03 / MSc PROJECT / CAPGEMINI FROG

Generative AI coaching for sales pitches

I researched how people practise sales pitches, then designed and tested an AI coaching experience for rehearsal and feedback.

Context: MSc User Experience Engineering
Industry partner: Capgemini frog
Focus: Sales pitch practice and feedback
My contribution: Research, prototyping and evaluation

EdenAI Pitch Practice interface showing two virtual coaches in a live coaching session — Working coaching experience built in Soul Machines, with an LLM connected via OpenAI and grounded in a sales training dataset I curated.

30survey participants

30user interviews

10participants in final prototype testing

RESEARCH AND DIRECTION

Practice needed to feel closer to the real thing.

I looked at how people prepared for sales pitches, how they received feedback and what made practice useful. Alongside surveys and interviews, I observed a training session and reviewed existing learning approaches.

57%

of survey respondents said they felt uncomfortable delivering a sales pitch.

60%

said existing sales training methods, including e-learning, needed improvement.

90%

felt AI should support sales training, not replace human coaching.

TWO PROTOTYPE ITERATIONS

Testing changed the experience from video-led to interactive.

01 / tED

Video-based practice

The first clickable prototype simulated pitch training through video-based learning. Feedback from nine participants helped shape the next version.

02 / EDENAI

Live AI coaching

The later prototype let people practise one-to-one with a virtual coach for ten minutes. An LLM connected via OpenAI used the curated sales training dataset to give live coaching and feedback.

EdenAI interface with a virtual coach giving spoken feedback during a pitch practice session — Feedback was part of the live practice experience, not a separate learning module.

FINAL PROTOTYPE TESTING

What testing showed

8/10Saw value in the concept

7/10Found the later version more engaging

8/10Were interested in using it

10/10Supported further development

Exploratory evaluation with 10 participants. Most participants saw the experience as a useful supplement to human coaching.

WHAT STILL NEEDED WORK

The coaching experience had clear limits.

Participants noticed delays in the AI’s responses and sensitivity to background noise. The prototype also could not reliably assess tone, pace, facial expression or emotion. These limits shaped the next questions for development: how to make feedback faster, improve the training data and test the experience with a wider range of learners.

The project brought research, prototype evaluation and AI product design together around a practical question: where could a virtual coach help people practise, and where did human support still matter?