Ecommerce ops · Production

OLX uses Prosus AI Assistant to extract and normalize job roles from job ad titles and descriptions

The problem

OLX lacked clearly defined job roles within its jobs taxonomies, with roles buried in ad titles and descriptions, creating a barrier to efficient and organized search for job-seekers.

Workflow diagram · grounded in source
1
Build job-role taxonomy tree
ai_action
“leveraging up to the top 100 profession-related searched keywords and up to 50 job roles extracted from randomly selected job ads within the specified category. The process was guided by a well-defined prompt, resulting in a hierarchical…”
2
Ad event triggers extraction
trigger
“A new service subscribed to ad events utilized Prosus AI Assistant to extract job taxonomy information”
3
Preprocess ad content
ai_action
“meticulous text cleaning, trimming to the first 200 words/tokens, and a touch of translation magic”
4
Extract job roles from ad
ai_action
“A dedicated prompt facilitated the identification of relevant job roles within the established taxonomy tree”
5
Send roles to Kinesis
integration
“sent job roles to the Kinesis platform used by our Search team”
Reported outcome

The A/B test showed positive uplift in most Successful Events metrics and a significant decrease in search extensions and keyword searches per user in the low-result segment, though some impact metrics had small effect sizes and not all reached statistical significance.
Operating cost was approximately 15K per month, prompting consideration of self-hosted alternatives.

Reported metrics
Job-seeker keywords focused on specific professions60%
Daily newly created or updated ads processedaround two thousand
daily API requests to Prosus AI Assistantapproximately four thousand daily requests
monthly cost of Prosus AI Assistantapproximately 15K per month
Show all 7 reported metrics
job-seeker keywords focused on specific professions60%
daily newly created or updated ads processedaround two thousand
daily API requests to Prosus AI Assistantapproximately four thousand daily requests
monthly cost of Prosus AI Assistantapproximately 15K per month
initial accuracy checknotably low incidence of flaws
search extensions and keyword searches per user (low-result segment)significant decrease
A/B test uplift in Successful EventsPositive uplift in most metrics
Reported stack
Prosus AI AssistantLangChainKinesis
Source
https://tech.olx.com/extracting-job-roles-in-job-ads-a-journey-with-generative-ai-e8b8cf399659
Read source ↗

Frequently asked questions

What did this team achieve with this AI workflow?

The A/B test showed positive uplift in most Successful Events metrics and a significant decrease in search extensions and keyword searches per user in the low-result segment, though some impact metrics had small effec…

What tools did this team use?

Prosus AI Assistant, LangChain, Kinesis.

What results were reported?

Job-seeker keywords focused on specific professions: 60%; Daily newly created or updated ads processed: around two thousand; daily API requests to Prosus AI Assistant: approximately four thousand daily requests; monthly cost of Prosus AI Assistant: approximately 15K per month (source-reported, not independently verified).

How is this ecommerce ops AI workflow structured?

Build job-role taxonomy tree → Ad event triggers extraction → Preprocess ad content → Extract job roles from ad → Send roles to Kinesis.

Related ecommerce ops cases