Test an ecommerce candidate in under an hour
Fluency is the one thing an agency or platform background guarantees, and a conversation interview measures nothing else. Four live work samples, fifty minutes, and you will know whether the person can do the job.
- Run four short live exercises over screen share with nothing sent in advance, because take home tasks select for candidates who are between jobs.
- The settlement report, the live listing, the three channel stockout and the ad account cut each test judgement under incomplete information rather than recall.
- Score it with two watchers giving one independent number each on a single question, whether they would let this person own the task unsupervised in month two.
- Never test trivia, tool names, commission slabs or anything a candidate could look up in one search, because none of it predicts performance.
Most ecommerce interviews at this size are ninety minutes of conversation followed by a decision made on fluency. Fluency is exactly what an agency or platform background guarantees. If you want to know whether someone can do the work, give them the work and watch how they move through it.
Set it up so it is short and fair
Four exercises, roughly fifty minutes, run live on a call with screen share. Send nothing in advance. Anonymise everything: strip the brand name, change the SKU codes, round the numbers. Tell the candidate up front that partial answers are fine and that you care more about the order they attack things in than about being right. Then stop talking and let the silence sit.
Sample one: the settlement report with something wrong in it
Hand them one month of a real payout report cut to about forty rows, with a genuine defect left in. A return deducted twice, a commission charged at the wrong slab for the price band, or a shipping fee on an order that was cancelled before dispatch. Ask one question. What is wrong here.
A good answer starts by tying the payout total back to the order total and looking at the gap, then works down the deduction columns. A rehearsed answer explains to you what a settlement report is. Watch for one specific thing: anyone who has actually filed a claim asks for the rate card within the first minute, because you cannot prove an overcharge without it.
Sample two: the live listing, in priority order
Open one of your listings, or a competitor listing in your category, on a phone. Ask what they would change and in what order, and why that order. The order is the entire test.
Weak candidates list everything: title, bullets, images, A plus, backend keywords, price. Strong candidates ask what the listing currently converts at and how much traffic it gets before they say anything, then pick the one change that moves that specific problem. If conversion is the issue they go to the first image and the price. If traffic is the issue they go to the title and the search terms. A candidate who wants a number before an opinion is telling you how they will operate all year.
Sample three: the stockout across three channels
Give a scenario out loud. Your hero SKU has eight days of cover in total. The marketplace fulfilment centre holds four days, the quick commerce dark store allocation holds two, your own warehouse has the rest, and a modern trade order just landed. What do you do first.
There is no single right answer, which is why it works. Listen for whether they think in contribution per unit and reorder lead time rather than in channel loyalty, whether they mention switching ads off on the channel that is about to run dry, whether they think of moving to a smaller pack rather than letting the listing go out of stock, and whether they ask when the next production batch lands. Candidates who say pause the ads on the short channel in the first thirty seconds have lived through a real stockout.
Sample four: the ad account and the target
Show a campaign export of thirty rows: campaign name, spend, sales, ACOS, orders. Tell them budget is coming down twenty percent next month and blended ACOS has to hold. Ask what they would cut.
The tell is whether they cut on ACOS alone. Cutting the worst ACOS campaigns is the answer everybody gives and it is often wrong, because the brand defence campaign with terrible ACOS may be the only thing keeping a hijacker off your own keyword, and the upper funnel campaign that looks bad on last click may be feeding the branded search that looks good. A candidate who asks which campaigns are branded before cutting anything has run an account with their own hands.
Scoring it without a rubric culture
You do not need a formal scorecard and a team of your size will not maintain one. You need two people watching the same exercise, each writing down one number from one to five before either of them speaks, against a single question: would I let this person own this task unsupervised in month two. Compare afterwards. If the two scores are more than two apart, one of you was scored by charm.
Then write one sentence per exercise describing what the candidate actually did, not what they said. Those sentences are what you will read in six weeks when four candidates have blurred into one.
What not to test
Do not ask what the commission rate is in beauty on a given marketplace, what a particular report is called, or the difference between two campaign types. It is all one search away and none of it predicts anything. Do not set a take home that costs more than two hours, because good candidates who currently have jobs will decline and you will quietly select for the unemployed. Do not ask for a strategy deck about your own brand, which is unpaid work and reads as such to exactly the people you want. Test judgement under incomplete information, because incomplete information is the actual job.