We use AI every day. We just don't let it near the bits that matter.
Every agency is saying they use AI now, which tells you approximately nothing. So here's the specific version: what it actually does in our work, what it's saved people, and the places we've decided it doesn't belong. Including the ones that would make us more money if we let them.
It's a very good junior. It is not a strategist.
AI is extraordinary at the work that is laborious but not hard — the first draft, the hundredth product description, the boring half of a migration. It's unreliable at the work that's actually hard: knowing which thing to build, why your customers hesitate at the same step every time, what your brand should sound like.
It does the typing, not the thinking.
The judgement is the thing you're paying for, and it stays human. If a decision needed taste or an opinion about your business, a person made it.
Everything gets reviewed.
Nothing generated goes live without someone here reading it properly. That sounds obvious. Look at the state of the internet and you'll see it isn't.
You get the time back, not a bigger invoice.
Where AI makes something quicker, it comes off the estimate. We price the work, not the hours we'd have spent without it.
Six things it does on a real store.
Not a list of model names. These are the jobs where it has genuinely changed what a project costs or how long it takes.
- 01
Product copy at catalogue scale
Two thousand SKUs with two lines of manufacturer copy each is a job nobody will ever get funded, so it just stays undone. We build a pipeline that drafts to your tone of voice, pulls real specs from your data rather than inventing them, and routes everything to a human queue for edit and approval before a word publishes.
A catalogue that's actually readable, and indexable, in weeks rather than never.
- 02
Art direction and image production
Not fake products — real ones. Generating and iterating on concepts before a shoot so you turn up knowing what you want, extending backgrounds to fit formats the photographer didn't shoot for, and producing the endless lifestyle crops that campaigns eat.
Fewer reshoots, and a shot list agreed before anyone hires a studio.
- 03
Research that would take a month
Thousands of reviews, support tickets and session recordings contain the answer to why people aren't buying. Reading all of them is the part nobody does. We use AI to cluster and theme the lot, then go and verify the top few by hand, because the clustering is often confidently wrong about the interesting ones.
A conversion roadmap built on what customers said, not what the room assumed.
- 04
Search and merchandising that understands intent
Shopify's native search matches words. Someone typing "something warm for a toddler" isn't using your product titles. Semantic search and AI-assisted recommendations close that gap, and they are genuinely one of the highest-return things you can add to a large catalogue.
Fewer no-results pages, and collections that reorder around what's actually selling.
- 05
Support that resolves rather than deflects
Grounded strictly in your real policies, product data and order status — never free-associating. It answers the where-is-my-order and does-this-fit questions instantly and hands over to a person the moment it's out of its depth, which we make it do early rather than late.
Your team spending their day on the conversations that need a human.
- 06
The unglamorous engineering half
Redirect maps for a migration, test coverage, data cleanup, translating a legacy theme's logic so we can rebuild it properly. This is where it saves us the most time and where the risk is lowest, because the output is either correct or it fails loudly.
Migrations and builds that cost less because the tedious part got cheaper.
The things we won't do, including the profitable ones.
Any agency can list the tools. This is the part worth reading.
We don't publish copy nobody has read.
Every generated word is edited by a person before it goes live. Not skimmed — edited.
We won't fake social proof.
No generated reviews, no invented testimonials, no synthetic "customers". It's fraud, it's illegal under UK consumer law, and it's the fastest way to lose a brand people trust.
No imagery that misrepresents the product.
Generated backgrounds and crops, yes. A product that looks like something the customer won't receive, no — that's a returns problem dressed up as a saving.
Your customer data doesn't go into public models.
Orders, addresses and support transcripts stay inside tooling with the right data agreements. We'll tell you exactly which services touch what.
It doesn't replace talking to your customers.
AI can summarise a thousand reviews. It cannot tell you the thing a customer says in month three that changes the whole roadmap.
We'll say when it isn't worth it.
Plenty of AI features on Shopify stores are demos that cost money and move nothing. If your budget is better spent on a faster product page, that's what we'll tell you.
The questions clients actually ask about this
Is a human actually involved, or is that just something you say?
Everything we hand over has been through someone here. The honest way to check that with any agency — us included — is to ask who reviewed a specific deliverable and what they changed. If nobody can answer, nobody reviewed it.
Will you use our data to train a model?
No. We use tooling configured so your content and customer data isn't used for training, and we'll tell you which services are in the stack and what each one sees. If you have a DPA requirement, bring it to the first call rather than the last.
Does AI make our project cheaper?
Often, yes — mostly on the laborious parts like migrations, catalogue copy and test coverage. It makes very little difference to design, strategy or anything requiring judgement, so a small project doesn't get dramatically cheaper. We'll show you where it landed in the estimate.
Is AI-written content bad for SEO?
Unreviewed content at volume is, and it's exactly what the recent Google updates went after. Content that's genuinely useful is fine regardless of how the first draft got written. The differentiator is whether a person with product knowledge improved it before it published.
Can you add an AI feature to our store?
Yes — search, recommendations, support, sizing and quiz-style guided selling are the ones that pay for themselves most often. We'll be straight with you about which of those suit your catalogue and which are a gimmick at your size.
What if we'd rather you didn't use AI at all?
That's a legitimate position and we'll work that way if you ask. It will cost more on the laborious parts, and we'd rather agree that up front than surprise you with it.