Bulbul · Shopee advertising

Step-Down Tracker

loading…

Are we moving along the curve, or has the curve moved?

Each dot is one day. The line is the diminishing-returns curve fitted on the frozen baseline window. Days sitting on the line behave exactly as that spend level always has — a healthy step-down. Days sitting below it mean the same spend now buys less than it used to, which is what ranking decay looks like.

Baseline day (before step-down) Since step-down Fitted baseline curve Break-even ROAS

Efficiency index, 7-day rolling

Actual revenue divided by what the baseline curve predicts at that day's actual spend. 1.00 means the spend performed exactly as it historically did. Sustained readings inside the shaded zones are the decay warning.

Are we actually executing the step-down?

Daily gross Shopee ad spend against the planned target for each phase. The tracker is only meaningful if spend genuinely lands where the plan says.

Actual daily spend (gross, incl GST) Phase target (expected realised)

How to read this

Falling ROAS is not the alarm

Cutting spend should raise revenue per ad dollar — that is diminishing returns running backwards, and it is the entire point of the exercise. Raw ROAS moving tells you nothing on its own.

The index is the alarm

It asks a narrower question: at the spend you actually ran, did you get the revenue that spend level has always produced? That isolates decay from the deliberate cut.

One bad reading is not decay

—

What decay would justify

A sustained run below the decay line means the cut is costing more than the model predicts, and the step-down should stop at the current level rather than continue down.

Method & limits

—

The baseline curve is frozen deliberately. If it re-fitted as new days arrived, post-step-down data would be absorbed into the baseline and the index would drift back toward 1.00 no matter what actually happened — the tracker would go quietly blind. The build script refuses to fit on any day at or after the step-down start.
Spend here is gross — raw Shopee expense plus 9% GST, before any ROAS-protection rebate. That is the basis the curve was fitted on, and it stops a rebate (a lagging refund) from flattering the single day it lands on. The net-of-rebate figure the P&L uses is shown separately in the table.
This is observational, not an experiment. The bidder spends more on high-demand days, so some of the spend–revenue relationship is demand, not causation. That bias inflates apparent ad efficiency, meaning the true marginal return is likely worse than modelled — the error runs in the conservative direction for the decision to cut.
Interpolation only. —
It cannot see why. A drop in the index is consistent with ranking decay, but also with a competitor campaign, a stockout, or a seasonal shift. The index tells you that something moved, not what.

Data table — every day in the tracker

Date	Phase	Spend (gross)	Spend (P&L)	Revenue	Predicted	Index	Index 7d	ROAS	Net	Orders