Every month, our promo campaign is a huge Canvas: close to 200 steps, with dozens of emails and pushes, more than 80 delays and a dozen splits, across two audiences (active and inactive users). Building it by hand took several days, and QA-ing it took even longer. At that scale, one wrong connection can send the wrong reminder to the wrong audience.
Attempt #1: let Operator build it from scratch. We gave BrazeAI Operator a full Canvas brief and asked it to build the whole thing. It didn't go well: it invented step names, left delays on default settings and rewrote our email copy. Lesson learned: never build from zero.
Attempt #2: duplicate, then edit. We simplified the Canvas into a single flow, and now we just duplicate last month's version and change what's different. Then I turned the method into a Claude Code skill:
- I share a screenshot of the promo calendar, and Claude turns it into a send-by-send table for me to validate.
- Claude reads last month's Canvas through the Braze MCP and lists every difference: new dates, extra reminders, removed sends, the campaign tag.
- It writes about 7 short, precise prompts for Operator, one type of change at a time.
- After each save, Claude checks the Canvas against the API before moving to the next prompt.
The first full test run also caught two issues a manual review had missed: a one-day gap in our new sign-up entry filters, and split filters still pointing to the previous month's promo tag.
The result: what used to take several days of setup and QA now takes about 2 hours, and every change is verified. Now that Operator Connect is live, Claude can also send audit prompts to Operator directly through the MCP.
My biggest takeaway: AI doesn't replace CRM rigor. It's what makes that rigor sustainable on a 200-step Canvas.
