This is the first post in a mini-series on my experience completing the Imperial College Business School Professional Certificate in AI for Business Transformation (delivered via Emeritus). The series covers this programme reflection, the Cashflow Copilot concept that emerged from the capstone work, the RAG evaluation framework I built as technical preparation, and finally cloning that framework to build an FCA complaints handling bot - the Module 1 use case I set aside commercially in favour of Cashflow Copilot - which is the cleaner demonstration of the eval framework in practice.

Why I enrolled

I've spent 20 years in enterprise technology across operations, consulting, and vendor roles. GenAI toolchains are deeply embedded in how I work. The reason for taking this course was to look at it from an alternative angle: the academic frameworks, the structured approach to innovation and competitive advantage, the business case for change and the ethics vocabulary. Not to learn about organisational change; to see what a formal AI strategy framework adds to the experience I already have.

The programme is explicitly business-focused and comprised of six modules, each with a mini-capstone submission, delivered over several weeks along with weekly cohort calls.

What each module covered and what I got from it

Module 1: AI fundamentals. The content wasn't new given my existing experience with AI tools, but the scoping exercise was useful. Identifying and framing three UK banking use cases (developer productivity, FCA complaints handling, policy knowledge management) forced a discipline around problem definition that set up the rest of the programme.

Module 2: GenAI tools and prompt engineering. This module introduced me to some new prompting techniques like Tree of Thought and Graph of Thought, which were frameworks I hadn't used formally before. Applying ToT to a real go-to-market analysis for one of my own projects rather than a course exercise helped bring it to life; I later added it into my claude-toolkit as a reusable skill. The capstone here also reinforced a view I'd been forming through hands-on experience: ML classification is the right tool for some tasks, and reaching for an LLM when an alternative approach fits better is a mistake.

Module 3: Automating workflows. The automation strategy content was familiar territory from experience of data centre automation use cases, but reframing it in the context of AI and business was obviously fresh. Having to articulate the benefits beyond efficiency (data asset creation, regulatory posture, employee value shift) forced me to dig deeper than I would have otherwise. The structured articulation was the value in this module and not necessarily the concepts themselves.

Module 4: Innovation and competitive advantage. The VRIO framework was a new element to me. I firstly applied it to Wiz capabilities (the security graph as a sustainable advantage through path dependency and data network effects; Wiz Code as competitive parity against pure-play AppSec vendors) and later to the Cashflow Copilot concept (where the bank's proprietary transaction data is the defensible asset, not the AI itself). It's now part of my toolkit as a reusable skill alongside the design thinking ideation framework from the same module.

This was also where I pivoted from the FCA complaints handling use case to Cashflow Copilot. The complaints use case was solid but it was an internal efficiency play and I wanted something more ambitious. The design thinking exercise helped to shape the concept into a customer-facing financial intelligence tool with three AI layers (ML forecasting, GenAI alerts, RAG scenario simulation). That concept carried through Modules 5 and 6 and is the subject of the next post.

Module 5: Strategic integration and organisational impact. Change management, stakeholder analysis, and integration strategy weren't new to me conceptually. But applying it rigorously to a specific AI use case forced a different depth: writing out who is affected and how, team by team, and building the integration strategy around protecting the data asset. That was something I don't have to do anymore in presales. Well delivered module and a great refresher on frameworks I discuss more informally with customers.

Module 6: Ethics and risk management. The standout module for genuine learning. The course introduced "ethics as architecture": the idea that ethical safeguards are structural design decisions baked into the product, not a compliance checklist bolted on afterwards. Before this module I'd probably have leaned more toward compliance and governance framing. The shift to thinking about opt-in architecture, system firewalls between intelligence and advice, plain-language explanations, and human escalation pathways as design constraints that make the product better (not just safer) was a useful change in perspective. I strongly believe in ethics in business; this module gave me a more rigorous framework for acting on that.

The cohort

The weekly cohort calls added something the content alone doesn't. Hearing other people's perspectives, their industries, their challenges with AI adoption; that context rounds out the learning in a way that modules and readings can't replicate and was the most enjoyable element.

Using AI to do the course about AI

I used Claude throughout all six modules with a discuss-first, draft-second workflow. The collaboration itself was largely applying patterns I already knew from Anthropic's 4D framework, but the course created situations where I found new ways to co-create and innovate that I wouldn't have discovered without it. Running design thinking exercises collaboratively, using graph of thought to surface non-obvious ethical connections, challenging Claude's cost estimates based on my own experience; these were patterns that emerged from the coursework rather than things I went in knowing how to do.

Was it worth it?

Yes. Not because it filled gaps in my knowledge; most of the operational and change management content was familiar ground after two decades in the industry. It was worth it because it gave me new frameworks (VRIO, ethics as architecture, new prompting techniques), a structured capstone process that produced a concept I wouldn't have developed otherwise, and cohort perspectives I wouldn't have encountered in my day-to-day. The programme teaches you how to think about AI strategically. For someone already comfortable with the technical and operational side, that different viewpoint is exactly the point.

What's next in this series

The next post covers the Cashflow Copilot concept in detail: the three-layer architecture, the bank data advantage assessed through VRIO, the integration strategy, and ethics as architectural design. After that, the RAG evaluation framework I built to systematically measure chunking strategy impact before building anything on top of it. And finally, cloning that framework to build an example bot that a customer services representative could use.