Skip to content

Compliance gating - Launch reads


What was the problem?

Our platform’s success, including strategic initiatives, IPO readiness, and Mall performance, relies on our ability to combat unauthorised brand products and illegal items. Unfortunately, our existing approach heavily depended on post-upload manual and data-driven processes to identify non-compliant listings, falling short of addressing the issue at its source and leading up to latency related leakages.

How did we solve?

With Compliance Gating V1, we’ve introduced synchronous nudges to suppliers immediately after the catalog upload. These nudges take the form of warning and error messages that guide suppliers in real-time, discouraging the upload of non-compliant products. By presenting these error nudges upfront, we empower suppliers with a clear understanding of which products are not permitted on our platform.
Objective -
Reducing flagging latency for keyword cases during uploads to -
Reduce net non compliant view share
Increase seller awareness with self rectification leading to reduced agent bandwidth for reactive deactivation flows

How does the new flow look -
Screenshot 2023-08-18 at 5.06.58 PM.png
Synchronous nudges on error keywords added post upload
Warnings highlighted in orange - Allowed
Errors highlighted in red - Not allowed

Experiment reads -

~40% drop in non compliant view share (Statistically significant)
Post live deactivation down for test cohort by ~25k listings in ~4 weeks, led to ops bandwidth reduction by ~38%
~80% sellers able to comprehend and take required next steps in case of compliance flaggings
Listings run-rate drop of ~1.09% due to higher validation failure by ~17.2%
Drop higher in initial weeks of experiment due to incorrect config for some generic keywords in Suraksha
Post iterative correction, numbers stabilised at ~1% LbyS drop at ~7-8% higher validation failure
Current listings drop in line with higher deactivations happening for control post catalog go live
image.png
Detailed view
Metric
Test
Control
Delta
Significance
NS - Non compliant VS
0.0004%
0.00067%
40%
Yes
NS - Brand VS
0.00032%
0.00049%
36%
Yes
NS - Illegal VS
0.00029%
0.00053%
45%
Yes
Check - LbyS
37.7
38
-1.09%
Check - Validation failure
3.2%
2.74%
17.2%
Check - Drafts
4.81%
4.8%
0.3%
QCed PID count
40k
65k
38%
There are no rows in this table

Qualitative reads (Basis 110 respondents) -

Feature comprehension for inline warnings / error stood at
~72% for single uploads
~64% for bulk uploads - Comprehension a bit lower due to seller’s not understanding the next steps in bulk upload errors
Folks who comprehended the feature complied by removing -
Error(s) -
Removing keyword / field - ~81%
~7% left the listings in draft
~7% raised support ticket, remaining didn’t action
Warning -
Removing keyword / field - ~65%
~8% raised support ticket
~17% submitted w/o any action, remaining didn’t action
Post-Error actions of these sellers -
40% understood the next step to register brand
~38% not aware of the same and said would reach out to support
Takeaway / Next steps -
Next / Rectification steps not super clear to sellers and requires better error messaging in bulk uploads
Feature awareness comms to be sent over to sellers for accurately understanding the flow + next steps

Next Steps -

As we continue to refine Compliance Gating, our next steps are geared towards even greater accuracy and efficiency:
Context Flagging Enhancement: We’re implementing context flagging on top of our current regex to avoid false positives. This excludes cases like considering “Comfort” as a brand for descriptions such as “provides the best comfort.”
Brand Recognition Improvement: We’re working on supporting substring regex matches for non-generic keywords like “Adidas” to enhance our brand recognition capabilities.
Image-Based Gating: Real-time image-based gating will significantly reduce flagging latencies, bringing them down from around ~6 hours to nearly real-time. This will help reduce the daily incidence of ~10% non compliant views.

A huge shoutout to all the folks who have made this possible and led the project to success!



image.png
Want to print your doc?
This is not the way.
Try clicking the ··· in the right corner or using a keyboard shortcut (
CtrlP
) instead.