How one Shopify brand clawed back 10 hours a week using AI (properly)

A case study in what happens when you stop treating AI like a novelty and start treating it like a tool.

Every ecommerce founder we speak to has heard the AI pitch. Most of them have tried ChatGPT. A smaller number have paid for a subscription. Almost none of them have rolled it out across their business in a way that actually changes how the work gets done.

That is a shame, because the businesses that get it right are quietly pulling ahead. Not because the AI is doing anything magical, but because they have taken the mundane, repetitive parts of running an ecommerce business and pushed them off the humans who used to do them.

Here is what that looks like in practice, drawn from a recent rollout we did with a UK skincare brand that sells direct to consumer on Shopify and to trade through wholesale.

The starting point

Before we got involved, the brand looked like most growing Shopify businesses. Customer service ran through a shared Outlook inbox where two people spent the first 90 minutes of every day triaging urgent messages, then the rest of the morning drafting replies. Every reply required them to open Shopify in a second tab, look up the customer's order, then write a personalised response.

The marketing team built the weekly dashboard by hand, copying numbers out of Shopify Analytics, Recharge, and Klaviyo into a Google Sheet. It took about three hours every Monday and by the time it was ready, the meeting it was for was almost over.

The B2B sales team was in a worse spot. Their CRM had recently been swapped from HubSpot to a niche B2B tool, which had left them without email open tracking, without bulk mail merge, and without a way to auto-log conversations. Everything was manual and nothing was measurable.

And ChatGPT? A few people had personal subscriptions. They used it "like a glorified Google". Nobody had a shared way of working. Nobody trusted its output enough to send anything it wrote without heavy rewriting.

What we did

The rollout took one training day and about three weeks of implementation work. Here is what actually went in.

A proper AI policy first, not last. Before we let a single person paste customer data anywhere, we wrote a one-page AI usage policy and a fuller admin handbook. Nothing scary, five rules that fit on a postcard: don't invent facts, don't publish anything without a human review, don't paste sensitive data, use the right tool for the job, and ask before you do something risky. Leadership signed it off in a morning. Every team member reads it during induction.

Team Projects with brand context baked in. Instead of everyone starting from a blank chat, we set up dedicated Projects for each function: customer service, retail sales, spa sales, marketing, finance, management. Each Project has role-specific instructions ("You are helping the customer service team at [brand]. Warm tone, UK English, never promise a refund without checking Shopify") loaded from a single markdown file. Every chat inside a Project inherits those rules for free. No more re-explaining who the brand is at the top of every conversation.

The customer service automation. This is the one that changed the most. We built a Microsoft Power Automate flow that watches the shared customer service inbox. When a new email arrives, it calls Shopify to look up any orders belonging to that customer, hands the email and the order data to Claude with a library of approved reply templates, and puts the drafted reply back in the Outlook drafts folder. The customer service team open their drafts folder in the morning and find most of the day's work already written. They review, tweak the ones that need it, and send from the same address as always. The customer sees no difference. The team gets 1 to 2 hours a day back.

Scheduled research. A weekly task now runs on Thursday morning that searches for new spa openings, hotel news, garden centre buying changes and personnel moves in their sector. The digest lands before the Friday sales meeting, giving the team specific, current things to talk about with prospects. Cold outreach becomes warm outreach.

Dashboards that build themselves. By connecting Claude to Shopify, Recharge and Klaviyo through their APIs, the weekly marketing dashboard now populates itself overnight. Monday morning starts with a discussion, not with copy-paste.

What changed

Numbers, honestly reported.

• Customer service morning triage: from ~90 minutes to ~25 minutes.
• Drafted replies per agent per day: from ~40 written from scratch to ~60 reviewed and sent, because the initial draft is already there.
• Weekly dashboard build time: from ~3 hours to ~15 minutes of sanity checking.
• Cold outreach open rates (early data): meaningfully up, though we want another 90 days before we quote a firm number.

Total time saved across the business, conservatively: 8 to 12 hours a week. Running cost, all in: around £70 a month.

The bit most rollouts get wrong

Two things.

The first: they treat AI as a tool for one person to use, not a system for the business to run on. That means each team member reinvents their own way of using it. Some get good, most get frustrated, nobody agrees on what "good" looks like. Ours was set up so that everyone in a given team works inside the same Project, with the same rules, drawing from the same templates. The consistency is what makes it stick.

The second: they skip the policy step because it feels bureaucratic. This is a mistake. The moment somebody pastes a full customer database into a chat because they were in a hurry, you have a problem. The policy is not there to slow things down, it is there to make everyone confident enough to move quickly.

What this means for your Shopify business

If you have a shared customer service inbox, a growing wholesale side, and a manual dashboard, you are almost certainly sitting on 5 to 10 hours a week of admin that could be automated. The blueprint is repeatable. The tools are all off the shelf. Most of the work is one-time setup that pays back inside a month.

The interesting question is not whether AI can help. It can. The question is whether your business has the discipline to roll it out properly, one team at a time, with a policy behind it and a shared way of working. That is the part we help with.

If any of the above sounds like your week, we should talk. Not a pitch, just a 30-minute conversation to see whether it is the right fit.

Okapi & Co helps ambitious UK businesses build AI into how they actually work, not just how they talk about it. We work with founders, marketing leads and operations teams who want the results without the hype. Get in touch.