By hand
AI automation forShopee, Lazada andTikTok Shop operations
Five jobs a marketplace team does by hand every week. This side is how each one is done today.
By hand
Settlement reconciliation
Download every Shopee and Lazada settlement statement. Match each order line to the invoice. Hunt for the mismatches. Rebuild the Excel. Every cycle.
By hand
Campaign stock forecast
Pull sales by SKU. Open the campaign calendar. Guess the uplift for the next mega sale. Email the reorder list and hope the warehouse reads it in time.
By hand
Listing generation
Open the brand website. Copy the spec. Write a title that fits the character limit. Repeat for every SKU, on three platforms.
By hand
Campaign pricing and vouchers
The platform changes its voucher mechanics. Recalculate seller vouchers across three price levels so the campaign price still clears the buffer. Again next campaign.
By hand
Customer service answer bank
The same twenty questions, answered from memory, in a different tone each time, across Shopee chat, Lazada chat and TikTok Shop.
You
How an agent gets built
- Write down how the job is done today. One page. We help you write it.
- Check the first run against a cycle you did by hand.
- Sign every output. The agent drafts; a person approves.
Not for you if
- You sell on one platform with a dozen SKUs.
- The process lives in someone's head and nobody will write it down.
- You want a chatbot on your website.
A battery brand
Dozens of SKUs across Shopee and Lazada Singapore. Reconciliation was a full day of matching statement lines to invoices, each cycle, by hand. Stock for the mega sales was forecast in a spreadsheet the week before.
- Settlement reconciliation
- Campaign stock forecast
- Listing generation
- Campaign pricing and vouchers
- Customer service answer bank
By agent
The same five jobs,run by an agent.
Built off your written SOP. Run on your own data, inside accounts you own. A person signs every output.
Book a CallBy agent
Statements in. Every line matched to the invoice. Mismatches flagged with the reason. Excel and PDF out, ready to send.
- Input
- Shopee and Lazada settlement statements, your invoice export
- Output
- A variance table with every mismatch and its reason, as Excel and PDF
- Runs
- Every statement cycle
A person signs the variance sheet.
By agent
Watches every SKU against the campaign calendar. Projects the stock-out date per SKU. Outputs reorder quantities before the campaign locks.
- Input
- Sales by SKU, stock on hand, the platform campaign calendar
- Output
- A reorder quantity per SKU with its projected stock-out date
- Runs
- Weekly, and before every campaign lock
A person approves the reorder.
By agent
Brand site and spec sheet in. Titles inside each platform's limit, bullets, descriptions and attributes out, in the upload format each platform takes.
- Input
- The brand website or spec sheet, your listing standards
- Output
- Upload-ready listings for Shopee, Lazada and TikTok Shop
- Runs
- Per SKU batch, on demand
A person runs the QC pass before upload.
By agent
Your pricing rules, encoded once. Feed it the platform's voucher change and it returns the recalculated seller vouchers, checked against the floor price.
- Input
- Your price levels and buffer rule, the platform's voucher change
- Output
- Recalculated seller vouchers per SKU, per platform
- Runs
- Every campaign, every mechanic change
A person checks one row by hand.
By agent
Drafts each reply from your answer bank and the order record. Flags the hard cases to a person. Drafts, never sends, without sign-off.
- Input
- Your answer bank, the order record, the incoming message
- Output
- A drafted reply in your tone, or a flag with the context attached
- Runs
- Every message, every platform
A person sends.
The agent
- Built off that page, on your real data, inside accounts you own.
- Runs in parallel with the manual process until the two agree.
- Runs on schedule. Every mismatch, exception and hard case comes back to a person.
For you if
- You run a brand on Shopee, Lazada or TikTok Shop with a statement to reconcile every cycle.
- One person loses a day a month to a spreadsheet the platform export could have filled.
- You already work with an enabler, or you are one.
Now
The statements come in, the agent reconciles them, flags the mismatches and sends the Excel and PDF to be signed. A second agent watches the same SKUs against the campaign calendar and outputs reorder quantities. Both were built off a written SOP, without writing code.
The grind moves across. Judgment stays.
Five jobs running on schedule, every output signed by a person. Bring one job your team does by hand and leave with a written scope.
Book a CallWhat is an ecommerce operations agent?
An operations agent is software that does one recurring marketplace job end to end. It takes the inputs your team takes (a settlement statement, a sales export, a brand's spec sheet), applies the rules your team applies, produces the output your team produces, and hands it to a person to sign. It is not a chatbot and it is not a dashboard. It is the weekly job, done.
Atsell builds them off a written SOP, runs them on the client's own data inside accounts the client owns, and keeps a person on every output. That is the model we run in our own operations: AI-assisted operations, human-led strategy. The agent drafts; a person signs.
What an agent will not do
Read a category. Negotiate with a platform account manager. Decide how your brand is positioned against a named competitor. Own the number when a target is missed. Those stay with operators, which is why "AI-powered" is not a strategy and why the jobs on this page are the ones we automate first.
Frequently asked questions
How long does it take to build an operations agent?
Most operations agents are running in parallel with the manual process within a few weeks of the SOP being written down. The gating item is the SOP, not the build. Jobs that touch more data sources take longer to reconcile.
Where does our data go?
The agent runs inside accounts you own. Contracts, brand pricing sheets and anything under NDA stay out unless you decide otherwise. The data rule is set on day one, in writing, before anything is built.
What happens when the agent breaks?
It stops and tells a person. Every agent is built to flag rather than guess: a mismatch it cannot explain, a SKU it has not seen, a question outside the answer bank goes to your team with the context attached.
Does this replace my team?
No. It removes the part of their week that a platform export could have done. The category calls, the platform relationships and the customer escalations stay with people, because that is the part that grows the store.
Do I need to be an Atsell client?
No. Atsell builds agents for brands that run their own stores and for other enablers. If Atsell already manages your store, the agents run inside that retainer.
How is it priced?
Per agent: a build fee plus an optional monthly run fee, scoped from the hours the job takes today. Book a call and bring the job. You will get a written scope, not a deck.
Human signs. Agent runs.
Bring the job. Leave with a scope.
Walk us through one job your team does by hand. You will get a written scope for the agent that runs it, priced from the hours it takes today.