Turning a static workbook into an AI-enhanced course and coaching program

The client

A leadership and culture development consultancy. Its core intellectual property is a proprietary leadership framework, delivered through keynotes, workshops, one-on-one coaching, an online video course and a substantial printed workbook.

The business is deliberately relational. Its value sits in interpretation, facilitation and trust rather than information. That strength was also the constraint: the depth lived with the founders, and the digital assets could not carry it on their own.

The challenge

The online program worked as a set of videos paired with a workbook the participant filled in by hand. It was good content, well structured and grounded in real practice. It was also completely static.

A leader could work through every module and finish with a stack of handwritten notes and no plan. Nothing they wrote was captured anywhere, and two leaders with very different teams received exactly the same experience. Turning reflection into a development plan still landed on the founders, one client at a time, so growing the program meant growing the founders' calendars.

The delivery itself was fragmented too, with videos, workbooks and course material spread across separate platforms and file stores.

The approach

The existing workbook was treated as the source of truth, not a starting point to improve on. The framework already worked; the job was to make it responsive. Every section was mapped back to its workbook page before anything was built, and each post opens with that page reference so participants keep using the physical workbook alongside the screen.

Two boundaries shaped the AI from day one:

• Professional limits. Where content touches governance, the AI prompts the leader to confirm the basics are in place and refers them to their own HR provider, accountant or lawyer. It does not advise, because it should not.
• Other people's intellectual property. The program draws on a licensed behavioural profile. Rather than reverse engineer the vendor's scoring, the AI gives a general overview and points participants back to their full report.

The solution

The program was rebuilt on one learning platform as a series of modular posts, each following the same shape: workbook page reference, the lesson video, structured questions, space for anything else the leader is thinking, and an AI-generated summary with action points.

What makes the output specific rather than generic is the AI design underneath:

• A thinking partner on every topic. Each topic has its own AI agent with a deliberately narrow knowledge scope, so it stays anchored to the section the leader is working on rather than wandering into general management advice.
• The client's framework as the lens. The workbook and framework are loaded in as the knowledge the AI works from, and every component has its own prompts, so responses use the same language and sequence the founders use in the room.
• Questions before advice. No advice is given while information is still being gathered. The AI asks first, then reflects the leader's own situation back through the framework.
• Personalisation. Leaders can add their behavioural profile, so the AI can take their working style into account.
• Progression that holds. Sections unlock in order, so leaders complete each step before the next opens.

Everything the leader contributes is carried forward into a development plan, which is designed to become the starting point for one-on-one coaching. Rather than spending the first part of a session establishing where someone is up to, the coach arrives already holding what the leader has said and what they have committed to.

Around the course, we also:

• Moved all 42 lesson videos, about 15GB, to secure embed-only hosting because the platform could not take files that large (up to 900MB)
• Made workbooks view-only so the client's material stays protected
• Added a custom domain, a new sign-up flow and an anonymous feedback form kept separate from the AI
• Brought course delivery together on one platform
• Trained the client's team to use Claude and wrote 13 how-to guides and a 26-page capability guide so they can run and improve the program themselves

Before launch, internal testers worked through the course, and the issues they raised were logged and tracked through to fixes.

The results

The program has moved from fixed content to a system that responds to the person completing it, holds what they say and builds towards a plan. It is launching to clients in October 2026, with 42 lessons in one protected home and the video hosting change projected to save around $2,700 a year.

Measurable participant and coaching outcomes will be added here after the first cohort.