Explore

PRD - FIF personalisation

Problem -

Gold VC is massively under indexed on search due to absence of intent channelisation levers to divert high intent traffic in relevant searches to premium gold listings.

We need to increase Gold VC by channelising intent from mixed PLP feeds across relevant PSTs and collection / CLPs.

Current context -

FiF widgets can be setup self serve via admin as follows -

⁠

⁠

⁠

⁠

Problems scaling the RE with current construct -

Each PST has to be defined manually against the widget which is an overhead reducing net coverage (Scope)

⁠

Each widget to collection / CLP mapping has to be defined manually since CLP creation doesn’t support reading from PST input (Not part of scope)

Objective -

Creating a bulk ingestion mechanism to showcase respective FIF widgets for relevant PSTs / collections.

Impact -

Coverage across head 60% PSTs / collections would ensure Gold VC boost by ~0.3-0.4pp

Recommended approach -

Adding a bulk PST ingestion mechanism that lets ops users tag -

Multiple PSTs per FIF widget group

Each PST with dedicated audience IDs / position and slotting

Data schema -

FIF widget group ID

PST

Audience ID

Slot

Priority

Position

Valid (To invalidate the config later)

The ingestion can be handled through a Presto table wrapper / CSV uploader (Presto table recommended)

Integrating the upload facility in admin panel for FIF widgets

Instrumentation -

Persist all the information persisted against FIF in bulk in silver presto tables

Metrics layout -

Northstar -

L0 -

Gold VC (Test vs control)

Gold OC (Test vs control)

L1 -

RE visibility effectiveness - % views for Gold out of net views for the feed (PST / CLP / Collection) per cohort

2 views here - First aggregated for all launched feeds and other broken down at a feed level

RE conversion effectiveness - Gold ObySe / O/Vi for the experimented feeds between test and control

Order cannibalisation spread - % of Orders coming to the experimented feeds that are now attributed to the new FIF RE (Just for test variants)

L2 / From long term tracking - NMV / Vi for the experimented feeds

Check metrics -

Overall O/Vi for the experimented feeds - Test vs control

Average feed scroll depth for the experimented feeds

Overall ObySe for experimented PSTs - Test vs control

Usability metrics -

Property CTR - Views → Interaction funnel for the FIF widgets wherever for the experimented PSTs (On the base of net traffic landing onto these feeds)

Property scroll depth → No of catalogs viewed on an average in FIF

Scaling metrics (To determine pockets where we can directly scale this) -

Cohort of users with significant interaction and conversion on FIF

PSTs where plat ObySe / O/Vi is not impacted for test users with introduction of FIF

Note - Test here refers to the 2 location variants where we showing FIF. We’d need to look at every cut broken down into Test1 and Test 2.

Tech solutioning doc -

https://meesho.atlassian.net/wiki/spaces/EW/pages/3237544019/Feed+in+Feed+experiment+Setup⁠

⁠

Reads: Ketan/Apoorv/Mahima

Want to print your doc?
This is not the way.

Try clicking the ··· in the right corner or using a keyboard shortcut (

CtrlP

) instead.