Seller retention in marketplaces: the first sale more than doubles the chance a seller comes back

Marketplace · Series A

Sales growth depended on supply: the sellers and the items they listed. The company wanted to know which seller profile was worth more and how to bring sellers back, but treated the whole base the same way.

Read the historical seller and inventory base, always by cohort, to find the signals that predict return and turn them into groups and actions.

• We crossed seller return with each seller's first entry: whether it sold, how long it took, how many items were listed and at what price.
• We analyzed seller cohorts month by month and inventory age, to see from when items stopped selling.
• We took the findings into two action groups: seller profiles by return potential and a plan for stranded items.

The first sale more than doubled the seller return rate. Time to that sale barely changed return, which stayed at practically the same level for sellers who sold in the first week and for those who waited months. Sellers who started with more items came back far more often than those who started with one or two. And a stranded item had a very low chance of selling. The analysis fed the seller account portfolios and a discount campaign for stranded inventory.

• In a marketplace, the seller's first experience decides whether they come back. More than a fast sale, what matters is that the sale happens.
• The size of the first entry is an early signal of value: those who start big tend to stay.
• Stranded inventory loses value. From a certain age on, the item's price or visibility has to change.

1. Do everything by cohort, grouping sellers and items by month of entry. Mixing cohorts hides the effect of time and distorts the rates.

2. Build three tables, all by cohort:

   • seller retention, with one row per entry month and, in the columns, the percentage that came back in month 1, 2, 3 and so on up to 12;
   • return by first entry: within each cohort, the return rate according to what happened in the first entry (sold or not, number of items, average value);
   • inventory age: within each item cohort, the chance of selling by month of age, to decide when a discount or more visibility kicks in.

3. Ask an LLM, after removing names and contact details from the base:

   I'll paste an export with one row per seller: entry date, items listed in the first entry, average value per item, whether the first item sold, days to sale and whether they listed again within 12 months. Group everything by monthly entry cohort. For each cohort, calculate the return rate by item band, by value band and by whether the first item sold, tell me which signals predict return most consistently across cohorts and propose 3 seller groups with one action for each.

   Copy