Automating the repetitive 80%
Ask Meera where her week actually goes and the glamorous answer — making things, telling stories — is a fraction. The honest ledger: answering "is this lamp in stock?" for the fortieth time, assembling the same weekly numbers, reformatting the same launch announcement for five surfaces, chasing the same follow-ups. Marketing for one is mostly operations, and operations is where automation-plus-AI pays best precisely because the work is repetitive: patterns a machine can learn, volume a machine doesn't resent. This lesson is how to hand it over without handing over your name.
The golden rule: automate toward a checkpoint
One principle governs everything here, continuing lesson 9's design: AI automation drafts; humans approve anything that leaves the building. The failure stories that go viral — the chatbot inventing a refund policy, the auto-reply gone surreal — share one root: generated text reached a customer with no human between. The fix isn't better AI; it's architecture. Every automation below routes output into a review point — a draft folder, an approval column, a "suggested reply" — never straight to send. Hold that line and automation failures cost you a chuckle in private instead of a screenshot in public.
The four automations that earn their keep
DM and comment triage. Meera's inbox sorts into perhaps six recurring questions (price, stock, shipping, customisation, care, wholesale) plus a long tail of genuinely individual messages. The automation: AI classifies incoming messages and suggests the matching answer from her own written replies — she taps approve or edits, and the fortieth stock question takes four seconds instead of forty. The individual messages — a bride planning a trousseau, a complaint, an artisan inquiry — get flagged straight to her, full attention. That's the shape to copy: automation absorbs the repetitive; humanity concentrates where it's felt. (Meta's business inbox has grown steadily better built-in versions of exactly this as of mid-2026 — check what your existing tools already do before building anything.)
The weekly numbers draft. A bridge (lesson 9) pulls the week's metrics into a sheet; the assistant turns them into five readable lines — what rose, what fell, what's odd — waiting in her notes every Monday. The reading habit this feeds becomes lesson 11's whole subject.
Launch choreography. A launch means the same information formatted eight ways (post, Reel, stories, newsletter, WhatsApp broadcast, website banner, ad copy, pinned comment). The core-story chain from lesson 6, run as a standing checklist with the model doing every derivation into a review doc: an afternoon's scramble becomes a twenty-minute review.
The follow-up net. Inquiries that went quiet, wholesale leads unanswered a week, the restock-notify list when stock lands: AI drafts each follow-up from the thread's context into a queue; Meera sends with edits. Nothing falls through, nobody gets a robot.
Maintenance is part of the deal
Automations rot quietly — the platform changes a field, the classifier meets a message type it's never seen, the "weekly" bridge has silently failed for three weeks. Two habits keep them honest: a monthly ten-minute audit (is each automation still running, still right, still worth it?), and watching the edits — when you find yourself rewriting the same suggested reply every time, that's the voice-doc loop from lesson 2 telling you a rule is missing; add it at the source instead of editing forever. And retire without sentiment: an automation you keep overriding is a process telling you it was automated wrong.
Do this today: write your six most-repeated DM answers into a doc — your reply library, the raw material for triage automation and useful even if you never automate (paste-ready answers beat retyping). Then pick the one automation from the four whose absence costs you most, and set up its simplest version this week.