Shopify marketing automation: what to automate, what to approve, what to keep

Shopify marketing automation

For most of the last decade, "marketing automation" on Shopify meant flows. A cart is abandoned, an email goes out. Someone signs up, a welcome series starts. The trigger was automated; the words were written once, by a person, and reused for years.

That's no longer the whole picture. AI can now draft the campaign, the blog post, the ad and the product-page fix — not just send them. Which means the question has changed from what can I automate? to what should run without me looking at it?

The answer that holds up is three layers. Some work should run entirely on its own. Some should be drafted automatically but published only after a yes. And some should never leave your hands.

Table of Contents

Layer 1 — automate completely

These are trigger-and-response jobs. The content is written once, reviewed once, and then the same message is right every time the trigger fires. Nobody should be approving these individually.

The rule for layer 1: if the message would be identical no matter who reviewed it, reviewing it every time is waste.

Layer 2 — automate the draft, approve the publish

This is the layer that didn't exist before AI, and it's where most of the time savings and most of the risk now sit.

Campaign emails, blog posts, ad copy, product-page and FAQ fixes — all of these can now be drafted from your own store data. What changed is that the draft is new every time, so it can be new and wrong every time. A drafted campaign doesn't know your shipping cutoff moved, that you can't make a health claim in one market, or that the discount ended yesterday.

So layer 2 runs on a rule: drafting is automatic, publishing waits for a yes. We wrote up the six checks that review should cover; in practice it's one to three minutes per piece once it's a habit, because you're deciding, not editing.

What belongs here:

Layer 3 — keep it

Some decisions shouldn't be drafted at all, because the draft would just be a guess dressed up as a recommendation:

A worked example: the automation on our own site

It's easier to see the layers in something real, so here is the lead-capture automation that runs on withbranva.com right now. It's not a Shopify store, but the pattern is identical to a store's signup and post-purchase automation.

When someone fills in the form on our homepage:

  1. The form validates name, email and phone before anything is sent, so bad data never enters the system.
  2. The team gets an email with everything the person wrote, and the subject line names them and their company — so the first human touch is fast and specific.
  3. A row is written to a Google Sheet. That's the system of record, and the hand-off point for anything else downstream, such as an SMS step.
  4. A three-email sequence is scheduled at the moment of submission — the first sends immediately, the next two are queued for day 3 and day 7. There's no cron job and no external scheduler; the email service holds the scheduled sends.
  5. The sequence is chosen by the page the person came from. A signup from a Klaviyo post gets Klaviyo material; a signup from the homepage form gets a sequence about how we'd work together. Callback-style forms skip the sequence entirely, because their promise is a person reaching out, not a drip.

Mapped onto the layers: steps 1–5 are layer 1 — they run on their own, every time. The email copy in that sequence was a layer-2 job: drafted, then approved before it went live. And deciding what to promise in that first email ("a person reads these, not a bot") was layer 3.

Three lessons from building it:

Running layers 1 and 2 as one loop

The hard part of marketing automation isn't any single flow. It's that the layers drift apart: flows are set up once and never revisited, drafts pile up in a doc nobody approves, and nothing connects what shipped to what it earned.

That's the problem Tilly is built for. It runs the marketing as one loop — audit, build, approve, publish, measure, next:

That's layers 1 and 2 in one place, with the approval gate built in rather than bolted on. The free plan runs the full brand audit and drafts without publishing anything, which is the cheapest way to see whether you'd approve what it writes. Paid plans that publish start at $49 a month. It's on the Shopify App Store.

For the bigger picture of where automation fits across a store — operations as well as marketing — see the role of automation in ecommerce. For the measurement side, how to measure AI-written marketing.

Frequently asked questions

Is Shopify Flow enough for marketing automation?

For layer 1 housekeeping — tags, alerts, inventory rules — Shopify Flow is excellent and free. It doesn't write or send marketing, though. Email automation runs in your email platform (usually Klaviyo), and drafting content is a separate job again.

What's the difference between Klaviyo flows and AI marketing automation?

A Klaviyo flow automates when a message sends; the message itself is fixed. AI marketing automation drafts what to send — new campaigns, posts, ads — which is why it needs an approval step that flows never did.

Will automation make my marketing sound generic?

Only if the inputs are generic. Drafts sound like you when they're built from your own products, reviews and past customer language, and when someone who knows the brand approves them before they publish. Generic output is almost always a briefing problem, not an automation problem.

What should I automate first?

The four Klaviyo flows that carry email revenue: abandoned checkout, welcome, browse abandonment and post-purchase. They're layer 1, they earn from day one, and they don't need ongoing approval.

What does Tilly cost?

There's a free plan that audits your store and drafts without publishing. Store is $49 a month, Growth $149 and Scale $399; the paid plans come with monthly credits that never expire. Nothing publishes until you approve it.