The AgentivUse cases

Every Shopee review answered, in your voice

An agent that reads new Shopee and Lazada reviews each morning, drafts a reply to each in the shop's tone and posts them after the owner approves

Function: Customer serviceEffort: About a weekPublished: 2026-10-20

The problem

Reviews pile up across two seller centres and most go unanswered, while the one star review with a shipping complaint sits in public for days. Replying by hand means copying the same thank you note forty times and still missing the one that needed a real answer.

How the agent works

  1. It starts

    Each morning the agent calls Shopee's get_comment for the shop's active items and Lazada's seller review list for reviews submitted since the last run, using the shop's own Open Platform credentials.

  2. The agent works

    It groups them by rating: four and five stars get a short, varied thank you in Thai or English matching the buyer's language; three stars and below get a draft that names the specific complaint, apologises once and offers the next step (a replacement, a refund through the platform, or a LINE OA contact). The drafts go to the owner's Slack or LINE group as a list with the review text, the star rating and the proposed reply; the owner approves, edits or skips each one. Approved replies are sent through reply_comment on Shopee and the seller reply endpoint on Lazada, and one row per review is appended to a Google Sheet so complaints can be counted by SKU at month end.

  3. The result

    Low rated reviews are never auto replied; they always wait for the owner.

Worked example

Both platforms expose the calls in their Open Platform APIs, documented in community SDKs: Shopee's v2.product.get_comment returns rating_star, comment, buyer_username and ctime per item and v2.product.reply_comment posts the reply by comment_id; Lazada's /review/seller/list/v2 returns review_content, four rating fields, can_reply and seller_reply, and /review/seller/reply/add takes the review id and a reply of up to 500 characters. Lazada's review history is fetched in windows of up to seven days, so a daily run fits comfortably. Neither SDK publishes reply rates or sales effects, so measure response time and complaint count on your own shop.

PDPA note. Reviews carry buyer usernames. Write only the review id, product, rating and reply to the sheet; do not copy buyer names or order details out of the platforms.

Join the Agent OS Club

8 weeks to the AI team a 50 person company would hire. Instant access to our agent blueprints. USD 15 a month, cancel any time.

Join the Agent OS Club

Building this for an organization? Learn more at theagentiv.com

Sources

Related use cases

Hear the bad review before it goes public

An agent that asks each customer for a one tap rating on LINE after the job is done, thanks the happy ones and alerts the owner to the unhappy ones with a drafted reply

Customer serviceEffort: About a weekLINE · Google Sheets · Claude

A clinic, salon, repair shop or delivery business only learns a customer was unhappy when the review appears on Google Maps or Facebook. Asking everyone by hand does not happen, and paper feedback forms end up in a drawer.

Your LINE OA takes the bookings

A LINE Official Account assistant that takes bookings in chat, checks Google Calendar, confirms the slot and sends a reminder the day before

Customer serviceEffort: About a weekLINE · Google Calendar · Gemini

Salons, clinics, restaurants and repair shops in Thailand take bookings through LINE chat, so a staff member reads every message, checks the paper book or calendar, types a reply and later remembers to send a reminder. Evenings and busy hours leave customers waiting, double bookings happen, and no-shows cost the slot.

Your LINE OA answers the same five questions

An agent that answers FAQs and order status on your LINE Official Account and hands the rest to staff

Customer serviceEffort: About a weekLINE · Claude · Slack

Most LINE OA chats ask the same five things: price, stock, shipping time, where my order is, and how to pay. Admins answer them one by one until late evening, and the one message that really needs a person, a complaint or a refund, waits in the same queue.