Customer Success · Leadership Development & L&D Analytics
Locked-away feedback. Unlocked.
World Class Manager runs online management courses for organisations around the world. Their participants complete questionnaires before, during, and after every programme. Thousands of responses. Years of data. And almost none of it being used.
The Problem
A soup of data. No way to taste it.
Wayne Clarke, CEO of World Class Manager, put it plainly in our first conversation. He had questionnaire data going back years across 150+ courses. He knew it was valuable. But when his publisher asked him to include data in a book he was writing, he had to admit the truth: he didn't know what he had, and he couldn't get to it.
“When we wrote the book in 2023, the publisher was saying, let\u2019s include some of your data. To be honest, I don\u2019t know what we have. I didn\u2019t know if it\u2019s good.”
Wayne Clarke, CEO, World Class Manager
This wasn't a data problem. It was an access problem. The data existed. Getting to it in any useful way did not.
Trapped behind a row limit
Moodle’s reporting tool hit a hard ceiling of 135,000 rows and had no filters for date, department, or organisation. Asking “what are the most common challenges our participants face?” was simply not possible.
150 courses, 150 manual exports
Getting all the data out meant logging into each course individually and downloading a separate file every time. Nobody was going to do that consistently. And they didn’t.
Free-text going to waste
The richest feedback — the open participant comments — was being copy-pasted into ChatGPT in chunks. Results were generic, lacked context, and couldn’t be trusted. The real gold was being left in the ground.
No story to tell
Without being able to connect pre-course challenges to post-course outcomes, demonstrating the programme’s impact to clients and in marketing was close to impossible. The data existed to prove it. They just couldn’t reach it.
What We Built
Ask a question. Get a real answer.
We extracted all the questionnaire data from their Moodle system, cleaned it up, consolidated it into a single analytics database, and built an AI assistant on top of it. Now anyone on the team can type a plain-English question and get back a clear answer — in text, charts, or tables — in under 20 seconds.
No SQL. No spreadsheets. No waiting for a report. Just ask.
Get all the data out, automatically
The first job was eliminating the manual export process. We built an automated pipeline that securely connects to the Moodle system and pulls all feedback data without anyone having to touch it. All 150+ courses, in one place.
We then handled the messiness baked into years of inconsistently built questionnaires. Using fuzzy matching, “Participant Strengths” and “Areas of Confidence” are correctly recognised as the same question. Everything ends up clean, consistent, and comparable across all courses and all years.
Free-text responses got special treatment. Rather than treating open comments as unsearchable noise, we used an AI model to read each response and tag it with consistent themes, so “I struggle to run productive meetings” and “meeting management is my biggest challenge” both count toward the same insight.
Build an AI that actually understands the data
We built a two-engine AI system underneath a simple chat interface. The first engine handles numbers. It takes a plain-English question, converts it into a precise database query automatically, and returns counts, averages, and trends across the entire dataset in under a second.
The second engine handles the qualitative side. All 55,000+ open-text responses are indexed semantically, meaning the system finds relevant comments by meaning, not just keywords. When someone asks about “leadership challenges,” it surfaces the actual participant quotes that are most relevant, not just the ones that happen to contain the word “leadership.”
Both engines feed into a final layer that writes a clear, coherent answer combining the numbers and the narrative. Every response also shows the exact database query that was run, so nothing is a black box and results can always be checked.
A simple interface anyone can use
The whole thing is wrapped in a clean web application that looks and feels like a chat tool. No training required, no BI platform to learn. Users type their question and the system decides the best way to present the answer — a written summary, a bar chart, a data table, or all three. Charts can be downloaded with one click and dropped straight into a presentation.
It is deployed securely on Google Cloud, so sensitive participant data never leaves World Class Manager’s own environment. The system also improves over time — when a user marks an answer as helpful, that question-and-query pair becomes a learning example for future responses.
“This is mind-blowing. Honestly, unbelievable. This is way beyond all of what I hoped we could see.”
Wayne Clarke, CEO, World Class Manager · on seeing the system for the first time
Real Questions, Real Answers
The kind of things you can now just ask.
These are not pre-built reports. They are genuine ad-hoc questions the system answers on the fly — the kind of questions that used to take days of manual effort or never got answered at all.
“This is already fascinating. Five people said that emotionally intelligent leadership was something they felt was their strongest area. This can give you really nice insights into what your customers are looking for before they start the course, and then you can start tailoring your messaging around that.”
Wayne Clarke, CEO, World Class Manager
The Outcomes
What actually changed.
The immediate benefit was obvious: years of participant feedback that had been sitting untouched was now actively informing decisions. But the conversation quickly moved beyond that.
Data finally in use
Years of participant feedback that was sitting untouched is now actively informing decisions across the business
Answers in seconds
Questions that used to take days of manual work, or never got answered, now take under 20 seconds
Sharper marketing
Real participant insights now feed directly into marketing messages, LinkedIn content, and thought leadership pieces
ROI is now provable
Pre- and post-course responses can be compared to show measurable participant growth across the programme
No specialist needed
Non-technical team members get their own answers without waiting for a report to be written
A new service opportunity
World Class Manager is now exploring offering this capability to their own clients as a value-add to the programme
“I think this is a massive value add for our product. Our HR and L&D clients don\u2019t understand much about AI or how valuable access to their data is. There\u2019s so many possibilities of how this is utilised, from producing thought papers and reports to generating insights for live streams.”
Wayne Clarke, CEO, World Class Manager
Before and After
Night and day.
Manual, slow, and mostly not happening
- 150+ individual course exports to do manually, one by one
- Hard row limits with no date or dimension filters
- Free-text pasted into ChatGPT for generic, context-free summaries
- No way to compare pre-course expectations to post-course reality
- Marketing messages based on gut feel, not participant data
- Insights that could prove ROI existed but could not be reached
Just type the question
- All 150+ courses consolidated into one queryable source, updated automatically
- Entire dataset queried in under a second with no row limits
- 55,000+ open responses semantically indexed and instantly accessible
- Pre- and post-course data linked for genuine before/after analysis
- Real participant insights feeding marketing messages and LinkedIn content
- Complex, ad-hoc questions answered in under 20 seconds
How It Is Built
The tech, briefly.
For those who want to know what is under the bonnet. Every choice here was made to keep things fast, cost-efficient, and secure. No expensive per-seat licences. No data leaving the client's own cloud environment.
AI question engine
GPT-4o-mini + instructor
Converts plain-English questions into precise, validated database queries with no hallucinated data
Analytics database
DuckDB + Parquet files
Sub-second queries on millions of rows with no server to manage or pay for
Semantic search
Postgres + pgvector + Cohere
Finds relevant free-text responses by meaning, not just keywords
Answer synthesis
RAG + LLM orchestration
Blends numbers and participant quotes into a single, coherent answer
Interface
Next.js + D3.js
Clean conversational UI with auto-generated interactive charts, downloadable as images
Hosting and security
Google Cloud Run + IAP
All data stays within the client’s own Google Cloud environment
The Bigger Picture
If your data is stuck, this applies to you too.
The platform was Moodle in this case, but the problem is not unique to Moodle. Most organisations are sitting on more data than they know what to do with: survey responses, customer feedback, training evaluations, HR pulse checks. And getting a fraction of its value out of it.
As Wayne put it: even a small company like World Class Manager had tons of data that could be looked at in a new way. Which means every company does.
The approach is the same regardless of the source. Get the data out. Clean it up. Make it queryable. Put a conversational interface in front of it that anyone on your team can use without needing to be a data person.
“Look at us as a really small company. Even we have got tons of data that can be looked at in a new way. So that means every company does.”
Wayne Clarke, CEO, World Class Manager
Got data that is not working hard enough?
If this sounds familiar — feedback you cannot access, questions you cannot answer, insights you know are in there somewhere — let's talk.