AI in Assessment: Moving from Experimentation to Responsible Implementation

While much of the conversation has focused on how AI is reshaping assessment, a more immediate question is now emerging: how are organisations beginning to implement it in practice?

Across the skills sector, organisations are beginning to move from exploration to early implementation, but questions around trust, quality, and control remain.

For awarding organisations and training providers, the challenge is not whether AI has a role, but how to introduce it in a way that maintains confidence, aligns with regulatory expectations, and supports assessors in practice.

Starting with a real assessment challenge

One area where AI is being explored is in the marking of extended written responses and portfolio-based assessments.

These assessment types are often:

• time-intensive
• cognitively demanding
• difficult to scale consistently

Maintaining quality and consistency, while managing increasing volumes of learner submissions, continues to be a challenge across awarding organisations.

Rather than approaching AI as a broad transformation, some organisations are beginning with targeted use cases, focusing on specific assessment pain points where support is most needed.

From manual marking to AI-assisted workflows

In practice, this has led to the emergence of AI-assisted marking models, where technology is used to support, rather than replace, assessor judgement.

A common approach involves:

• generating structured, criteria-aligned feedback using AI
• enabling assessors to review, edit, and approve outputs
• retaining final decision-making with the human assessor

This “human-in-the-loop” model is increasingly seen as a more viable route for adoption, particularly in regulated environments where accountability and explainability are critical.

In a recent panel discussion hosted by Test Community Network, Kaylë Brightwell, Director of Education at Safety Training Awards (STA), highlighted that the intention is “not to hand marking over to AI, but to use it in a way that supports consistency while keeping professional judgement central.”

Early insights from implementation

While evidence is still emerging, early-stage pilots suggest some common themes:

• Reduced time spent drafting feedback, allowing assessors to focus on review and quality assurance
• Greater consistency in feedback, particularly against defined marking criteria
• Improved clarity for learners, in some cases leading to increased resubmission rates
• Continued reliance on assessor oversight, particularly for final decisions

While these insights are encouraging, it is important to note that outcomes vary depending on assessment type, implementation approach, and the quality of training data used.

At this stage, much of the evidence remains early-stage and contextual, rather than standardised across the sector.

Trust, regulation, and the role of the assessor

One of the central considerations in adopting AI in assessment is trust.

Questions around fairness, bias, explainability, and accountability remain critical, particularly in high-stakes or regulated qualifications.

Regulatory bodies such as Ofqual have already signalled that AI cannot operate as the sole mechanism for marking, and that human judgement must remain central to assessment decisions.

This places emphasis on models where:

• AI supports consistency and efficiency
• assessors retain control and accountability
• decisions remain explainable and open to challenge

In this context, AI is not replacing the assessor, but changing how their role is carried out, shifting from generating feedback from scratch to reviewing and refining AI-supported outputs.

A gradual, use-case-led approach

What is emerging across the sector is not a single model, but a more gradual approach to adoption-reflected in guidance and early activity from regulators, awarding organisations, and training providers.

Rather than large-scale transformation programmes, organisations are:

• starting with specific assessment types
• piloting within controlled environments
• evaluating impact before scaling
• aligning implementation with existing quality assurance processes

This approach reflects a wider emphasis on building confidence over time, while managing risk and maintaining standards.

What this means for the sector

The conversation around AI in assessment is often framed in extremes, either full automation or complete hesitation.

In practice, a more balanced approach is beginning to take shape.

For awarding organisations and training providers, the focus is shifting towards:

• identifying where AI can provide meaningful support
• introducing it in a controlled and transparent way
• ensuring alignment with regulatory and quality expectations
• building trust through practical, real-world use

The question is no longer simply whether AI has a role in assessment, but how it can be implemented responsibly within existing systems.

Final thought

AI in assessment is still at an early stage of adoption, and many questions remain.

However, early exploration across the sector suggests that when introduced carefully, with clear boundaries, human oversight, and a focus on real assessment challenges, it can begin to support more consistent, manageable, and scalable marking processes.

The opportunity now is for organisations to move beyond theory, and to explore where this approach may, or may not, work within their own assessment models.

By Kavitha Ravindran, Co-Founder & Chief Growth Officer at sAInaptic