Every week a founder asks us some version of the same question: "Should I just let the AI run my marketing?" The honest answer is that the best Shopify brands we work with don't choose. They run a hybrid, and they're deliberate about which jobs go to the machine and which stay with a person. Here is how we draw that line, and the guardrails that keep the automated half from quietly spending your money badly.
Two kinds of marketing work
Strip marketing back and there are two kinds of work in it. The first is judgement: who is this brand for, what do we promise, what do we say, what do we test next, what does this result mean. The second is execution at scale: bidding on ten thousand auctions a day, splitting a list into segments, resizing creative for nine placements, choosing which product to show which visitor.
AI is extraordinary at the second kind and unreliable at the first. Most of the disappointment we hear about ("we turned on the AI and sales didn't move") comes from handing it judgement work, or from handing it execution work with no human reading the results.
What to hand over to AI
Bidding and budget allocation
Meta's Advantage+ campaigns and Google's Performance Max are, at their core, automated bidding systems. They are better than any human at deciding, in real time, how much to pay for a particular impression. Let them. Your job moves up a level: feeding them clean conversion data, a sensible budget, and creative worth showing.
Segmentation and send-time
Predicting who is likely to buy again, who is about to churn, and when a given customer opens email is pattern-matching over thousands of rows. Email and WhatsApp platforms do this well. Use the predicted segments; write the message yourself.
First drafts and variations
Ten subject-line options, six ad-copy angles, a product description outline, alt text for two hundred images: this is where a language model saves real hours. Treat everything it produces as a draft from a bright intern who has never met your customer.
Product feed hygiene and on-site personalisation
Categorising products, filling missing attributes, recommending "you might also like". Machines do this tirelessly and improve with data. Check the recommendations occasionally so the model doesn't learn something silly from a stock-out.
What to keep human
Strategy and the offer
Which customers you want, what problem you solve, why someone should pay your price rather than the cheaper one: no model has that context, and a wrong answer here makes everything downstream efficient at the wrong thing.
Brand voice and creative direction
AI copy converges on the average. Your brand needs to sound like one particular person, not the mean of the internet. Use the drafts; keep the pen.
Reading the results and deciding
Dashboards report; they don't interpret. "ROAS fell 20%" could mean the creative is tired, a competitor launched a sale, or the tracking broke on Tuesday. Someone has to ask which, and that someone should understand your business.
Customer conversations
Chatbots handle "where is my order" fine. The angry customer, the wholesale enquiry, the influencer who wants to collaborate: those are human moments, and how you handle them is your brand.
Guardrails for the automated half
- Keep a control. When you turn on an automated campaign, leave a small manual one running for a few weeks. If the automation can't beat it, you've learnt something valuable.
- Feed it real conversions. Automated bidding optimises to whatever event you give it. If your pixel fires on "add to cart", you will get very cheap add-to-carts and few orders. Check your events before you check your ads.
- Set spend limits at the account level, not only inside the campaign. Automation is confident, and confidence spends.
- Read every AI draft out loud before it goes live. If it doesn't sound like you, it isn't yours yet.
- Review the segments monthly. Predictive models drift as your catalogue and seasons change.
What this looks like at your stage
A brand doing its first few thousand in monthly sales doesn't have enough data for most machine learning to be clever; keep things simple, run automated bidding with tight budgets, and spend your time on the offer and the product page. As volume grows, the automated half earns more responsibility (more of the budget, more of the segmentation, more of the creative testing) and the human half spends more time reading and less time doing.
That's the hybrid: machines for the ten thousand small decisions, people for the ten big ones. Neither works well alone.
