Train a Custom GPT on your brand guidelines doc, paste any draft, and get a line-by-line compliance score with violations flagged before anything ships. The 30-minute build.
Stand up an Airtable base fed by GA4, ad spend, email and CRM, a Claude prompt that answers natural-language questions about it, and a Slack slash command so any teammate can ask 'what's our blended CAC by channel this month' and get a chart back. Full hour-by-hour build, with the prompt, the schema, and the four gotchas that will eat your weekend if you skip them.
A battle-tested 5-email cart abandonment sequence with exact send times, subject lines, body copy, and the AI prompts that generated them — plus what to fix by hand.
Wire a Slack slash command to an n8n workflow that fans out to Perplexity for cited research, hands the result to Claude for a 3-paragraph rewrite, and posts it back into the same channel — full JSON, prompts, and the dispatch_failed trap included.
A 6-step workflow that uses Apify to scrape 50 competitor product pages into a Google Sheet, has Claude flag any price change above 5%, and posts a Slack alert with the diff. 90-minute setup, ~$0.10 per daily run.
A copy-paste AI prompt that turns 90 days of Klaviyo events into 25 high-conversion micro-segments — grouped by lifecycle, behavior, predictive signals, channel, and strategic niche. With the actual segment definitions you can paste into Klaviyo today.
One morning, five dashboards requested, zero patience. I wrote 10 base SQL queries against the GA4 BigQuery export, fed them to Claude with a Looker Studio schema, and shipped 10 working dashboards before close of day. Here are the SQL, the prompt, and the traps I hit.
Build one 8-slide carousel, generate four different cover openers with Claude, A/B test them for a month, and let the data name the winner — a step-by-step hook-test protocol I now run on every carousel.
A battle-tested Customer.io sequence for trial users stuck at day 3-5 with no activation event. Three personas, one hard exit, and the one instrumentation decision that makes the whole thing work.
Stop A/B testing two subject lines and hoping. Pre-write 15 falsifiable hypotheses (length, emoji, personalization, urgency, numbers, case, brackets, sender, preheader...) in a 15-row spreadsheet, design 2-4 cells per hypothesis, then let the data pick the winners — one row per send, not a multivariate stew. The 6 starter rows, the Klaviyo test setup, and why one row per send is the rule.
Generate 100 subject lines with ChatGPT, filter them down to 5 with a 4-criterion rubric, then A/B/n test the finalists. The exact prompt, the 5 that made it through, and the test result — including the variant I would never have written on my own that beat my favorite by 19%.
