A Shopify store manager that actually replies to customers

For 30 days, we ran a Vezra store manager agent on a real Shopify store — a small apparel brand doing about $80k/month, one founder, no full-time staff. The brief was simple: handle every customer message, keep the storefront tidy, and ship one abandoned-cart campaign a week. No marketing polish in this post, just what we saw.

Day zero — the setup

Setup took 11 minutes. We connected Shopify (OAuth), Gmail (for customer replies), and Klaviyo (for the cart flow). We gave the agent a one-page brand voice doc: "warm but not cutesy, always offer a solution, never apologize more than once." It read the store's existing order history and most recent 200 support threads before doing anything.

Nothing dramatic happened in the first hour. That's what we want.

Week one — support on training wheels

For the first week we ran the agent in draft mode. Every reply landed in the founder's inbox as a drafted Gmail thread, ready to send. This is the single highest-leverage thing anyone hiring an AI agent can do in their first week, and we preach it constantly.

By Thursday the founder was sending the drafts without reading past the subject line. The voice was right, the refund logic was right, and the agent had stopped asking to escalate trivial things. We promoted it to full send on Friday.

That week, by the numbers:

• 142 customer emails handled. The founder intervened on 9 — mostly edge cases (one fraud suspicion, two order mixups where the agent correctly flagged them).
• Median reply time dropped from 14 hours (the founder's old baseline) to 6 minutes.
• Zero "why are you replying like a robot" complaints. Two compliments on the voice.

Week two — proactive work

The agent started doing things without being asked. This is the moment AI teams start earning their keep.

It noticed the store had 23 products with inventory at zero but restock dates in the future. It wrote a short email explaining the fix (mark them as "coming soon" instead of unavailable) and waited for approval. The founder approved. It fixed them in 11 minutes.

It also noticed the abandoned-cart email was using the store's default copy — generic and weak. It drafted three variants matching the brand voice and proposed an A/B test. We shipped variant B. Cart-recovery revenue that week went up 38% vs. the 4-week baseline.

Week three — the mistake

This is the honest part. On day 19 the agent made a real mistake. A customer in the UK asked about a return. The agent correctly pulled the order and correctly quoted the return policy — but then misread the UK-specific free-returns rule as applying to the US policy. It told the customer they'd have to pay postage. They didn't.

The customer pushed back. The agent escalated to the founder. Nothing actually broke — no money was lost, no shipment went wrong — but the first reply was wrong and we corrected it in the next message. We updated the agent's brief with a region check, and we haven't seen the bug again.

This is exactly what weeks two and three look like for every single Vezra team we watch. Small calibrations. Nothing catastrophic.

Week four — numbers

At the end of the month, we pulled the data. The agent had:

• Replied to 611 customer messages.
• Shipped 4 Klaviyo campaigns, with 2 A/B tests.
• Closed 47 open cart-abandonment tickets.
• Flagged 12 inventory issues the founder hadn't noticed.
• Drafted the weekly store-performance summary every Monday.

The founder's weekly time-on-Shopify dropped from roughly 18 hours to about 4. The 4 hours were spent reviewing drafts, approving campaigns, and answering the agent's occasional escalation questions.

What it didn't do

We want to be precise here. The agent did not:

• Invent marketing strategy. The founder still decided what the brand was promoting and why. The agent executed.
• Talk to suppliers. That stayed human.
• Handle anything outside the tools it was connected to (e.g. no customer voice calls).

A Vezra agent is a teammate, not a replacement for judgment. The founder is still the founder.

Is this a one-off?

No. We've run versions of this on four other Shopify stores, two DTC brands on WooCommerce, and one marketplace. The specifics vary — voice, policies, which products get flagged — but the shape of the rollout is identical. Week one is calibration. Week two is proactive gains. Week three is one small correction. Week four is "how did we ever do this by hand."

Hire one yourself

You can spin up the same agent on Vezra in under 15 minutes. The first thing it'll do is read your last 200 support threads. The second thing it'll do is ask you how you want it to sound. After that, it works.