Toolkit
Practical frameworks based on the posts of this blog, built to be used, not just read. Score your organisation, run your own numbers, or take a diagram straight into your next presentation. Each one links back to the post that explains the thinking behind it.
How the downloads work: Spreadsheets open in Excel or Google Sheets, with yellow cells for your inputs and a worked example to overwrite. Documents are editable Word files with shaded boxes to fill in. Diagrams are editable PowerPoint slides, each with a blank template version. Every file has a short “when this fits” section so you can tell quickly whether it suits your situation.
Data Foundations & Strategy
The groundwork that decides whether any data or AI investment pays off: how mature your data really is, what you own, and what your stack is built on.
Data Maturity Self-Assessment
Score your organisation on the five-stage Data Maturity Model and see what to fix before your next data or AI investment.
First-Party Data Maturity Self-Assessment
Find where you sit from platform-dependent to first-party led, across owned identity, consent and activation.
Collected, Connected, Activated
Check each customer touchpoint to see where its data stops: collected, connected to the customer record, or used in decisions.
WhatsApp Consent Audit
Audit your WhatsApp consent setup against five checkpoints, and check every place opt-ins are captured.
The Modern Data Stack
See the seven layers of a data stack and map your own tools against them, starting from the foundation.
Identity Resolution Resilience Scorecard
Score yourself across six dimensions of identity resolution resilience, from first-party identifier capture to consent architecture, and see which tier you sit in and what to fix first.
Marketing Analytics & Measurement
Tools for understanding customers and giving channels fair credit, including how to measure discovery now that AI answers take the click.
RFM Segmentation & Persona Model
Score your customers on recency, frequency and monetary value and assign each one a named persona.
Attribution Model Comparison
Run one customer journey through all five attribution models and see how differently each credits your channels.
SEO vs GEO Framework
Compare how search engines rank you with how AI models represent you, then check your readiness for AI-driven discovery.
The GEO Measurement Framework
Combine Search Console with AI visibility tools to find where AI is taking your traffic and whether you appear in the answers.
Building & Running AI Systems
How AI systems are put together, from prompts and retrieval to voice and connections, and how to know when one stops working.
Prompt Engineering Swipe File
Copy tested prompt patterns with ready-made templates, and log your own to build a prompt playbook.
RAG Architecture
See how retrieval-augmented generation works, what you control in it, and plan the four decisions before you build.
The Voice AI Pipeline
See the four layers of a voice AI system, where each tool sits, and where failures start.
MCP Overview
Explain the Model Context Protocol in one picture, and map which systems your AI assistant should connect to.
Evaluation Framework by AI Type
Pick your AI type, check which evaluation metrics you already measure, and build an eval plan from the gaps.
Eval Maturity Ladder
Check how far your AI system is up the four-step ladder, from golden dataset to mandatory regression runs.
The AI Observability Loop
Set up the six-stage loop that keeps a live AI system accountable, with the five metrics to watch.
AI Product Development
Lessons from building real AI products: whether and how to build, how products mature, and how to check the work before you trust it.
Build vs Buy Decision Framework
Score one build-or-buy decision on seven factors and compare three-year costs before you decide.
Build Tool Selection Framework
Compare six ways to build with AI, from no-code automation to a full engineering team, and find the best fit for your situation.
Three-Generation Product Evolution
Map your AI product against the Selector, Advisor and Platform pattern and see what the next generation needs.
The Five-Layer Capability Model
Plan how your product grows in layers, starting from the outputs users need, with a regression checklist.
Knowledge Base Depth Framework
Measure how deep your knowledge base is across seven categories and 28 fields, and find the thin spots.
Independent Cross-Check Verification
Confirm AI outputs with two checks that cannot influence each other, and test that the checks themselves work.
Governance, Risk & Strategy
For leaders deciding what to fund, in what order, and how to keep AI accountable once it is making decisions.
Five Questions Every Board Should Ask About AI
Run a non-technical AI oversight review, with space to record the answers and the actions that follow.
The Four Governance Gaps
Check which of four common governance gaps apply to you, and plan how to close each one.
AI Risk Tiering Register
List your AI systems, tier each by the consequence of an error, and check that governance matches the tier.
Data & AI Adoption Roadmap
Check your progress through the four phases of data and AI adoption, and which phase to finish first.
Leaders vs Laggards
Place your organisation on eight dimensions that separate data and AI leaders from laggards.
Priority Matrix
Plot your initiatives by effort and impact on a live chart, and see which to do first and which to park.
Tools
Two shortlists for anyone building with AI: MCP servers worth connecting, and Claude Code plugins worth installing. Both are starting points for your own checks, not endorsements.
MCP Servers Worth Knowing
A shortlist organised by what each server connects to. Not a ranking, and not exhaustive: the ecosystem has grown past 20,000 servers by some counts, most of them side projects nobody maintains.
Before connecting any of these: a 2026 BlueRock Security audit of public MCP servers found that roughly a third carry SSRF vulnerabilities, four in ten have no authentication at all, and only about one in twelve use OAuth. The protocol standardises how a connection works. It does not vet who built the server on the other end, or how carefully they handle the credentials you give it. Favour vendor-maintained servers over community ones where both exist, check that a server is still actively maintained before connecting it to anything real, and treat everything below as a starting point for your own verification, not an endorsement.
Already covered in MCP: Connecting AI to Your Business Systems: Filesystem, GitHub MCP, Google Drive MCP, Slack MCP, Playwright MCP, Brave Search MCP.
Docs & Knowledge
- Context7: injects live, version-specific documentation into a prompt so a coding agent stops guessing at APIs
- Notion MCP: (official): read and write access to pages, databases and wikis
- Obsidian MCP: no confirmed official server. Obsidian does not publish a first-party MCP server, and the community options (MarkusPfundstein/mcp-obsidian, StevenStavrakis/obsidian-mcp and others) are fragmented with no clear leader. Check current options before choosing one.
Project & Team Management
- Linear MCP: (official, hosted)
- Atlassian MCP: (official): Jira and Confluence, OAuth-backed
Databases
- Postgres MCP Pro: index tuning and health checks, not just raw query access
- Supabase MCP: (official)
Search & Web Data
Business Systems & CRM
- Salesforce MCP: no confirmed official server. Community options exist (for example advancedcommunities/salesforce-mcp-server), but none are maintained by Salesforce as of this writing.
- Stripe MCP: (official): payments, also hosted remotely at mcp.stripe.com
Memory
- Memory MCP: (official, Anthropic): a persistent knowledge graph across conversations, in the src/memory folder of the reference servers repo
Claude Code Plugins & Skills Worth Knowing
A shortlist of what is useful to install, not a full catalogue. For how the trust tiers work and what I tested myself, see Claude Code: The Strategic Case for Building with AI.
Before installing anything: Anthropic reviews two tiers directly. claude-plugins-official is Anthropic-built or vetted, and claude-community is automatically screened third-party work. Most of the wider ecosystem, where some marketplaces run past 300 plugins, has no review at all. Install count reflects popularity, not quality or safety.
Core Workflow
- frontend-design: UI generation with a deliberate aesthetic direction, not generic AI-looking output
- code-review / pr-review-toolkit: reviewing your own work versus reviewing incoming PRs
- commit-commands: git commit, push and PR creation as one workflow
- security-guidance: flags roughly 25 vulnerability patterns in real time while you edit
Building & Extending
- skill-creator: guided creation of a custom skill instead of hand-writing a SKILL.md
- mcp-builder: guided creation of a proper MCP server
- agent-sdk-dev: building custom agents on the Claude Agent SDK
- context7: the same Context7 from the MCP list, also available as a plugin
Testing
- playwright: real browser automation for end-to-end tests
Worth trying, not unconditionally recommending
- knowledge-work-plugins: (Sales, Marketing, Finance, Legal, HR, Brand Voice): directly relevant to non-technical work, but hands-on testing found overlap and quality issues in at least two of the role plugins. Test them against real work before trusting the output.
Install pattern: /plugin install <name>@claude-plugins-official for the default marketplace, or /plugin on its own to browse interactively. Anything outside that marketplace needs /plugin marketplace add <owner>/<repo> first.
Learning Tracker
Not part of the framework themes above. A single, self-contained tool you can start using to avoid the sheer information overload, especially in the AI space.
The T-Shaped Learning Tracker
A working triage tool for deciding what's worth learning, and what's fine to let pass by. Write your stem in one sentence with a built-in test for whether it is specific enough, log what you encounter, and let the status column flag anything in Watch Closely once its revisit date has passed. Seeded with the nine items from the meetup so you can see the triage worked through before you touch it yourself.