25 days to first sale: the marketplace activation benchmark
Here is the number that should keep every marketplace operator up at night.
Twenty-five days. That's the gap, across 65,000 sellers, between "a merchant decided to sell on your platform" and "a customer actually bought something from them." Almost a month in which a motivated seller who wanted to make money on your marketplace couldn't. Every day in that window is GMV that doesn't exist, a seller wondering whether they backed the wrong platform, and a catalog that isn't earning.
The instinct is to blame the seller — slow to upload, bad photos, incomplete data. Sometimes true. But when the same 25-day drag shows up across tens of thousands of sellers on enterprise platforms, it isn't a seller problem. It's a process problem, and it lives almost entirely in one place: catalog review.
Catalog review is the hidden tax
A seller signs up, uploads products, and then their listings enter a queue. Someone has to look at each one. Is this categorized correctly? Are the images actually of the product? Is anything copyrighted, counterfeit, or against policy? Is the content good enough that a buyer will actually click? None of that is optional — a marketplace that skips catalog QA drowns in garbage listings and buyer complaints within a quarter. But all of it sits between the seller and their first sale.
And it doesn't scale by adding people, at least not gracefully. Every new seller adds products; every product adds a review; every review adds a human-minute to a queue that already had a backlog. The queue is the constraint, and the queue grows faster than any QA team you can hire against it. So the 25 days isn't the seller being slow. It's your review capacity being the bottleneck, quietly capping how fast anyone can go live.
What the tax costs, before and after
Getir ran exactly this queue, at a scale that makes the problem impossible to ignore: 90,000 restaurants, roughly 6,000 images reviewed every day, and a 25-person manual QA team that was the bottleneck. Restaurants couldn't take orders until their products were approved, so every hour of review delay was GMV lost — not eventually, but that day.
The team wasn't slow. The work was just enormous and almost entirely mechanical: classify the request, check for copyright issues, judge whether the content is good enough, approve. Ninety seconds a task, thousands of tasks a day, one human doing all four steps on every single item. Here's what changed when an agent took over the assembly and humans moved to approving in batches:
| Metric | Before | After |
|---|---|---|
| Time per task | 1.5 seconds | |
| QA team size | 5 people | |
| Photo coverage | +30% | |
| Time to go live | Same day |
~€150K/yr in savings, plus €1M+ upside from restaurants going live the same day instead of three days later.
Read the last row first, because it's the one that moves GMV. Restaurants that used to wait three days to go live now went live the same day. Multiply that across a stream of new merchants and it's worth more than €1M — dwarfing the ~€150K/yr the team saved on headcount. The savings are nice. The activation speed is the business case.
And note what didn't happen: photo coverage went up 30%. Automating the review didn't lower the bar, it raised it, because the agent could touch every item consistently instead of a stretched team sampling what it had time for.
The flow that did it
The work didn't disappear — it got re-shaped. Four stages, agent-driven, with a human approving in batches at the end instead of grinding through every item one at a time:
Click a step. The agent runs all of them; a human confirms the last call.
The agent reads each incoming product and sorts it into the right category and review path automatically — no human deciding where a listing belongs before anyone can even look at it.
Nothing here removes the human judgment that keeps a catalog clean. It removes the 90 seconds of mechanical assembly that stood in front of every judgment call. The reviewer stops classifying, copyright-checking, and formatting by hand, and starts confirming batches of work that's already done. Five people now do what took twenty-five, and they do it on more items, faster.
Model your own activation drag
Getir's numbers are Getir's. Yours depend on your seller volume and how much of your catalog review is genuinely judgment versus mechanical assembly. Put your own daily review volume in and see how much of it a flow like this clears automatically versus how much still needs a human's call:
Drag to your daily case volume. Qrambo clears the routine ones; your team stays on the 25% that need judgment.
Illustrative, based on a 75% auto-resolution rate and 1.5 min per manual case. Your numbers are set in the pilot.
The exact split will move with your policies and your catalog. But the shape holds: most catalog review is classification, policy-checking, and formatting — mechanical work that an agent does in a second and a half instead of ninety. The genuinely subjective calls, the edge cases that need a human eye, are the minority. When the mechanical majority clears itself, your reviewers spend their day on the calls that actually need them, and your sellers stop waiting three days — or twenty-five — to make their first sale.
The 25 days isn't one wait — it's a chain of small ones
Part of why the benchmark is so stubborn is that nobody owns the whole 25 days. Break the seller journey into stages and each one adds a few days that look reasonable in isolation and lethal in aggregate.
A seller uploads their catalog, and it enters the review queue — a few days, depending on backlog. Some listings bounce back for a fix: wrong category, poor image, missing attribute. The seller corrects them, re-uploads, and re-enters the back of the queue — another few days. A policy flag needs a second look — another wait. By the time everything clears and the products are live, three or four rounds of a-few-days-each have stacked into most of a month, and no single handoff ever felt like the problem.
This is why "hire more reviewers" barely dents the 25 days. Adding people speeds up one stage of a multi-stage chain, and the seller still round-trips through the queue every time a listing bounces. What actually collapses the timeline is cutting the number of round-trips: catch and fix the fixable problems in the flow instead of bouncing them back to the seller, and clear the routine majority automatically so the queue never builds the backlog that makes each stage slow in the first place.
That's exactly what Getir's "enhance content quality" stage does — the agent improves and standardizes listings so more of them clear the bar on the first pass instead of ricocheting back to the restaurant. Fewer bounces, shorter queue, faster activation. The +30% photo coverage isn't a side effect; it's the mechanism. Better first-pass quality means fewer of those a-few-days-each round-trips, which is where the 25 days actually lives.
What to check in your own funnel
If your marketplace is anywhere near the 25-day benchmark, the diagnosis is usually one queue. Pull your last cohort of sellers and measure the gap between signup and first sale, then find where the days actually accumulate. If it's catalog review — items sitting in a queue waiting for a human to classify, check, and clean them — you have Getir's problem, and it responds to the same fix: let an agent do the assembly, keep a human on the approval, and clear it in batches instead of one item at a time.
The move generalizes past food delivery. Any marketplace where listings must pass review before they can sell — retail, services, B2B catalogs, rentals — has the same anatomy: a review queue that grows faster than the team reviewing it, with real GMV bleeding out the gap while sellers wait. The retail and marketplace playbook covers catalog QA, seller activation, and the batch-approval model in more depth. The one number to walk away with: your reviewers' time is your activation speed, and right now most of it is going to work a machine should be doing.