The ethical application of AI in learning analytics

With the launch of our newly updated learning analytics code of practice that now includes AI considerations for data teams, we’ve been exploring the responsible use of artificial intelligence (AI) to enhance student engagement monitoring and support organisations to identify learners who may benefit from additional support.

The code sets out the guiding principles of learning analytics, namely that analytics should:

• Benefit students
• Be based on sound data
• Be governed by accountable people

Building on existing learning analytics capabilities, we will use AI to analyse engagement data and identify patterns, trends, and indicators that may suggest a student requires attention, whilst ensuring essential human oversight.

Our aim is to provide teams with timely, evidence-based insights that enable earlier and more informed interventions, helping organisations to take a proactive approach to supporting student success, retention, and wellbeing.

This companion piece provides more in-depth guidance and practical examples around the ethical use of AI in learning analytics, in particular where AI changes the nature of the analytics process.

Straightforward applications

AI is already being used within learning analytics products and workflows in several ways:

• Advanced prediction: models that go beyond traditional descriptive approaches and are not always fully transparent
• AI-assisted interpretation: tools used to summarise or explain analytics, acting alongside skilled analysts
• AI-enabled data capture: tools used to analyse written content provided by students such as surveys, expanding analytics beyond structured data such as attendance or grades

For many uses of AI, the code’s existing requirements around data protection, validity, responsible intervention and data minimisation provide a clear basis for safe practice.

In practice, this means:

• Data shared with AI tools must meet the same privacy and data protection standards as any other analytics processing
• Processes must be validated, monitored and reviewed by qualified staff
• Interventions must remain proportionate, auditable and under human control
• Data collection must be limited to what is necessary

Third-party tools and stewardship

Where AI tools are provided by third parties, organisations remain responsible for how student data is used.

Before AI tools are used in learning analytics, organisations should understand:

• What data is processed
• Whether data is retained or reused
• How outputs are generated and checked
• What safeguards apply

Explainability

Where models cannot be fully inspected, we recommend colleagues apply the code’s emphasis on transparency through clear, meaningful explanations of outcomes rather than relying solely on technical detail.

We often hear from peers across the sector about the challenges with systems and solutions that operate like a black box.

‘Understanding’ a model means being confident in its behaviour through validation and monitoring, not necessarily being able to read its internal workings.

Pre-emptive prediction and purpose limitation.

AI can forecast future scenarios and outcomes for students, which may help organisations offer earlier and more targeted support.

However, this also raises the risk that students are labelled unfavourably, treated unfairly, or subject to decisions that go beyond the original purpose of learning analytics.

Where AI is used to generate predictions, summaries or interpretations about students, students should have a meaningful route to understand, question or correct the information used. This is a practical application of the code’s emphasis on transparency, access and responsible intervention.

In line with the code’s focus on student benefit and supportive intervention, predictions and classifications should prompt offers of support, not restrictions, penalties, or disciplinary action.

Interpretation and validity

Where AI is used to support analysis, its role should be to assist people rather than replace them.

AI-generated explanations can sound convincing even when they are wrong or misaligned, so outputs must be treated as drafts or supporting material, not decisions.

We recommend a trained and experienced member of staff remains responsible for checking the output, interpreting context and making any judgement or action that follows. Transparency arrangements should also make clear whether AI is used to analyse student data, as well as who can access that data.

Data capture and minimisation

AI makes it easy to analyse large amounts of written content that would once have been impractical to review at scale.

However, the same ease of analysis is exactly why the code's principle of data minimisation needs robust application: student content should only be analysed when it is provided for that purpose, and only to the extent necessary.

A cross-cutting risk: bias

The involvement of AI within learning analytics brings the issue of bias into sharper relief. The relevant response sits within the code’s existing commitments to validity, responsible intervention and minimising adverse impacts.

We recommend managing bias as part of ongoing governance.

AI-enabled analytics can introduce or amplify bias in several ways:

• Complex models may identify indirect proxies for protected characteristics, even where those characteristics are not included directly
• Generated interpretations may present biased or unsupported conclusions in confident, persuasive language
• Interventions based on predictions may affect some groups more than others, even where the underlying model appears accurate overall

Taken together, the safeguards recommended in the code of practice provide the basis for managing these risks. In practice, organisations should test AI-enabled analytics not only for technical performance, but also for whether they produce fair, proportionate and beneficial outcomes for students.

Where risks are identified, the response should be to adjust the model, narrow the use case, strengthen human review, or stop using the approach.

Our plans for AI in learning analytics

As part of our product development, we will adopt an ethical, human-centred approach in which AI acts as a decision-support tool rather than a decision-maker. Any insights generated by AI will be transparent, explainable, and designed to assist staff in understanding potential areas of concern, while ensuring that professional judgement remains central to all intervention decisions.

Above all, we are committed to ensuring that the technology is used fairly, responsibly, and in a way that delivers value to both organisations, teaching teams and learners.

Data Matters 2027

Our principal technical consultant Jim Keane will be at Data Matters 2027, talking about the human side of learning analytics with South Bank university.

Go and see the learning analytics team in the Data Matters demo zone to discuss your needs and see what our platform can do.

Tom Moule, our senior AI specialist, is also hosting a breakout session at Data Matters about AI, student insight and the next generation of learning analytics.

We’ve sent free ticket codes out to all our member organisations, so check with your internal teams if you haven’t seen them.