There's a lot of valuable expertise sitting within consulting firms, often captured in frameworks, methodologies and other IP. But what if you could use AI to turn some of that knowledge into interactive tools that clients and prospects can use for themselves?

In this episode of Professional Services Tech Talks, Adam Maze and George White take us through how we've used Claude Code to build our new online Value Builder assessment. The tool draws on Garwood's existing IP to assess a professional services firm's maturity across eight business domains and generate a personalised report with observations and recommendations.

George shares how he approached the build, from getting started with Claude Code to working out which systems needed to connect behind the scenes. They also discuss what it took to move beyond an initial working version to something ready for prospective clients to use.

They get into:

• How to turn your firm's existing expertise and IP into useful AI-powered tools
• Where to start with Claude Code or ChatGPT Codex, and how AI can help you plan the build
• Why a clear idea of what you're creating and well-documented IP make a difference
• What we learned from testing and improving the AI-generated reports
• How we've built in checks to keep recommendations consistent with Garwood's approved IP
• The practicalities of hosting, integrations, data security, GDPR and insurance
• Choosing different AI models for different tasks and keeping running costs manageable

There's plenty of practical advice here for anyone considering building something similar, including some of the things George discovered through trial and error. It's a useful reminder that while AI can now handle much of the technical work, building something you're confident sharing with clients still takes thought and testing. And it helps to have a clear idea of what you're trying to achieve.

Next time

Adam and George look at how we're using AI to automate some of the manual work involved in keeping our CRM up to date.

Meeting transcription tools are hardly new. But automatically producing a set of notes after a call is only part of what they can do.

In this episode of Professional Services Tech Talks, Adam Maze and George White talk through how Garwood chose its meeting transcription platform, Fellow, and what they wanted from it beyond transcription accuracy.

They also look at what becomes possible once meeting content is connected to the rest of your technology. George shows how he can use a conversation captured in Fellow as context in Claude, while Adam explains how he sends actions from meetings straight into Microsoft To Do.

They get into:

• What we looked for when choosing a meeting transcription platform
• Why security, GDPR and data ownership were important parts of the decision
• Bot versus botless recording and where each approach works
• Why we chose Fellow from the platforms we assessed
• How meeting transcripts can become context for Claude
• Turning feedback from a meeting into a plan for the work that follows
• Moving meeting actions into Microsoft To Do and the daily digest George is building

There’s also a brief detour into George’s autonomous stock-picking experiment, followed by Adam putting him on the spot with a problem from Daniel Kahneman’s Thinking, Fast and Slow. George handles the pressure rather well.

Next time
Adam and George get under the hood of Garwood’s Value Builder Diagnostic, looking at how we’ve turned our own IP into an interactive lead magnet and how other professional services firms could approach something similar.

Creating a client deck often starts the same way: find an old one, strip out the content and spend far too much time getting fonts, colours, logos and layouts where they’re supposed to be.

We’ve been experimenting with a different approach.

In this episode, Adam and George go behind the scenes of the AI skills we’ve built at Garwood. George opens up our branding skill in Claude, shows what sits behind it and demonstrates how we can turn unformatted content into a branded PowerPoint or Word document.

They get into:

• What an AI skill is and where it can be useful
• How we’ve combined instructions, brand assets, templates and references
• How we decide when the skill should automatically kick in
• What the first output actually looks like
• How feedback helps improve the skill over time
• Why replicating tone of voice has proved much harder than visual branding
• How we’re using Claude and ChatGPT for different parts of the job

It’s not a case of pressing a button and sending the result straight to a client. The output still needs human review and refinement. But as George puts it, if the first attempt gets you 60% of the way there, that’s already time saved.

And for anyone wondering whether AI has finally cracked tone of voice: not quite. That one is still very much a work in progress.

Next time

Adam and George look at meeting transcription tools: how we chose ours and what becomes possible when meeting content connects into the rest of your technology stack.