About change attribution in Ads traffic navigator

In complex serving data funnels, identifying the root cause of a performance shift is challenging. Our change attribution algorithm provides a scientific way to isolate the impact of each serving stage in Ads traffic navigator, helping you determine exactly where a drop or gain originated.

Jump to a section of the article:

• Review key metric symbols
• Core Logic: Isolating the impact to a final metric
• Understand the algorithm
• FAQ

Review key metric symbols

To understand the algorithm that follows, review these key metric symbols: 

• : The initial entry metric (for example, "Total ad requests").
• : The current stage metric being analyzed.
• : The metric immediately preceding the current stage.
• : The final funnel metric (for example, "Competing bids").
• : A specific rejection reason or error (for example, "Bidder error").
• : Refers to the previous period's value (for example, "Yesterday").

Core logic: Isolating the impact to a final metric

Let’s look at the "Ad requests allowing programmatic" view as an example. The final metric we care about is "Competing programmatic bids" (the C metric symbol). To decide if a serving stage or a rejection reason warrants a highlight in the interface, we ask:

• Did this specific change independently cause a 5% or greater shift in the final competing bids (C)?

By calculating expected metric values in "virtual states," we isolate one stage while holding all other factors constant based on the previous period (such as yesterday).  

For the Line items view, the final metric we care about is "Competing line items." The core logic is the same. 

Note: Examples in the remainder of this article use the "Programmatic" view and the "Yesterday" comparison period.

Understand the algorithm 

Impact of main stages

Every stage is connected to the next. The final outcome is simply the result of multiplying the starting volume by the performance of every step along the way. To understand how performance shifts, think of the funnel as a series of cascading rates, where each rate represents how much volume of a serving stage feeds into the next stage.

The cascading rates equation

The final result, competing bids (C), is determined by the initial volume  and a chain of conversion rates (r):

• : The rate at which initial requests move to the second stage.
• : The rate at which those progress to the third stage, and so on.

The "cascading rates" represent the efficiency of each stage (for example, the rate at which total programmatic bids become competing bids).

The impact equation

When the final results change (C vs. C’), it is because one or more of these cascading rates shifted. Every rate in the chain has a direct, proportional impact on the final result (C).

• If a specific stage's rate ratio drops by 10%, it means your final result (such as competing bids) also drops by 10% because of that stage alone.

By looking at the rate change ratio ()  we can see exactly how much each stage contributed to the final outcome:

Identifying the root cause

Instead of blaming a stage for a general drop in the final result, we isolate the efficiency of each specific stage (i).

1. Calculate the rate change: We isolate the specific multiplier for that stage i (for example, )
2. Determine the impact: We subtract 1 from that ratio to see the percentage shift:

   

3. The 5% threshold: If any single rate in the chain has shifted by 5% or more, it has independently moved your final results by the same percentage. We highlight these stages so you know where to focus your analysis.

Real-world example: Efficiency loss

If your "Bid requests sent" are exactly the same as yesterday, but your bid rate ("Total programmatic bids"/"Bid requests sent") drops from 10% to 9%:

• The rate change ratio is changed to 0.09 from 0.1 (a 10% drop).
• Assuming every other stage performs exactly the same, your final competing bids will drop by 10%.
• The interface highlights the "Total programmatic bids" metric in red because it is one of the primary drivers of revenue or volume loss.

Impact of rejection reasons (the E metric symbol)

The impact of a stage can be decomposed into the impact of its various rejections. The contribution that a rejection reason had on the stage’s total impact is:

In Ads traffic navigator, a rejection reason will be highlighted if its impact is more than 3%.

FAQ

Why is a metric highlighted even if its volume stays the same? 

If the preceding stage increased in volume but the current stage did not, the ratio has dropped. This indicates an efficiency loss that independently impacts the final result.

For the same reason, the highlighted trend could be opposite to the comparison from yesterday.

How does the algorithm handle the very first metric M_start? 

For the first metric (total ad requests), the impact is calculated as the simple change in volume:

Does the attribution account for the total drop? 

Yes. By multiplying the isolated impact of every stage from left to right, the values mathematically equal the total change in your final metric .

What does the tooltip mean when it says "independently shifted"?

It means the algorithm has mathematically removed the influence of other stages to show you what would have happened if only that specific metric had changed.

How can the impact of a rejection reason be greater than 100%?

If the volume of the rejection reason is greater than the volume of the next stage metric itself (that is, most of the previous stage volume is rejected), a single reason’s impact may be greater than 100%. Note that the rejection reason impacts are additive, so the net effect is their sum.