Blog

Sourced, dated and updated. Mostly about the gap between what training completion data says and what actually happened.

Article 5(1)(f) and education institutions →

Schools, universities and training providers are named explicitly in the EU's prohibited AI practices. What the rule covers and what stays lawful.

Lawful alternatives to emotion recognition for measuring engagement →

If facial emotion analysis is off the table for legal or cultural reasons, these are the signals that actually work — and the research says several of them outperform it.

Average e-learning completion rates →

What counts as a normal completion rate for corporate e-learning, compliance training and MOOCs — and why the benchmark you're chasing may be the wrong number entirely.

Beyond completion rates →

Seven engagement metrics that tell you what completion rates cannot — which section lost people, when, and what to change. With the browser APIs to collect each one.

The 90% completion problem →

Mandatory compliance training has the highest completion rates in corporate learning and the weakest evidence of behaviour change. Why that should worry you.

How to find where learners drop off →

A practical method for locating the exact point your training loses people — what to instrument, how much data you need, and how to read the result.

Is emotion recognition legal in the workplace? →

In the EU, no — prohibited since February 2025 with fines to 7% of global turnover. In the UK and US the picture differs. The answer by jurisdiction.

Why learners drop out of online courses →

Six causes of drop-off in online training, what each one looks like in the engagement data, and which of them you can actually do something about.

Zoom attention tracking alternatives →

Zoom removed its attention tracking feature in 2020. Here's what it did, why it went, and what you can legitimately use now to measure whether people are engaged.

Does Zoom have attention tracking? →

No — Zoom removed the feature in April 2020 and has not reinstated it. What it did, why it went, and what hosts can actually see now.

A DPIA template for learner engagement analytics →

A practical Data Protection Impact Assessment structure for engagement monitoring in training and education, with the AI Act questions built in. Copy, adapt, use.

Emotion recognition vs behavioural analytics →

These two get sold as the same product. Legally they are not remotely the same, and the difference decides whether you can deploy in the EU at all.

The Kirkpatrick model explained →

The four levels, what each actually measures, where the model still earns its keep after 65 years, and the specific things it cannot tell you about digital content.

L&D KPIs executives actually care about →

The metrics that survive a board conversation, the ones that get discounted on sight, and how to translate learning data into the language finance already uses.

The limits of facial emotion recognition →

An honest account of what facial emotion recognition can and cannot do — accuracy ceilings, the posed-versus-spontaneous gap, and the contested science.

How to prove training ROI to a CFO →

Only 29% of L&D leaders feel confident proving ROI. The problem is usually presenting numbers finance discounts on sight. What they accept instead.

Signs of learner disengagement →

What disengagement looks like in behavioural data before someone actually leaves — the observable signals, in the order they typically appear.

Identifying disengaged learners early →

Building an early-warning signal from behavioural data — what predicts drop-off, how much warning you get, and why you should usually act on the content instead.

UK vs EU rules on emotion recognition at work →

The EU prohibits it outright. The UK does not — but the ICO has been pointedly sceptical, and UK GDPR still applies. A practical comparison for multi-jurisdiction employers.