I work in B2B sales as a Solutions Consultant at AiCore, a UK apprenticeship provider that trains companies' teams to use AI in practical ways. Here I share personal projects that show how AI can make everyday commercial work clearer, faster and easier to check. The examples are fictional or have had real details removed. I share what worked, what didn't, and what I'd change.

A library of AI workflows for the everyday parts of B2B sales, each with its own one-page recipe card:

• Preparing for a call
• Following up after one
• Building a business case
• Briefing a champion
• Chasing a quiet prospect
• Handling an objection
• Handing over an opportunity
• Reviewing one that did not close
• Keeping a CRM honest
• Finding a new prospect in the first place

The evidence-status matrix shows, job by job, whether it has a method, a skill, a fictional test, real use or an independent test. The last column matters most. Nobody independent has logged using any of these yet, and that's the biggest gap across everything here.

Watch one actually work → shows a real skill turning a fictional call transcript into evidence-labelled output, live. Every line traces back to where it came from.

Rather read than watch? Follow the same example as a written walkthrough, from the fictional transcript to the finished output and its scored review.

There's also a guide to setting up your AI for sales work, whatever tool your company gives you, and a scored comparison of the same task run with no setup in Claude, ChatGPT and Gemini.

Plain-English guides for people who want to use AI at work but don't have a technical background. They start with writing a better brief and checking what comes back, then build up to useful workflows.

A workflow for sales and commercial people thinking about a career change. It checks the evidence at each step.

I use the same standard in two more places: a real worked example with a hard case built in on purpose, and a review that checks whether the tool caught it.

Nineteen free tools. Most take a pattern from the sales repo and rebuild it for a wider audience: keeping facts and assumptions apart in any meeting write-up, building a scoring rubric for any AI output, checking whether the evidence backs up a tracked status, deciding what to send when someone has gone quiet, and more.

Not sure which one fits? Try the interactive picker → or, if you'd rather paste a description into an AI chat, use the router.

Want to build your own skill from scratch instead? Book to Skill turns a book you already own into a Claude skill. Along with the sales workflows above, it's one of the two patterns most of these tools are based on. The newest, Do These Actually Match?, is based instead on the Conflicting evidence label in the sales repo's methodology.

I'm also trying the same idea for commercial teams more widely. These are still early: AI for Commercial Teams, Sales Conversation Gym, Sales Proof Bench, and Sales Value Workshop.

The roadmap covers all of this, not just the sales repo. In short: fewer assumptions, easier to pick up, more real testing, and more work on what already exists before I build anything new.

Not mine, but useful: Wispr Flow Media Pause pauses whatever you're playing while you dictate with Wispr Flow, instead of muting it. It's built for USB audio interfaces that macOS can't mute by itself. A small thing, but it annoys you every time until someone fixes it.

Also worth having if you dictate on macOS: macos-mic-keepwarm keeps the microphone awake. That fixes the 2-5 second delay before push-to-talk starts, so Wispr Flow, SuperWhisper and similar apps start listening at once, instead of losing the first half-second of what you say.

Found on LinkedIn: Premortem, a Claude skill by Ollie Scheers. It assumes a project has already failed and works back to why. Then it sorts the causes into tigers (real risks that need an owner and a plan to reduce them), paper tigers (risks that sound scary but don't deserve much attention) and elephants (things people already suspect but nobody says out loud). It ends with actions that each have an owner, and tripwires, not a longer list of worries.

• Start with a real problem, not a tool looking for a use case
• Keep facts, estimates and assumptions clearly separate
• Keep a person responsible for the final decision
• Don't put sensitive information into tools that aren't approved
• Publish the honest score, not just the polished result

I'm learning as I go and sharing the useful bits here.