cohort-revenue-pull
Pulls last N days from Shopify, segments by source/campaign, joins to Klaviyo flow performance. The skill that started it all. Day 2.
Jolly Commerce is a 7-person Shopify CRO agency. The team had Claude open all day, used it to draft emails, clean up briefs, rough out audit decks. Same 5% of the toolset every team starts with. We ran the Wedge sprint internally to break the typewriter habit. Walked in confused. Walked out shipping.
Jolly Commerce runs CRO retainers for DTC brands on Shopify. Four PMs, one designer, one developer, one founder (Kevin). Claude was already on every laptop. Every PM had a Pro plan. Nobody had touched Claude Code. Nobody had wired an MCP. Nobody had written a shared skill.
So every analysis still looped through engineering. "Pull last 30 days of Shopify orders, segment by source, cross with Klaviyo flow performance." That request used to cost the developer half a day. Three times a week.
The brief was simple. Move every recurring analysis off engineering and onto the PM running the account. Five days. No exceptions for "we'll need a custom tool for this." If the tool was missing, we built it during the sprint.
The "wait, what?" moment. Everyone watched the founder run a full Shopify cohort analysis live, in 12 minutes, from the terminal. No SQL, no developer, no agency tool. The team realised they had been using 5% of the toolset.
Claude Code installed on every laptop. MCP set wired into Shopify Admin, GA4, Klaviyo, Slack. First skill written together: cohort-revenue-pull. Shared via the team's hosted skill registry. Every PM ran it on their own client by EOD.
Each PM brought one recurring task they hated. By 6pm we had skills for each task shipped to the registry. weekly-test-recap, klaviyo-flow-audit, pdp-heatmap-summary, cart-abandon-cohort. All version-stamped. All tested on a real client.
Skills that ran more than 3x/week got promoted to internal tools. Built and hosted four. jolly-radar (daily client digest), jolly-cro-brief (auto-formatted Hub doc), jolly-test-results (A/B reporter), jolly-knowledge (team-shared learnings DB).
Friday afternoon: every PM ran a full client review using only their own stack. Engineering didn't get a single ad-hoc request all afternoon. Sprint receipt printed. Team retros what stuck. Next sprint scheduled in 90 days.
Every artifact is in production today. Every skill is version-stamped, used weekly, owned by the PM who wrote it.
Pulls last N days from Shopify, segments by source/campaign, joins to Klaviyo flow performance. The skill that started it all. Day 2.
Daily 7am digest per client account. Yesterday's CVR/RPV deltas, ad spend variance, top-3 alerts. Replaced a 30-min manual ritual every morning.
Reads VWO/AB Tasty exports, computes RPV with proper stats, drops the formatted recap into the client's Hub doc with a one-click "ship to client" button.
Audits abandonment, post-purchase, winback flows. Flags drift from playbook, highlights flows under-sending vs the segment's natural cadence.
Reads Hotjar/Clarity exports + GA4 page-level data, ranks PDP elements by RPV-per-click, surfaces the buried-but-converting blocks. PMs use it on every CRO audit now.
Team-shared learnings DB. Every test result, every client artifact, every "we tried X, here's what happened." Searchable from any PM's terminal. Notion-killer for the agency.
The whole team uses Claude Code now. Not because we forced it. Because the day-2 demo broke their brain. They saw what they had been missing and they refused to go back to the chat tab.
If your team uses AI like a faster typewriter, we walk in, we wire the stack, we walk out. €700 intro for the first 3 clients. €1,200 standard.