Michelle Dunivan explains why the best data governance is invisible, how to scope it by use, and what it enables at Best Friends Animal Society.
During a recent conversation on The Executive Outlook, Michelle Dunivan challenged one of the most persistent assumptions about data governance: that good governance has to be visible.
Ask most people in an organization what they think of data governance, and the reaction is familiar. It sounds like paperwork, approvals, and someone telling them no. It sounds like the part of the work that slows everything else down.
Michelle sees it differently.
She is the Senior Director, Insights & Analytics, at Best Friends Animal Society, where she leads work across data, data science, governance and AI. Her view of data governance turns the usual assumption inside out. Done properly, almost nobody should notice it is there.
The less people have to interface with governance, the better. If the work has been done properly in the background, colleagues should receive something they can trust without needing to understand every process that made it safe, secure and reliable.
Before Best Friends, Michelle spent six years working with one of the largest judicial districts in the United States. She also helped build the Shelter Pet Data Alliance, a data-sharing platform serving thousands of animal welfare organizations. Her PhD in communication shapes the way she approaches the work, particularly the relationship between human behavior, trust and measurable outcomes.
Michelle did not enter data through engineering. She began with an interest in communication and human behavior, particularly how messages influence people and why.
The deeper she went into the research, the clearer it became that observation alone was not enough. She wanted to measure those effects properly, which meant developing a stronger understanding of statistics and research design.
That became an unexpected foundation for a career in data.
When she finished her education, Michelle looked for a role that could bring together the two things she cared about most: understanding the human experience and measuring what was really happening.
In many ways, that is what data teams inside organizations are trying to do. They are using information at scale to understand behavior, identify patterns, and make better decisions.
Michelle began that work in the courts, where she helped teams understand what was happening inside courtroom processes and where those processes could improve. About five years ago, she moved to Best Friends Animal Society, an organization that had already invested significantly in using data to support animal welfare across the United States.
Animal welfare does not have a single federal authority defining how sector data should be collected, interpreted or reported.
That makes consistency especially important.
The first step was not technology. It was agreement.
Michelle's team worked with other national organizations to establish common definitions and align on the information being collected. That consistency allows organizations to share data more effectively and prevents shelters from having to report the same information in several different ways.
This is where data governance begins at its most fundamental level.
If one organization uses a phrase such as return to field, everyone involved needs to understand exactly what that means and which situations it includes. Without that shared understanding, the data may look consistent while actually describing different things.
Governance therefore starts before the platform, dashboard or model. It starts with agreement about meaning.
The value of that data governance work becomes clear in the insights it has helped the sector produce.
Best Friends collects data from a large number of shelters across the United States. That allows the team to identify where the largest life saving gaps exist and what types of interventions may help close them.
The organization also combines shelter data with external sources such as US Census information and data from the Centers for Disease Control. This helps teams understand how broader community conditions may influence animal welfare outcomes.
One analysis provides a clear example.
Several free adoption platforms in the United States allow shelters to upload photos and profiles of animals available for adoption. Those platforms attract significant audiences, but some shelters were not using them.
Michelle's team analyzed the difference.
They found that using those platforms increased adoption rates by around five percentage points. For a typical shelter where roughly twenty five to thirty five percent of animals may be adopted directly, a five point increase can make a meaningful difference in the number of animals leaving alive.
The insight could then be shared across the shelter network so organizations could understand the potential value of participating.
That is where governance becomes more than a control function. Common definitions and trusted data make analysis possible, and the analysis can then influence action.
Michelle also describes a challenge familiar to many analytics leaders.
If you only analyze data from organizations willing and able to provide it, there is a risk that the picture you create does not represent everyone who is missing.
Her team needed to understand what was happening in shelters that had not submitted data. The mission required a broader picture of animal welfare across the country, not only a picture of the organizations already participating.
The team developed a statistical method to estimate intake and outcome patterns for those shelters and tested the method for accuracy.
It performed particularly well when identifying whether a shelter was no kill, reaching approximately 98 percent accuracy. That allowed the organization to build greater confidence around where support might be needed, even when a shelter had not provided data for several years or had never provided it at all.
The approach was later developed into a more automated machine learning process.
But the next challenge was not mathematical.
The business had to trust it.
Building an accurate model was one job. Helping people feel confident enough to act on it was another.
Accuracy alone was not enough to create trust.
Michelle's team had to demonstrate that the necessary controls and protocols were in place and then make the output useful to the people making decisions.
The answer was not to give everyone a deeper technical explanation of the model. It was to build tools around the model that connected its output to the work people were already doing.
Someone deciding which communities might need ordinance changes has a different need from someone trying to identify which program could create the greatest life saving impact for a particular shelter.
People did not need to understand every detail of the machine learning model. They needed access to the right information within their existing systems and workflows.
That principle is important beyond animal welfare.
The rigor can sit underneath. What reaches the user should be clear, relevant and usable.
Best Friends is actively investing in AI, but the rollout did not begin by simply giving everyone a collection of tools.
It began with philosophy.
Michelle's team first established what the organization believed about AI and then communicated those principles clearly. Employees were encouraged to experiment with approved tools, but they were also given boundaries around ethical use, security and the types of information that could be entered into different systems.
Practical questions mattered. Which data can go into which tool? Can employees use personal accounts? What happens when someone wants to use a system outside the approved options?
That foundational AI governance gave employees something important: clarity.
People understood that they were encouraged to explore AI, but they also understood where the boundaries were.
In this case, AI governance did not prevent adoption. It helped create the confidence for adoption to begin.
Training then introduced another leadership challenge. Not everyone works in front of a computer all day. Some Best Friends employees spend most of their time in the field caring for animals and may only use a computer occasionally.
The goal cannot therefore be to turn every employee into an advanced AI user.
The better question is how those employees can benefit from AI without forcing more computer time into their work. In some cases, that may mean working through managers and division leaders to identify where AI can improve the process around employees rather than changing the nature of their jobs.
Data leaders have spent years trying to make information more accessible and create genuine self service.
AI could accelerate that significantly.
But greater access also intensifies an old concern: if everyone can use the data, how do you know they are using it consistently and that the outputs influencing decisions can be trusted?
AI does not eliminate that challenge. It makes it more important.
Michelle's approach is not to choose between experimentation and control. It is to create room for both.
People need enough freedom to explore because AI is still new to much of the organization. At the same time, clear boundaries need to exist around risk, security and responsible use from the beginning.
As experimentation moves toward implementation, those controls become more important.
If something will be used across a team, adopted throughout the organization or shared externally, the organization needs clear quality standards and approval processes. When those expectations have already been established, employees do not reach the final stage asking who needs to approve the work or what standard it must meet.
The process is already understood.
Where possible, AI can even help automate parts of those reviews.
One of Michelle's most practical ideas is about scope.
Data governance programs can become overwhelming when they begin with the ambition to govern the entire data estate equally.
Best Friends takes a more selective approach.
The organization focuses first on the areas where people need to share the same understanding and where trust matters to an important use case. Michelle is clear that not every piece of information needs the same process, definition or level of governance.
The scale alone makes that necessary.
An organization can have tens of thousands, potentially hundreds of thousands, of tables. Applying the same level of governance to every one of them would consume enormous effort without creating equivalent value.
The level of governance should therefore be connected to use.
If a data asset will feed an AI system, support an analysis or influence an important decision, it needs the appropriate definitions, security, context and controls.
If nobody uses it, it does not need the same level of attention.
There is also a useful consequence to that approach. If an ungoverned data asset lacks the definitions or security information required for important use, it simply cannot move into that use case yet.
The question then becomes practical: what needs to happen to make this data ready?
Governance becomes a path toward trusted use rather than a reason to say no.
Michelle offers another insight that matters for leaders trying to build internal support for governance.
Most people do not want to design the governance process themselves.
They simply want to know what the process is.
Data is complicated, and governance requires decisions that many employees do not have the expertise, context or time to make. When a governance team takes on that complexity and gives people a clear route forward, work can move faster.
Instead of asking employees to determine every requirement themselves, the team can say: this is the process, this is what we need from you and these are the checks that have to happen.
That clarity removes decision making burden from the user.
But invisible governance does not mean governance happens in isolation.
The team still needs partners across the organization. Definitions must reflect what people are actually entering into systems. Data scientists need enough context to use the information correctly. And when data is going into AI, the system needs the semantic layer required to understand what the values actually represent.
Without that context, more data does not necessarily create better intelligence.
Three leadership lessons from Michelle's approach can apply well beyond animal welfare.
First, scope governance according to use and risk rather than trying to govern everything equally. Identify the information feeding important analysis, AI systems and decisions, then give those assets the level of attention they deserve.
Second, absorb complexity wherever possible. A governance function creates more value when it makes responsible use easier for everyone else. The goal should not be to make employees think constantly about governance. It should be to make the trusted path obvious.
Third, allow exploration within clear boundaries, then strengthen the guardrails as work moves toward implementation and publication. People need room to learn and experiment with AI, but anything that begins influencing wider teams, customers or external organizations needs stronger quality and accountability controls.
That balance matters because data democratization and governance are not opposing goals.
Done well, governance is what makes democratization sustainable.
None of this is particularly glamorous. Michelle readily describes governance as the less exciting part of the work. But behind every trusted model, reliable insight and responsible AI application sits a layer of decisions about definitions, security, quality, context and accountability.
The strongest governance may be the governance most employees never notice.
They simply know that when something reaches them, they can trust it.
Are you a CIO, CDO, data leader, AI executive or technology founder with a perspective that can help global leaders make better decisions?
Share your experience, leadership lessons and industry insights with The Executive Outlook and join the conversations shaping the future of data, AI and digital transformation.
Listen in for business strategies and interviews with business experts to help you build a business and a life that you love.
About Us
