Explore

Auto CPC algo

Overview:

Current State Implementation:

⁠

⁠

Real time bidding is currently live only in PDP reco feed

All other REs are live on dynamic bidding framework

We will migrate all the REs to the RTB flow in this R2R cycle

Including Rev/View Aspect in Target ROI framework

For every sscat, compute the required rev/view, basis historic rev/view values

Apply adjustment factors on the sscat level rev/view based on

ECP or OC-CP of the sscat in the last 7 days

BU% of the sscat

Arrive at a rev/view value for each catalog belonging in the sscat with different adjustment factors

Catalog goodness - CTR and ObyC

T.Rev/view = f (Historical Rev/view, ECP, BU%, catalog goodness)

Compute the average target CPC of a catalog from the above T.Rev/view value:

T.Rev/view = avg_Ad_CTR x Target_CPC

T.CPC = T.Rev/View /avg_Ad_CTR

Using the target CPC value in computation of Target ROI for auto CPC framework

T.ROI = avg_Ads_ObyC x AOV / T.CPC

T.ROI = avg_Ads_ObyC x AOV / (T.Rev/View /avg_Ad_CTR)

T.ROI = avg_Ads_ObyC x avg_Ad_CTR x AOV / T.Rev/View

This should be the target ROI committed to the supplier for onboarding them in ROI-A or CRA programs

Using this Target ROI in the computation of CPC (real time bidding flow)

CPC = PCVR x AoV / T.ROI

CPC = PCVR x AOV / (avg_Ads_ObyC x avg_ad_CTR x AOV / T.Rev/View)

CPC = PCVR x AOV * T.Rev/View / avg_Ads_ObyC x avg_ad_CTR x AOV

CPC = T. Rev/View x (PCVR / avg_Ads_ObyC) x (1/ avg_ad_CTR )

Extrapolating this value of CPC to a world where ad catalogs are ranked based on rev/view:

Ranking score =a * Goodness_score + b* CPC * pCTR

The rev/view component of ranking will look as follows:

b* (T. Rev/View x (PCVR / avg_Ads_ObyC) x (pCTR/ avg_ad_CTR ))

Accuracy in computing the target Rev/view is of paramount importance

Target Rev/view guardrails across different Campaign Types

Include an adjustment factor on the rev/view computation basis the campaign type of the catalog for the following reasons:

To bridge any disparities between auto CPC and ROI A catalogs -

Guardrails like, the ROI A catalogs can have only upto “x%” lower rev/view than non-ROI A catalogs and should not get any lower

To migrate budget from fixed catalogs into auto CPC and ROI A programs

The rev/view targets for the same catalogs in fixed CPC campaign should be higher by “y%” than the same catalog in Auto CPC campaign, which in turn will be “z%” higher than the rev/view target of the catalog in ROI A campaigns

T.Rev/view = f (Historical Rev/view, ECP, BU%, catalog goodness, campaign_type)

Using the same Ranking and Charging CPC:

Once we migrate to inclusion of rev/view based Target ROI computation in the CPC framework, there need not be any difference between the ranking and charging CPCs, and the dependency on discounting framework to deliver the promised ROI will become minimal to negligible

Fixed CPC campaigns: Future State

If fixed CPC campaigns will continue to exist as of today, then migration to a fixed CPC at a catalog level is imperative

The T.Rev/view, and hence the CPC would be computed at a catalog level

This will be a significant tech effort in migrating from base CPC at a sscat level to min CPC at a catalog level

Sequence of Events and Timelines:

Task

ETA for Experiment

ETA for scaleup

Primary Owner

Extending RTB to all REs:

CLP/Collections

Search: Nov 14

CLP/collections: Dec 14

FY: Jan

Search: Dec 16th

CLP/Collections: Jan 14th

FY: Feb 14th

Experimenting without Smart Spends framework

Nov 14

Dec 9th

Product / DS

Computing target Rev/View framework for all catalogs

Dec 1

Dec 31

Product

Rev/view based Ranking:

Reco

CLP/collections

Reco: Nov 14

Other REs: TBD

DS / Product

Fixed CPC at a catalog level

TBD

Product / Tech

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.