I work in B2B sales as a Solutions Consultant at AiCore, a UK apprenticeship provider helping companies upskill their teams in practical AI use. This profile is where I share personal projects showing how AI can make everyday commercial work clearer, faster and easier to review. The examples are fictional or sanitised, and I share what worked, what did not, and what I would change.

A working 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. That last column is the honest headline: no independent use has been logged for any of these yet, which is the single biggest gap across everything here.

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

Prefer to read it rather than watch it? Follow the same example as a written walkthrough instead, from the fictional source transcript to the finished output and its honest scored review.

There's also a guide to setting up your own AI properly for sales work, whatever tool your company gives you access to, and a scored comparison of the same task run cold in Claude, ChatGPT and Gemini.

Plain-English guides for people who want to use AI at work but do not have a technical background. It starts with a better brief and checking what comes back, then builds towards useful workflows.

The same standard used above, a real worked example with a deliberate hard case built in and an honest review checking whether it was actually caught, is also applied in two further places.

Nineteen free tools, most a pattern from the sales repo rebuilt for a wider audience: keeping facts and assumptions separate in any meeting write-up, building a scoring rubric for any AI output, checking whether a tracked status is actually supported by evidence, deciding what to send when someone has gone quiet, and more.

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

Building your own skill from scratch rather than picking one of these? Book to Skill turns a book you already own into a Claude skill, and is one of the two patterns (alongside the sales workflows above) that most of these tools were generalised from. The newest, Do These Actually Match?, instead generalises the Conflicting evidence label named directly in the sales repo's own methodology.

Also experimenting with the same idea for commercial teams more broadly, still early: AI for Commercial Teams, Sales Conversation Gym, Sales Proof Bench, and Sales Value Workshop.

See the roadmap for the full picture across all of this, not just the sales repo. In short, the next stage is fewer assumptions, easier adoption, more real testing, and deeper work on what already exists, ahead of building anything new.

Not mine, but genuinely useful: Wispr Flow Media Pause pauses whatever you're playing instead of muting it while you dictate with Wispr Flow, built for USB audio interfaces macOS can't mute on its own. Small, but the kind of thing that's annoying every single time until someone fixes it.

Also worth it if you dictate on macOS: macos-mic-keepwarm fixes the 2-5 second push-to-talk activation delay by keeping the microphone hardware awake, so Wispr Flow, SuperWhisper and similar apps start listening instantly instead of eating the first half-second of what you say.

Found via LinkedIn: Premortem, a Claude skill by Ollie Scheers. It assumes a project has already failed and works backwards to why, then sorts the causes into tigers (real risks that need an owner and a mitigation), paper tigers (risks that sound scary but do not deserve much attention) and elephants (the things people already suspect but nobody is saying out loud). Ends in owned actions and tripwires, not a longer list of things to worry about.

• 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
• Do not put sensitive information into tools that are not approved
• Publish the honest score, not just the polished result

I am learning as I go and sharing the useful bits here.