Beaphar UK - practical AI tools inside a marketing team

Most marketing teams "use AI" by pasting things into ChatGPT and hoping for the best. That's fine for brainstorming, but it doesn't scale and it doesn't enforce the things that matter - brand voice, compliance, consistency.

At Beaphar I had three recurring problems that were eating time: writing on-brand social content across platforms, monitoring partner stock levels after a key retail partner went bust, and assembling monthly digital reports from multiple data sources. Each one was repetitive, predictable, and ripe for automation.

So I built tools for them. Not pitch-deck ideas. Working tools the team can actually use.

The problem: writing social content for Beaphar means following specific brand voice rules - "our pets" shared-ownership language, British English spelling, empathetic tone, no overclaiming, max 100 words per post. Getting that right every time while producing content for Facebook and Instagram is slow and error-prone.

The solution: a single-file React application (380 lines of JSX) that runs as a Claude artifact. You paste in rough draft copy, and it generates on-brand, platform-ready content through a three-step workflow:

• Step 1 - Input: Paste your draft. The tool auto-extracts any URL and sets up brand context.
• Step 2 - Angles: Claude generates 3-4 genuinely different creative angles on the same topic - different emotional hooks and framing, not just rewording. You pick the one that fits.
• Step 3 - Output: Instantly builds platform-specific versions. Facebook gets the copy with URL appended. Instagram gets the copy with URLs stripped and a "link in bio" CTA. Simultaneously generates 3 first-comment options.

The system prompt encodes roughly 650 words of Beaphar copywriting rules covering tone, British English mandates, product sensitivity, copy constraints, and first-comment guidelines. Every piece of output runs through a sanitisation pipeline that strips markdown, normalises dashes, and cleans formatting.

Visual design matches Beaphar's exact brand colours and typography. The whole thing is self-contained - no build step, no external dependencies.

The problem: a key click-to-buy retail partner went out of business in 2024. Suddenly we had buy-now links across the site pointing to a dead storefront, and no automated way to know which products were still available through remaining partners.

The solution: an automation built on Zapier that monitors partner stock levels and validates buy-now links on an ongoing basis. It checks availability, flags dead links, and alerts the team when stock status changes - replacing what would have been a manual audit that nobody had time to do regularly.

Not glamorous, but it solved a real business problem that was actively costing us conversions.

The problem: monthly digital reports meant pulling data from multiple sources, formatting it, adding context, and assembling it into a presentable document. It was hours of work that followed the same pattern every month.

The solution: a Claude Code skill that automates the entire workflow - pulls the data, structures the report, and generates the output. What used to take hours now takes minutes, and the format is consistent every time.

The skill is reusable and adapts to each reporting period. I built it because I'd rather spend my time on analysis and decisions than on copying numbers between spreadsheets.

Three principles I followed across all three builds:

• Solve a real problem first. Each tool started with a specific pain point, not "let's use AI for something." The technology choice followed the problem, not the other way around.
• Make it actually usable. A tool nobody uses is worse than no tool at all. The Social Post Workshop has a three-step flow because that's how the team already thinks about content. The reporting skill runs with one command. Low friction or it doesn't get adopted.
• Brand and compliance aren't optional. In a company like Beaphar, every piece of content has rules. The AI tools enforce those rules by design - they're baked into the system prompts and output processing, not left to the user to remember.