# Emotuit

> Emotuit measures whether people are actually paying attention to digital content — training modules, e-learning courses, webinars and recorded meetings — and pinpoints the exact moment and section where attention breaks. Completion rates tell you someone finished; Emotuit tells you where they checked out, what was on screen when it happened, and what to change.

Emotuit was founded in 2014 by Ross Jones and rebuilt on modern tooling. It combines behavioural signal analysis (tab-switching, focus loss, dwell time, window state) with optional client-side facial expression analysis to produce a continuous engagement index mapped against the specific content a person was viewing at each moment.

The core output is content-level, not person-level: not "learners are disengaged" but "60% of learners disengage at module 3, slide 14, during the section on incident reporting."

## Products

- [Analyse](https://emotuit.co/analyse): Post-event analysis of recorded sessions. Upload a Zoom recording, meeting or training session video and receive per-participant engagement scoring, an interactive engagement timeline, transcript-aligned flagged moments, AI coaching suggestions tied to specific timestamps, and a downloadable PDF report. Priced per analysis.
- [Learn](https://emotuit.co/learn): A JavaScript SDK (under 50KB) that drops into any LMS or learning platform via a single script tag. Provides real-time engagement tracking, per-section content effectiveness scoring, drop-off point identification and disengagement prediction. Works with any LMS, SCORM or xAPI platform. Priced as an integration fee plus monthly per seat.

## How it works

- Face detection and 68-point landmark extraction (MediaPipe / DeepFace)
- Seven-emotion probability vector based on Ekman's Facial Action Coding System (happiness, sadness, anger, fear, surprise, disgust, contempt)
- Per-user baseline calibration — every reading is a deviation from that individual's neutral expression, not a population average, so naturally downturned features are not misread
- Behavioural signal capture via native browser APIs (Page Visibility, Intersection Observer): tab-switching, face-loss, window minimisation
- Continuous engagement index (0–1) fused from emotional and behavioural signals
- Content-timestamp correlation producing per-section engagement heatmaps
- Retroactive disengagement labelling that builds a predictive model of the emotional signature preceding dropout

## Privacy and compliance

- All face detection and emotion classification runs client-side in the browser via TensorFlow.js or ONNX Runtime Web. No facial images or video are transmitted — only numerical engagement scores.
- Explicit opt-in consent is required before webcam activation.
- Content-level analytics function with no personally identifiable information at all.
- Air-gap compatible: the pipeline can run fully offline and on-premise with zero external network dependencies.
- Designed for GDPR, COPPA and institutional data governance requirements.

Note on the EU AI Act: Article 5(1)(f) prohibits AI systems that infer emotions from biometric data in workplace and education contexts. Emotuit's behavioural signal layer does not rely on biometric emotion inference and is intended to be deployable where facial affect analysis is not permitted. Buyers in EU workplace or education settings should confirm their configuration with their own legal counsel.

## Research validation

Each architectural component is supported by published research:

- Ekman seven-emotion FACS model — Springer, Smart Learning Environments, 2025
- Normalised emotion vector (softmax) — PMC / Multimedia Tools and Applications, 2022
- Per-user baseline calibration — Frontiers in Psychology, January 2026
- Tab-switching ("cyberloafing") as the strongest single predictor of online disengagement — ScienceDirect, 2024
- Multi-signal fusion improving classification accuracy from 91.5% to 94.6% — OUCI/DNTB online learning study
- Retroactive labelling achieving 73.3% disengagement prediction accuracy with 40% of the session remaining — Boote, Agarwal & Mostow, 2021
- Content-timestamp correlation framework — PMC MOOC engagement framework, 2021

## Who it is for

- Corporate learning and development teams that need to prove training impact beyond completion rates
- E-learning platforms and course businesses diagnosing where learners drop off
- Higher education and institutional digital learning teams identifying at-risk learners
- Internal communications and enablement teams measuring town halls, all-hands and webinars
- Market research, UX and media testing teams needing browser-native facial coding
- Government, defence and regulated training environments requiring air-gapped deployment

## Known limitations

- Facial obstructions (glasses, beards, poor lighting) reduce detection quality; partial occlusion is handled by MediaPipe and baseline calibration reduces the impact
- Cultural variation in expression intensity is normalised by per-user baselines rather than population thresholds
- Spontaneous and posed expressions differ in classification accuracy; Emotuit tracks change over time rather than absolute emotional state
- Webcam quality varies; minimum resolution is detected with graceful degradation

## Contact

- Email: ross@emotuit.co
- Founder: Ross Jones
- Website: https://emotuit.co