Explore

WS

Meesho Context & Learnings - What is happening & what can guide us to become better

Context & Objective

Context

Strategic initiatives, IPO readiness and Mall, are heavily dependent on the platform’s capability to arrest Fakes (Unauthorised Brand products) and Other Illegal PIDs (Sexual, Pills/ Drugs, etc.)

Currently, the identification methodology of these non-compliant listings is indexed on DS + Manual Ops processes after Suppliers upload catalogs. DS interventions have been implemented in the past 6 months with the primary of reduction of QC manpower as part of the overall Capital Efficiency goal

Objective

The document intends to build a lens on Cataloging Compliance to

Identify, measure & track Compliance metrics indexed towards platform Compliance

Prioritise & align solutions for maximum impact on the metrics

Current Compliance Process Brief

Currently, 2 parallel processes run post Catalog upload by Supplier before Catalog Go Live on Platform

Tech related checks (Not in scope of this document) - Checks like Same supplier duplicate or not, Visibility assignment for CT Views, Wrong Category mapping, Image related errors, etc.

Compliance checks (Scope of this Document) - Checks where DS models are run on the entire PID inflow, post which DS model flags non-compliant PIDs. After the DS model output, the Ops team performs manual check on i) All flagged PIDs & ii) Holdout group of all PIDs passed by DS

Compliance checks can be further divided into the following

Proactive Brand Infringement Process Checks (DS + Manual) post Catalog Upload

Proactive Other Illegal Classes (Sexual, Pills/ Drugs, etc.) Process Checks (DS + Manual) post Catalog Upload

Reactive deactivation of PIDs post Escalation (TAT < 1 hr post escalation)

Details of all Compliance Checks can be checked

here⁠

⁠

Learnings from VoS, Outside In & Meesho Data Analysis

VoS Summary

Lack of Awareness & Clarity - For both Brand Infringement and other Illegal classes, Suppliers have called out the lack of awareness/ understanding about what can be sold & what cannot be sold

~40% of Suppliers mentioned awareness as an issue quoting “Meesho blocks all catalogs repeatedly, causing lots of loss to business”

Suppliers do not understand the reason for deactivation indicated by Low open rates of emails for deactivation (for both Brand infringement & Other Illegal Classes)

Absence of Consequence & Punitive action - While Suppliers mentioned that they were aware about the deactivations of the non-compliant listings, that was not enough to stop them from uploading non-compliant listings in the future due to absence of any consequence/ punitive action

Meesho Data Analysis Summary

95% Fakes contributed by 10k Suppliers - Around ~3k Suppliers contribute to ~84% of Brand Fake deactivation on platform, the next 7k Suppliers contribute to around 11% of Brand Fake deactivation on platform

93% Other Illegal Listings contributed by 1k Suppliers - Top 1k suppliers (20% of all sellers who upload illegal listing) contribute to around 93% of all illegal listing on the platform

Outside In Summary

Strong Brand Gating Mechanism at Time of Upload - FK & AZ have a strong brand gating mechanism at the time of Catalog upload with nudges/ prompts in the Catalog upload journey to provide documentation

Continuous Scanning of All Live Listings - As part of Amazon Project Zero, AZ performs a detailed scanning of all platform listings through automated bots running a check on Image, Logo, Name, attribute to check for counterfeits, etc.

Self Serve Removal for Brands - AZ also has a Self Serve tool provided to the Brands with the ability to report counterfeits and hence deactivate any Brand Fakes reported

Learnings Define Gaps & Guide Us Towards The End State

Gaps in Current Compliance Process at Meesho

Missing Output metric to drive initiatives - The metrics around Compliance (Precision & Recall of individual DS models) are all input oriented without any North Star guiding the initiatives

Lack of checks/ gating at Catalog Upload Stage - There are no checks/ gating mechanism at the time of Catalog Upload. All the checks in current process happen post the DS output is provided which happens with a TAT of 5-8 hours post Catalog Upload

Supplier Awareness on Platform Compliance at the time of Upload - Current communication to Suppliers happen post PID deactivation through E-mails, WhatsApp, etc. which suffer through low CTRs and hence fail to build awareness

Lack of Consequences for Repeat Offenders - There are currently no punitive actions for the repeat offenders who keep submitting non-compliant listings

End State View of Compliance At Meesho

End State view of Compliance can be summarised as follows:

Non-compliant listing identification capability (pre-Go Live)

Ability of non-compliant listing identification would be hinged on Keywords used & Images Uploaded at time of upload

Keyword based flagging of non-compliant listings would be close to ~100% for exact matches

Image based flagging of non-compliant listings would be at 90%+ accuracy for all DS models

Non-compliant listing identification capability (pre-Go Live)

This would be based on 2 sub-streams

Automated DS model based scanning of entire active listing base frequently to flag non-compliant listings

Flagging of non-compliant listings based on input from internal stakeholders (Mall, Legal, etc.) & external stakeholders (Brand PoCs)

Feedback loop to Pre-Go Live models based on flagged listings to improve accuracy further and keep models relevant to catch hold of newer ways of Supplier bypassing

Supplier Behaviour Shaping

Supplier Behaviour Shaping would be based on

Identification of Suppliers would be done in 3 categories based on their past behaviour (frequency of deactivated listings, % of deactivations of overall listings uploaded) - (i) Normal Suppliers; (ii) Potential Fraud Suppliers; (iii) Fraud Suppliers

Shaping behaviour for Normal Suppliers to make corrections at time of listing who might be made aware of non-compliant listings & correct their listing based on Warnings/ nudges on panel

Shaping behaviour for Potential Fraud Suppliers to stop turning into Fraud Suppliers by proactive messaging on panel to socialise the punitive actions that could be levied if the behaviour persists & place softer punitive actions like throttling of CT views, etc.

Shaping behaviour for Fraud Suppliers to stop newer uploads among other harder punitive actions like stopping Catalog Edit, Monetary penalties, etc.; suppliers can resume uploading new catalogs based on submission of an undertaking of not uploading non-compliant listings in the future

⁠

image.png

⁠

North Star Measurement, Target & Roadmap

Defining North Star

Guiding Principle for North Star

North Star should be able to quantify the overall Compliance risk of platform at a given point in time - taking into account the risk

North Star should be indexed towards checking compliance for the Suraksha list (currently containing 1350+ brands & 500+ illegal classes)

North Star & Its Calculation

View Share of Non-Compliant Listings*

* Proxy for the non-compliant listings would be calculated through a newly setup process of DoD manual scanning on pareto RE

Pareto RE are

1. Search (contributing 55% of views of leaked non-compliant listings) - manual scanning & removal of non-compliant listings from top 100 catalogs in descending order of views for the previous day for each of the Brands/ Illegal classes (keyword searched) of the Suraksha list

2. FY Feed (contributing 25% of views of leaked non-compliant listings) - manual scanning & removal of non-compliant listings from top 1000 catalogs in descending order of views for the previous day (taking top 1000 to remove personalisation bias instead of 100)

3. PDP Reco (contributing 14% of views of leaked non-compliant listings) - manual scanning & removal of non-compliant listings from top 1000 catalogs in descending order of views for the previous day (taking top 1000 to remove personalisation bias instead of 100)

Aditional details on calculation are

here⁠

⁠

North Star Target Taken With Prioritisation Rationale

NS Target → By April End, drive North Star from 1.06% (Feb 15th) to 0.22%

The NS Target achievement would be done through the V1 Scope which is indexed on

Keyword based non-compliant listing identification - prioritisation of keyword based since most of the visibility is gathered through Search & also due to the fact that any brand or government authority have typically used Search to discover non-compliant listings

Shaping Supply Behaviour - Identifying repeat offenders & penalising them; identifying signals for potential offenders & warning them of upcoming penalties

Scanning on Platform - Scanning & removal of non-compliant PIDs on platform - through manual QC for Pareto RE & entire platform through DS models

The Goal given by Legal team of having 0 Fakes in Top 60 & upto 3 Fakes in Top 100 would also be subsumed in the KRs that address this North Star

Detailed views of all KRs of V1 Scope & other details are provided below

V1 Scope & Roadmap

V1 Scope consists of initiatives to drive North Star (View share of non-compliant listing) from current 1.06% to 0.22%

⁠

image.png

⁠

Summary of all V1 KRs

Table 67

Table 67

KRs

Scope of KR

NS Target

KR1: Compliance Gating (Keyword based) at Catalog Upload Stage

Arresting non-compliant listings based on keyword & creating supplier awareness

0.14% Viewshare Drop

KR2: Repeat Offender: Blocking + QC

Identifying Suppliers based on past offences & also suppliers who could potentially become offenders & taking punitive action against them

0.09% Viewshare Drop

KR3: DoD Search Cleanup (part of Daily NS calculation & deactivation)

For 0 Fakes in Top 60 & upto 3 Fakes in Top 100

Indexed towards pareto Real estates of DoD tracking & clean up of non-compliant listings

This would target the Legal relevant KR of having 0 Fakes in top 60 & upto 3 in Top 100

0.18% Viewshare Drop

KR4: Monthly Cleanup Based on DS Scanning of All Live listings

First scan (Jan), missed the brand attribute scanning

Second scan (Feb) has the brand attribute enhancement

Large goodness expected due to the addition of brand attribute in the scan

0.43% Viewshare Drop

There are no rows in this table

⁠

Beyond V1

V2 Scope

Principles of V2 are

Increasing the scope of identification of non-Compliant listings through images at time of upload & through external stakeholder

Increasing the intensity of punitive actions for Suppliers

High level Scope

Compliance gating (Image based) at Catalog upload Stage - Catalog Upload

Self serve tool for internal stakeholders like Mall & Legal teams & external stakeholders like Brands to flag non-compliant listings on platform

Installing Monetary penalties for Suppliers that are repeat offenders

V3 Scope

Principles of V3 are

Adding a feedback loop for flagging non-compliant listings

Adding punitive actions for Suppliers based on their past offences during the Go Live stage

High level Scope

Self serve tool for flagging non-compliant listings → first for internal Legal & Mall team owners; then to be scaled to external brands

Adding platform level sanctions like blocking of CT views of all listings for Repeat offenders among other benefits/ features for suppliers

Alignment Required

Alignment Required

Alignment Needed

Details

Alignment Gathered

Upvote

Downvote

North Star

Definition & calculation methodology

End State

Overall alignment on the End State envisaged for the workstream

Roadmap & Target

Alignment on overall target of moving North Star from 1.06% to 0.2% by Apr End

There are no rows in this table

⁠

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

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

CtrlP

) instead.