Wonkhe reports King's and Napier experiences of AI feedback analysis

Updated Sep 07, 2026

A Wonkhe commentary published on 10 June 2026 describes how King's College London and Edinburgh Napier University used AI-assisted analysis of NSS comments. Its authors are Daniel Robson, Associate Director for NSS and PTES Strategy at King's, and Helena Lim, whose listed roles are at Queen Mary University of London and evasys.

This is a practice account, not an independent comparative evaluation. It explicitly discusses Student Voice AI supplied via evasys at Napier and recommends human validation. That commercial connection is relevant when reading this summary on Student Voice's own website.

What the account reports

At King's, Robson describes processing more than 1,700 comments in 2025 within two weeks of survey publication, producing over 5,000 category combinations. The account connects the analysis with wider assessment reforms and reports improving measures, but it does not isolate AI's contribution from those reforms.

For Napier, the article reports 1,701 comments, with at least one category assigned to 98 per cent of substantive comments and 5,340 comment–category combinations. School results were shared within one to two weeks of NSS publication. The earlier manual process is described as five to six working days of work; those are different measures of time, so this is not a like-for-like speed comparison.

Reported action-plan topics at Napier include timetable changes, programme restructuring, final-year groupwork, contact arrangements and practical experience. These are examples identified for local action, not evidence that every action had already improved outcomes.

What the account cannot establish

The article does not publish a validation dataset or a controlled comparison that separates analysis software from changes in staffing, assessment or institutional practice. The reported similarity between Napier's manual and automated themes is a local observation, not a numerical accuracy benchmark for every topic or cohort.

The source also reports assessment improvements at King's, not Napier. We have not reproduced those percentage changes here because the commentary alone is insufficient to establish their denominators, comparability and causal interpretation. Providers considering a similar approach should ask for the method behind any performance claim.

This is not an NSS methodology change or a regulatory mandate. Related OfS research announced in May and Jisc's proposed human-oversight pilot are separate developments, not endorsements of this implementation.

A reviewable local workflow

Our practical recommendation is to start with the decision that analysis should inform. Identify a defined comment set, its eligible population and the date by which a team can use the findings. Then agree:

• How reviewers will inspect source comments behind a theme.
• How ambiguous, multi-topic and minority experiences will be handled.
• What comparison is valid between cohorts or years.
• Who can authorise action and how students will hear the response.

Keep analysis turnaround, reviewer time and implementation time as separate measures. Faster categorisation does not guarantee earlier action, improved response rates or greater trust.

Our NSS open-text methodology and student comment analysis governance checklist explain questions to ask about coverage and review. Student Voice Analytics can support comment analysis, while institutions retain responsibility for interpreting evidence and evaluating changes.

FAQ

Q: Was this an independent product evaluation?

A: No. It is an institutional practice commentary that includes an evasys-affiliated author and describes Student Voice AI's use. The account is useful context, with that relationship disclosed.

Q: Does one to two weeks prove AI was faster than manual analysis?

A: Not by itself. Elapsed time to share results and working days spent coding are different measures. Compare equivalent activities and include human review time.

Q: What should teams take from the examples?

A: Define the question, review the output and connect it to an accountable decision. Treat reported local results as prompts for investigation rather than guaranteed outcomes.

Correction, 7 September 2026: we corrected the co-author, King's processing timeline and the institution associated with assessment improvements. Unsupported programme-count, September-timing and quotation claims have been removed, and the commercial context and limits of the comparison made explicit.

References

[Wonkhe]: "AI can help providers read and act on the student feedback they never usually get to" — Daniel Robson and Helena Lim. Published: 2026-06-10; current account checked 2026-09-07.