A private-label seller was answering a third of its marketplace reviews, days late. We built an AI pipeline that drafts a tone-matched reply to every review and queues it for one-tap approval. Today 100% of reviews get answered — median time, six hours.
Typical results
6h
Median response time
400
Reviews handled monthly
1 tap
To approve a reply
Figures reflect typical results for this kind of automation.
The challenge
Around 400 reviews landed each month across Amazon and two other marketplaces. The team answered maybe 30% — whatever they caught between shipments — and the median reply took five days. Praise went unthanked, complaints festered publicly, and a one-star review could sit unanswered all weekend, quietly costing the buy box and future sales.
Before and after
Reviews receiving a response
3.3×Median response time
−95%Weekly hours writing replies
−86%The automation
We connected every marketplace account to an hourly polling flow. Each new review is classified by sentiment and passed to OpenAI with the product context and the brand's tone guide; it drafts a reply — thankful for praise, solution-oriented for complaints. Drafts arrive as WhatsApp cards the team approves, edits or rejects in one tap, then post back to the right listing. One- and two-star reviews also open a support ticket with full order history attached.
How it works
- 01
Poll
New reviews pull from every marketplace account hourly.
- 02
Draft
AI writes a tone-matched reply using product context and the brand voice guide.
- 03
Approve
The team taps approve or edit on a WhatsApp card — ten seconds per review.
- 04
Publish
Approved responses post back to the right listing automatically.
- 05
Escalate
One- and two-star reviews also open a support ticket with the order attached.
Starts from
- Marketplace reviews
- Listing catalog
Delivers
- One-tap approval card
- Published reply
- Support ticket
The impact
Every review — praise or complaint — now gets a brand-voice answer, with a median response time of six hours instead of five days. The team spends about an hour a week on approvals instead of seven writing from scratch, and unhappy buyers hear from support before they escalate.
Questions about this automation
How does the AI match the brand's voice?
Each draft is written by OpenAI from three inputs: the review text and its sentiment, the product details from the listing catalog, and the brand's tone guide. Praise gets a short thank-you, while complaints get an apology and a practical next step. The team can edit any draft before it is published.
Does anything post without a person checking it?
No. Every draft arrives as a WhatsApp card with approve, edit and reject options, and only approved replies are posted back to the listing. Most reviews take about ten seconds to approve. Rejected drafts are dropped, and a team member can write that reply by hand instead.
What happens with one- and two-star reviews?
Low ratings get a drafted public reply like any other review, and they also open a support ticket with the buyer's order history attached. That way support can reach the buyer through the marketplace's messaging and fix the issue while the public reply is still being approved. This is a multi-step build, usually rolled out in stages.