Marketplaces · Reviews
RelayKitchenware & marketplacesReading every Amazon and Flipkart review before ops hears about it
A lid that cracks in the dishwasher shows up in one-star reviews weeks before it shows up in returns data. These field notes cover a Relay that reads every marketplace review, flags defects to ops and drafts replies where a marketplace allows them.
Valnora & Elric team5 min read
Saanjh Kitchenware is a fictional brand, used here as an illustration. It sells about 90 products on Amazon and Flipkart and on its own Shopify store. Reviews on the marketplaces are read by whoever has time, usually on Friday, usually the one-star ones. In this example, a lid hinge started cracking in dishwashers after a new supplier batch. Reviews said so within days. Returns data said so a month later.
These notes cover a setup where every review is read and tagged daily, and a Relay turns patterns into a ticket for ops and replies for a person to post.
Where do the reviews come from?
Neither Amazon nor Flipkart is a built-in integration, so there are two ways to collect reviews in Agent Serve, both run from a scheduled workflow in the workflow builder:
- Apify. The Apify block runs an actor or a saved task and reads the results as a dataset. Several public actors return marketplace reviews as structured rows: rating, title, text, date, verified purchase and review ID.
- Browser Use. A cloud browser agent that reads pages the way a person does. Give it a listing URL and ask for reviews since a date. It is billed per task and suits a smaller catalogue or pages no actor handles.
Before you build either, read each marketplace’s terms on automated access, and check what the seller portal already exports. If a report gives you what you need, use it and skip the collection step.
Each review becomes one row in a reviews table in Tables, keyed by marketplace and review ID so a re-run never creates duplicates.
How are reviews tagged?
From marketplace reviews to a defect ticket and a reply draft
Workflow · Collect and tag
- 01Daily scheduleRuns at 6 am for every product listing in the listings table.
- 02Apify or Browser UseAn Apify actor returns reviews as a dataset, or a Browser Use task reads the review pages.
- 03Tag each reviewTheme from a fixed list, defect yes or no, part affected, sentiment and language.
- 04Reviews tableOne row per review, de-duplicated by marketplace and review ID.
Relay · Review watcher
- 05Watch the tablePicks up new defect-tagged rows and builds a weekly theme summary.
- 06Product and reply guideKnown issues, care instructions and the brand’s reply tone.
- 07Defect ticketA ticket for ops in Jira, Linear or ClickUp, created after approval.
- 08Marketplace managerEdits reply drafts and posts them in the seller portal where replies are allowed.
In the same run, an agent block tags each new review against a fixed list you write: delivery and packaging, product defect, size or fit, quality versus price, listing mismatch, and other. For defects it also names the part, such as lid, clip or seal, from a per-product list.
A fixed list matters more than a clever model. If the theme names change week to week, nothing can be compared. Keep the original text in the row next to the tags, so anyone can check a tag against what the reviewer wrote.
What does the Relay watch for?
The Relay watches the reviews table, checking for new rows as often as every minute, though every hour is plenty here. It runs in Assist mode with three jobs:
- Defects. When defect-tagged reviews for one product and part pass a threshold you set, for example five in ten days against a low baseline, it drafts a ticket with counts, quotes and timing.
- Weekly themes. Every Monday at 9 am, on a schedule, it writes a summary of themes by product and marketplace for the marketplace and ops leads.
- Reply drafts. For reviews where a reply could help, it drafts one from the reply guide in the Knowledge Base.
What the solution produces
Defect ticket drafted for ops (illustrative)
Record- Product
- Glass lunch box set, 3 pcs · Amazon and Flipkart listings
- Signal
- 9 reviews in 10 days mention the lid cracking; 1 in the prior 60 days
- Part
- Lid, clip hinge
- Pattern
- 7 of 9 mention the dishwasher. All from orders after 12 September.
- Quotes
- "Clip snapped after the second wash." "Lid cracked at the hinge in the dishwasher."
- Suggested owner
- Quality lead, to check the September batch from the supplier
Weekly theme summary (illustrative)
Checklist- 412 new reviews this week, 3.9 average. Amazon 268, Flipkart 144.
- Top themes: delivery packaging (61), lid fit (38), size smaller than expected (24).
- New defect theme: lid hinge cracking, lunch box set. Ticket raised.
- Size complaints cluster on the 500 ml jar; the listing does not show dimensions in the first image.
- 18 reply drafts waiting where the marketplace allows a public reply.
Who sees a defect, and how fast?
Creating the ticket is a write to Jira, Linear or ClickUp, so under Assist it waits for approval. The marketplace lead approves it in one click, and the ticket lands with the quality lead with the quotes attached. The approval and the ticket are recorded in the audit log.
In the lunch box example, the quotes pointed at dishwashers and orders after one date. That is enough for the quality lead to ask the supplier about the September batch the same week, and to add "top rack only" to the listing while they wait.
What about replying to reviewers?
Rules differ by marketplace and change over time. Some marketplaces do not offer public seller replies on product reviews at all; others allow them in specific places. Check each one’s current seller policies before you plan any reply work.
The Relay does not post replies. It has no marketplace access to do so. It drafts, the marketplace manager edits, and a person posts in the seller portal where that is allowed. Drafts follow the reply guide: thank the reviewer, address the specific issue, point to care instructions or the support channel, no discounts or requests to change a rating.
Do not use the setup to contact reviewers outside the marketplace. Most marketplaces restrict that, and it is not what this is for.
How do you know it is working?
Add a small dashboard over the reviews table: average rating by product per week, share of reviews tagged defect, and theme counts by marketplace. Watch whether a defect theme falls after a fix ships. That is the measure that matters.
Controls and governance
Approvals before action
Relays never send email or book meetings without an approval, under any autonomy setting. An approval covers the exact payload the reviewer saw and expires if nobody decides.
Audit trail
Proposals, approvals, declines and executed actions are written to the workspace audit log, which admins can export.
Run logs
Every workflow run records each block’s inputs, outputs and errors, so a reviewer can trace how a result was reached.
PII detection
A Guardrails block detects personal data and blocks or masks it before text reaches the next step.
Roles and permission groups
Admin, write and read roles per workspace, plus permission groups that limit which models, blocks and integrations a team can use.
Deployed versions
Workflows run from a deployed version, so a change is reviewed before it reaches production runs.
More on how Agent Serve handles access, data and audit on the security page.
Mask reviewer names in logs with PII redaction, since they add nothing to the analysis. For the same pattern on your own channels, see Instagram social monitoring, and for complaints that arrive by phone, returns and refunds on inbound calls.
Questions teams ask
Will the Relay reply to reviews on Amazon or Flipkart?
No. It drafts. A person posts replies in the seller portal, and only where that marketplace allows a public seller reply. Check each marketplace’s current seller policies first.
Can it contact the reviewer directly?
Do not set it up to. Marketplaces generally restrict sellers contacting buyers outside their own messaging tools. The Relay’s job is to fix the product and the listing, not to chase reviewers.
What you need
- A listings table: product, SKU, marketplace, and the listing URL or ASIN or FSN
- An Apify account with an actor that returns reviews, or Browser Use for pages without one
- A check of each marketplace’s terms on automated collection, and the seller reports it already offers
- A fixed list of review themes and the parts of each product that can fail
- Your reply guide and each marketplace’s rules on seller replies
- A Jira, Linear or ClickUp project where quality issues are tracked