers who want adoption and recognition for their data products
• Stakeholders: Compliance, security, and leadership who need verification that governance requirements are being met

How ICA Gruppen applied design thinking to drive governance

ICA Gruppen’s first governance attempt failed because it started with compliance requirements rather than user needs. Existing routines seemed to work well enough for users, so governance felt like extra homework with no payoff.

The team pivoted to a psychology-driven strategy, identifying their key user groups and designing for each group’s specific needs.

Then, ICA Gruppen rolled out a four-phased approach:

1. Find the champions. ICA Gruppen identified passionate data product managers who were already thinking about data quality and discoverability. 
2. Build the data marketplace. These champions created well-documented data products that attracted consumers looking for trustworthy data. 
3. Use consumer presence as an incentive. With users actively searching for data, it became easier to recruit more owners. 
4. Layer in governance requirements. Once teams were already on the platform and finding value, adding governance tasks became a much easier change to manage.

When data consumers loved the catalog, data owners wanted their assets included in it. Once owners participated, stakeholders finally had a governance structure that wasn’t theoretical.

With this intentional and incremental approach, governance went from “homework” to “help” that made people’s jobs easier.To see how ICA Gruppen brought this ecosystem model to life, watch the CONTEXT 2025 session: Driving data catalog adoption through psychology and design.

5 tenets to drive data governance adoption: Lessons from ICA Gruppen

Apply these five tenets to transform governance from an obligation into a resource teams actually want to use.

1) Obsess about users and product design, not governance frameworks

Governance tools are competing for attention in a crowded landscape of priorities. Think about your users’ motivations and position your governance tools as core products for them.

2) Use product design as psychology

Your design choices affect behavior in ways you might not expect. Make the user interface clean and modern. This signals that you respect users’ time and want the experience to be easy. A clunky interface might be all it takes for someone to decide they have more important things to do.

3) Use familiar language

Small changes in terminology can dramatically affect whether users feel the tool is for them. For example, ICA Gruppen changed its DataHub configuration to show BigQuery tables instead of dbt models because its key users (analysts outside the data team) were used to searching for data in BigQuery.

4) Deliver value before asking for effort

Features like data discovery and data lineage don’t require user input, but provide immediate value. Lead with these user benefits to build trust and momentum.

5) Build for your culture today, not tomorrow’s perfect state

Work with what you have and deliver incremental value. Adjust frameworks to fit your specific company setup.

Your data governance action plan

Follow this phased approach combining shift-left technical practices with psychology-driven design to implement data governance successfully.

Phase one: Discovery and planning

Understand your landscape before making technical decisions.

• Map your user ecosystem. Who are your consumers, owners, and stakeholders? What does each group need? What motivates them? Talk to real users in each category.
• Identify your champions. Look for people already passionate about data quality or documentation. These early adopters create network effects.
• Audit the current state. What governance activities happen today, even informally? Where are the pain points? What workarounds exist? Understanding existing patterns helps you meet people where they are.
• Choose your starting scope. Focus on published assets, high-performing teams, or high-value use cases for quick wins.

Phase two: Quick wins and momentum

Build trust by delivering value before asking for effort.

• Deliver no-effort value first. Enable search, discovery, and lineage that provide immediate benefits without changing behavior. This builds goodwill before you ask for contributions.
• Implement governance at the source. Set up integrations with tools teams already use, like dbt, Snowflake, Databricks, and Airflow. Capture metadata automatically wherever possible.
• Start with minimal requirements. The goal is reducing barriers to entry, not achieving perfect metadata coverage on day one. Expand requirements later once teams experience the benefits.
• Measure and communicate early wins. Track time saved, questions answered automatically, duplicate work prevented. Share these stories broadly.

Phase three: Scale and expand

Once you have momentum, thoughtfully expand the scope.

• Use early adopters to recruit the next wave. Let champions tell the story. Peer influence beats top-down mandates. When teams hear from respected colleagues about how governance made work easier, they engage.
• Layer in compliance requirements gradually. Now that teams see value, you can introduce additional requirements with less resistance. 
• Expand coverage based on demonstrated return on investment. Let success guide expansion. Focus on areas where governance has the highest impact, like reducing incidents, supporting critical processes, or enabling new analytics.
• Build feedback loops with all user groups. What’s working? What creates friction? What unexpected use cases are emerging? Iterate constantly based on what you learn.

Key data governance solution capabilities to look for

As you evaluate data governance platforms, prioritize features that align with the shift-left philosophy. ICA Gruppen relies on DataHub for capabilities including: 

• Automated metadata capture with native connectors that extract metadata from source systems without manual documentation
• Programmatic policy enforcement that defines governance rules as code applying automatically across your data estate
• Self-service access workflows that enable users to request access and data owners to approve via Slack or Teams with full audit trails
• Dynamic compliance forms that route annotation tasks based on asset type with real-time tracking of completion status
• Data contracts that formalize quality and schema expectations between teams with automated validation
• Structured properties that extend metadata with custom fields specific to your compliance or business needs

Not every organization needs every capability on day one, but the underlying philosophy matters more than the feature list. Choose platforms that reduce friction rather than add steps, that capture information where it’s created rather than asking teams to duplicate effort, and that make governance feel like infrastructure rather than overhead.

Making data governance work for your team

Data governance only works when it earns its place in people’s workflows. ICA Gruppen treated governance as an experience to design around existing behaviors—not a set of requirements to impose on resistant teams.

This approach is what allows data governance to improve business outcomes without sacrificing velocity. Done right, governance becomes your competitive advantage.

To go deeper, watch CONTEXT 2025 sessions for implementation ideas and data governance change management strategies you can apply to your program today.