AI-Assisted Product Registration System
To ease the manual burden of filling out the required fields on the product listing screen, I built an assistant feature where, depending on the field, an ML model or an LLM generates and recommends a value. This walks through how a system that increased the listing conversion rate by 4pp and listings per user by 47% was built.
Situation
Users were entering the product listing screen to sell an item, but a significant share dropped off before completing registration, so the listing conversion rate was measured lower than expected. We suspected the main driver was the manual effort of filling out required fields — title, category, description, and more. So we introduced an assistant feature where AI automatically generates and recommends values as users fill out these fields, aiming to improve the listing conversion rate.
Task
- Soft landing: Bungaejangter's product listing UI generally guides users through filling in various product-information fields from top to bottom, so we needed to build an assist feature that wouldn't significantly disrupt this existing user experience.
- Field-specific design: Each field has a different input/output modality (image → text, text → classification, multimodal → text), so we had to design a suitable model and calling pattern separately for each field.
Action
Result
To evaluate whether the AI-assisted registration feature achieved its goal of helping a user who has decided to sell list their item without friction, our internal data analytics team tracked user behavior on this screen for one month after launch and observed the following changes.
- A weekly-average 4pp increase in listing conversion rate for the AI group versus the non-AI group
- A 47.2% increase in listings per AI user, comparing before and after the feature launched
- A 13.5% reduction in time to first listing for new AI users, measured from their first app visit
In addition, we surveyed roughly 260 users of the AI-assisted registration feature: 85% said the feature was "helpful," and 68% said they noticed a reduction in the time it took to list an item.
Combining this behavioral and qualitative data, we confirmed that AI worked to reduce the burden of listing a product and increase actual listing activity. In particular, we saw clear improvement along two axes: (1) speeding up first-time listings for new users, and (2) driving repeat listings among users of the feature. This suggests AI lowered the barrier to entry early in the selling journey and reinforced a virtuous cycle of repeat listing behavior.