7 November 2026
Welcome to my weekly roundup of eLearning, LMS and AI news. Each week, I share new product releases, emerging technologies and research findings, with a focus on what they mean for workplace learning and frontline training. I’ll look at what the data tells us about how people learn and apply new skills on the job.
Synthesia reports early results for AI role-play.
In a voluntary pilot with Hays, Synthesia reported over 70% engagement and a 29% average improvement in measured sales skills among the pilot cohort.
A retail pilot could test one conversation, such as explaining the value of a new watch, and have store managers assess the same behaviours before and after practice.
This is directly relevant to your Synthesia partnership. A retail pilot could test one conversation, such as explaining the value of a new watch, and have store managers assess the same behaviours before and after practice.
A large AI tutoring trial found weaker outcomes.
An October working paper on a randomised trial involving 2,379 undergraduates found that access to a course-integrated AI tutor reduced final grades and LMS participation in comparisons between sections of the same course.
The study was conducted in higher education. edsi.umd.edu
If you are using AI chatbot, measure whether employees can recall and apply product knowledge a week later. Usage and answer speed tell you how the tool is used. A follow-up assessment tells you what people learned.
What happened in the university trial
The October 2026 working paper covered 2,379 undergraduates and 30 instructors at a large US public university.
The GPT-4o tutor was built into the LMS and drew on course materials selected by instructors. In the same-course comparison, students offered the tutor scored 4.12 points lower on final grades and recorded substantially less LMS participation.
Only about 15% of students offered the tutor used it. The experiment measured the effect of offering access, and the researchers couldn’t establish whether the lower grades came from tutor use, changes in study habits, or other factors. edworkingpapers.com
It means that the classes offered the AI tutor did worse on average, even though only about 15% of students offered it actually used it.
What other studies found
Study Result, High-school maths · nearly 1,000 students
Students with a ChatGPT-style tool scored 48% higher during practice and 17% lower on the subsequent exam without AI, compared with students who never had access. A version that offered teacher-designed hints largely eliminated the exam penalty. pmc.ncbi.nlm.nih.gov
Harvard physics · 194 students
A tutor designed specifically for two physics lessons produced more than double the median gains on immediate tests compared with active classroom instruction. The study measured learning just after the lessons, leaving long-term retention open. ---OPEN
Professional development · 78 participants with complete dataAn online course and a custom AI tutor each improved knowledge scores. The researchers found no clear additional knowledge gain from combining them. The small sample and uneven dropout make that result preliminary. mdpi.com
Separate two jobs in a retail pilot.
A knowledge assistant helps an associate verify an approved product detail in seconds. A learning coach asks the associate to explain the product in their own words, gives feedback and lets them try again.
For a product launch, you could measure lookup speed for the first job and, a week later, use a new client scenario without AI to see whether associates can explain the craftsmanship and recommend the right model.