Two different problems: testing automation is not handing over daily store work
Testing automation and daily task automation are two different problems. Search “Android multi-device automation” and the top results speak to developers and QA: widget-based frameworks, device farms, the official multi-device test harness, remote-control tools. They assume you verify features, not that you take store listing and reply work off one person's desk.
Testing automation maintains test cases; operational task automation maintains judgement and error tolerance. Taking daily tasks with a testing framework runs into a different failure: after a UI change, element locators stop resolving and the script fails at the first step (see why scripts break after a UI change).
Three routes, and who each one suits
- Widget-based frameworks such as Appium or uiautomator locate elements through the widget tree, and when the structure changes existing locators stop resolving.
- Device farms and cloud phones are rented per cloud device and built for testing and parallel runs at scale; when the target is a local store back office, account environment and marketplace rules stay your job.
- The screen-understanding route uses no root and no fixed coordinates, reading what is on screen and deciding where to tap; teams with no developer resources usually take it, and its boundaries are covered in choosing a no-root Android automation tool.
The screen-understanding route builds on Android's official accessibility service: it lets an app read screen content and perform taps and scrolls without root. Google Play also limits what the accessibility API may be used for, and prohibits using it to remotely control other people's devices.
Listing tasks: what can be handed over whole
Tasks worth handing over whole: batch listing and delisting the same product set across devices; format-fixed field sync such as titles, images and category fields, without prices; timed runs that need nobody standing by; and a log for every run.
Tools such as AnsClaw describe this capability as mobile batch operations — running the same set of actions across multiple devices. What differs between tools is how each decides what is on screen, and what it costs to maintain.
Replies: only half can be handed over
Keep a human on anything that touches money — haggling, top-ups, order changes — anything that becomes a promise to the buyer, such as shipping time, compensation or after-sales wording, and anything heading for a dispute or platform involvement.
Pool unread messages from several devices so none is missed, draft replies for a human to approve, and keep a library of reusable wording (see handling messages from multiple phones in one place). Read your marketplace's own terms on automated actions and automated messages before starting.
Where handing over too much sends work back
Wrong judgement: replying to conversations that need a human with a template can damage the customer experience and escalate to platform involvement. Wrong action: a mistaken price or stock edit costs money directly. Broken locators: fixed-coordinate scripts tap empty space after a UI change, hitting every device in that batch at once.
The test is not whether something can be automated, but whether you can absorb the cost of a mistake: hand over whole the cheap, checkable work; keep a human on anything that goes outwards or cannot be withdrawn, especially anything involving discounts or limited-time campaigns (see timed campaigns: reminders only, no auto-ordering).
Rollout order
- Pick one purely repetitive action, such as batch listing and delisting, and trial it on one or two devices only
- Run it several times and watch for taps that hit nothing or hit the wrong thing
- Once it holds, add devices in small batches
- Write the human confirmation step down as a checklist: who confirms, and within how long
- Review the run logs regularly and fold new repetitive actions into the process
One test before handing anything over whole: can the mistake be undone, and is the success condition unambiguous? Only when both answers are yes.