By Robyn Defelice
Early in my career, I misunderstood data governance. The word made me think of an old-timey governess from a finishing school, waiting to crack the back of your hand with a ruler if your data got out of line. I assumed it was some kind of formal oversight or arbitration process that mostly got in the way of otherwise awesome L&D ideas and solutions.
I also had a rather narrow view of data. Learning data came from an LMS, a survey, or an assessment. It was L&D’s data, so of course we were supposed to take it and use it. Not to mention how optimistic I was about everyone else’s data: I assumed that if HR, sales, operations, or another part of the organization had data that could help us understand a problem or create a better training product, they would want to share it. Why wouldn’t they? The data already existed, sharing it seemed free (data has no strings attached, right?), and we were going to use it to help the organization address performance.
I was not yet thinking about ownership, access, protection, context, or what could happen when data from an organization system or even org-sponsored survey was combined with learning data and used to inform decisions.
Everything I have written in this series so far has been about getting closer to data: taking stock of its value and what it means for L&D and using it to help us guide and advise the business about what we do. Governance is what protects the integrity of that work.
My understanding of what being data-capable actually requires has obviously grown: I know that finding, interpreting, and communicating data is only part of data literacy and data fluency. The other part is being able to make responsible decisions with and about data.
What Data Governance Actually Means
Data governance is how an organization establishes responsibility for its data. Governance addresses who owns data, who can access it, how it is protected, how data from different sources can be combined, and what it can responsibly be used to decide.
That is much less exciting than cracking knuckles with a ruler, but it is also far more relevant to the work of the Learning function.
You may be participating in governance and not realize it. I had participated in governance long before I understood that was what I was doing. I sat on councils that reviewed technology and made decisions about tools, access, and acceptable use. I worked on teams that had to decide who could see completion data, who could access assessment results, and whether managers should be able to view their direct reports’ scores.
Nobody called those conversations data governance. At least, not in a way that registered with me. They were simply the conversations we had to “get through” before an idea could move forward. What I did not recognize was that they were part of designing the solution.
Governance was not separate from the work. It was part of determining how the work could move forward without losing sight of the people, systems, and responsibilities behind the data.
Learning Data May Not Be L&D’s Alone
Learning teams have often used the data that they can readily access in reporting. For example, completion rates and assessment results, typically available through an LMS, help answer a familiar question: Did employees learn what the training was intended to teach?
Those data cannot answer every question an organization might ask.
- Did the training change performance?
- Did people become proficient faster?
- Did the change in performance reduce costs, improve efficiency, or affect a business result?
The advent of AI has made these broader questions harder to avoid. When organizations invest in AI-enabled tools and learning products, they may want to know what business outcomes changed as a result. Answering that question often requires information from outside the systems that the Learning function owns or manages.
Suppose leaders are trying to determine whether an AI-enabled sales program helped representatives improve the quality of their customer conversations. The LMS can show who completed the program and how they performed on an assessment. It cannot show whether customer interactions changed or whether those changes affected sales results.
That information might sit in a CRM, an HR system, or in reports managed by sales operations or finance. When decision-makers need information from those systems, governance comes into play.
It’s equally true that data from learning platforms might be useful to leaders outside of the Learning function: The Learning team might administer the LMS, design the learning, and pull the reports, but the data describes employees, their activity, their assessments, and sometimes their performance. Managing the system does not necessarily mean that every use of that data is decided by the Learning team.
In either direction, access to data is only the beginning. Learning managers and leaders must also determine whether the information is appropriate for the intended use and reliable enough to support the question being asked.
| Ask | Why It Matters |
| Who owns each data source? | The owner can help determine whether and how (in aggregate, individually, scrubbed of specific details, etc.) the data should be used. |
| Who approves its use? | Having technical access does not necessarily mean the Learning team has permission to use the data for a specific purpose. |
| Why was the data originally collected? | Data collected for one purpose may not accurately or appropriately support another purpose. This is why shared governance discussions on data are so important! |
| Do the systems define employees, teams, performance, and time periods consistently? | Data cannot be reliably compared or combined when the systems use different approaches to measurement. |
| Who can explain what the data means and what it does not mean? | Proper context helps prevent the Learning function from drawing conclusions that the data cannot support. |
These questions help Learning managers and leaders determine whether an analysis is credible enough to inform a decision. They can reveal that two systems define the same measure differently, that Learning leaders do not need direct access to the underlying data or are not permitted to access it directly, or that the available information cannot answer the question being asked.
They also help Learning leaders make more useful requests. Rather than asking another function to hand over its data, Learning can explain what it is trying to determine, identify what the learning data can and cannot show, and work with the data owner to decide what information is appropriate. That may result in direct access, an aggregated report, a limited data set, or an analysis completed by the function that manages the data.
A Good First Step
Before requesting access to data, especially if it’s your first time, do a little homework so you can have a more informed discussion—whether with a specific department or with a shared governance team/council.
Start with the business question, then identify what data your team already has, what additional information is needed, and who is responsible for it. Map that question to the data:
- What are we trying to determine?
- What can our existing data answer?
- What is still missing?
- Who owns that information?
- What form of access is actually needed?
As an example, I had a client I worked with who had decided to sunset a SharePoint resource library that provided customers with additional teaching content and activities. The decision was based on usage data showing little activity.
When I asked what period the data covered, I learned that it represented only three months: June through August. Because the primary customers were teachers, that timeframe raised an immediate question about whether summer activity reflected how the library was used during the school year. I asked for three years of usage or access patterns to see whether the data supported the decision over time. That was when we learned that the IT department retained only three months of SharePoint usage data for everything on the platform.
What changed the discussion was learning that the three-month window was not an analytical choice; it was all the data the organization retained. That points to another governance responsibility: how data is maintained and cleaned so it remains useful when a business or operational question arises.
AI Makes Data-Cleaning Harder to Ignore
Data-cleaning has often been treated as routine system maintenance, but it includes more than correcting records in a system. It is the process of identifying and addressing inaccurate, incomplete, duplicated, outdated, inconsistently defined, or incorrectly formatted data so that retained data remains reliable and usable. Data-cleaning also includes examining the standards and conventions used to enter and maintain the data, such as how items are named and tagged, what different statuses mean, when information is updated or archived, and who is responsible for those changes.
Routine cleaning may include removing inactive users, retiring outdated content, correcting duplicate records, or standardizing tags. In the Learning function, these tasks are commonly associated with an LMS, CMS, assessment management system, or survey platform. Other cleaning happens when Learning needs to answer a business or operational question and must confirm that the relevant records, definitions, and time periods can support the analysis.
AI has made the consequences of poor data quality more apparent. AI can process outdated, incomplete, unnecessary, or inconsistently defined data without recognizing that they are problems. It does not know that a course was retired three years ago. It does not know that completion records belong to employees who have left the organization. It does not know that the tagging in a content library was developed by several people using different approaches over many years.
For Learning teams, maintaining accurate and clearly defined data is therefore a practical governance responsibility, whether or not a formal governance framework exists.
A Good First Step
A useful first step is to conduct a basic data health check on one source your team manages. Start by deciding whether the goal is routine maintenance or preparation for a specific business or operational question.
For example, for routine data management, select the LMS, SharePoint, or a survey platform and identify:
- Who maintains it
- How information enters it
- How often it is updated
- What naming, tagging, status, retention, and archiving conventions are used
Then review the most important fields for outdated, missing, duplicated, or inconsistently labeled information. If the data is stored or processed by a vendor through an AI-enabled or cloud-based product, additional governance considerations apply.
Vendors & Data Rights
Most Learning functions rely on learning platforms, authoring tools, survey products, analytics tools, coaching platforms, and content libraries owned by third-party companies. Many of these products are cloud-based, and a growing number include AI features.
That means the governance conversation does not stop with the data inside the organization. Vendor contracts, data processing agreements, privacy terms, and terms of service may address questions such as:
- Who owns customer inputs, outputs, content, and metadata?
- May the vendor use aggregate or de-identified customer data?
- May customer data be used to develop or train AI features?
- Can additional companies or sub-processors access the data?
- Where is the data stored?
- How long does the vendor retain it?
- Can the organization retrieve or delete it?
- What happens when the contract ends?
There is limited published guidance developed specifically for Learning teams to help them navigate this as well-informed data stewards. What current information I could find comes from higher education, European government, privacy, and technology procurement.
The Higher Education Community Vendor Assessment Toolkit, or HECVAT, is one useful resource. Although it was not designed specifically for corporate Learning, its current version includes questions about vendor security, privacy, data ownership, sub-processors, and AI use. HECVAT, along with much government and privacy guidance, recommends involving experts beyond your organization’s legal counsel when reviewing how a vendor will process, retain, share, return, or delete organizational data.
A Good First Step
Select one vendor that your Learning function currently uses, preferably one with a cloud-based or AI-enabled product. Locate the current contract, data processing agreement, privacy terms, and terms of service. Then identify who within the organization should review them with you, such as procurement, legal, IT, privacy, information security, the business owner, or a data governance council.
Use the questions above to guide that review and document any gaps, unclear terms, or decisions that require follow-up. Also determine who will monitor future changes to the product, its AI features, sub-processors, or terms, and when the agreement should be reviewed again.
If no formal vendor-governance process exists, this review can provide a practical starting point for bringing the appropriate people together before Learning adopts, renews, or expands use of the product.
Governance Is Part of Data Capability
Governance is already part of the work of the Learning function, whether we call it that or not. Governance shapes how we manage our own data, request data from others, prepare it for use, and understand what happens to it when vendors are involved. AI has made those responsibilities more visible, but it did not create them.
That is why governance belongs in a series about data literacy. Recognizing data, interpreting it, and asking better questions are not enough if we cannot also use the data responsibly. Together, these capabilities help the Learning function guide and advise the business. They are the difference between having data and being capable of using it well.
Boost Learning Impact with Evidence-Informed Design!
Join us October 14-15 for the Evidence-Informed Design online conference, where we will survey proven approaches to creating learning that works. Sessions will explore ways to reduce cognitive load, improve learner recall, and deepen engagement. Register today!
Image credit: ismagilov
