# AMC Audience Segment Analysis for DSP Targeting

Canonical page: https://www.kuudo.com/guides/amazon-ads-dsp-audience-segment-insights/

Use AMC audience segment analysis to compare off-Amazon converter rates, eligible reach and cost across DSP segments before choosing a targeting test.

Author: Thomas Spicer

## Summary

Compare each segment's share of eligible users who converted with the pooled rate from the same run. Shortlist segments with enough reach and an acceptable cost per converter, then use a controlled targeting test to measure the effect of the change.


Use an AMC audience segment analysis to shortlist Amazon demand-side platform (DSP) audiences for a targeting test. Compare each segment’s share of users who converted with the rate for the eligible pooled cohort. Check reach and cost per converter against your off-Amazon goal before choosing a segment.

A page-view pixel and a purchase event answer different questions about the same campaign in Amazon Marketing Cloud (AMC). Your team's chosen event, campaign scope and cost ceiling decide which segment is worth testing. Keep those choices with the AMC result. An authorized Amazon Ads MCP connection can help prepare a fresh analysis for that scope. [Amazon Ads MCP](/features/amazon-ads-mcp/) · [AMC](/features/amc/)

## The aggregate rate describes the users retained by the analysis

The example query starts with impressions from selected Amazon DSP campaigns. It keeps those impressions for users with no selected pixel event in the run. For event-bearing users, it keeps only impressions on or before the earliest selected event and within the configured lookback. If no impression qualifies, that user drops out of both the segment and aggregate rows. This can happen when the earliest event precedes all selected impressions.

The `Aggregated` row describes the users left after that filter. Its reach and cost can change when you change the pixel or lookback, because the impression filter treats users with and without a selected event differently. Use the aggregate from the same run and record how the filter defines the comparison population.

The example reads `conversions` by conversion time within the AMC run dates. Its 21-day lookback controls how far back from the first selected event an impression can qualify within those records. Waiting can give in-window events time to arrive. To follow a fixed exposure cohort beyond the run end, validate a query with separately bounded exposure and conversion dates. [Amazon off-Amazon conversion guidance](https://advertising.amazon.com/API/docs/en-us/guides/amazon-marketing-cloud/playbooks/off-amazon-conversions)

## Read the rate multiple alongside reach and cost

Converter rate divides flagged users by retained reach. Each flagged user counts once in that rate. The event total separately sums the selected pixel events for those users across the run dates, including repeat events. The query scopes events by pixel ID and conversion date. It pools impressions by selected campaigns and impression date, then joins the two by user. The rate describes observed behavior within that filtered population.

The instructional example uses an off-Amazon **page-view pixel** and a 21-day lookback. The table uses cost units because the example gives no currency.

| Metric | Pooled cohort | Example segment |
| --- | ---: | ---: |
| Eligible reach | 6,601,103 | 383,360 |
| Flagged converters | 68,967 | 5,177 |
| Converter rate | 1.04% | 1.35% |
| Rate multiple | 1.00× | 1.29× |
| Cost per converter | 4.12 | 3.12 |

Divide the rates before rounding to get **1.29×**. The segment has a higher observed rate and lower retained impression cost per flagged converter than the aggregate. It becomes a candidate if that reach meets your test floor and a page view is the outcome you want to improve. If the baseline rate is missing or zero, leave the multiple undefined. If the segment has zero converters, leave cost per converter undefined even if the export displays zero.

## Segment membership and overlapping reach shape the shortlist

The `dsp_impressions_by_user_segments` table records membership at the time of each impression row. The example joins those labels by user across the run dates. A label can therefore come from a different campaign or a later impression than the exposure retained for analysis. Verify the marketplace and segment ID. To identify targeting matches for the selected campaigns, tie the segment records to the selected impressions and check `behavior_segment_matched`. [Amazon DSP segment impression tables](https://advertising.amazon.com/API/docs/en-us/guides/amazon-marketing-cloud/datasources/dsp_impressions_by_segments)

One user can contribute reach, events and cost to several segment rows. Summing those rows counts overlapping users repeatedly. Use the aggregate row for the overall population and a dedicated overlap analysis for the unique reach of a group of candidates. Amazon's related **Conversions by audience segment with impressions** instructional query uses traffic-time attributed events and has its own measurement rules. [Amazon's related segment analysis](https://advertising.amazon.com/marketing-cloud/use-cases/fd48dc055ccf6ea343f8ff5af99cd937f4e723af3e1b743a22c9f9ea4672ea91)

Set a minimum reach based on the scale of the planned test. The example's reach flag only checks the selected cutoff. Look at converter counts as well, especially when a few additional converters could change the ranking.

## A shortlist gives you a targeting hypothesis to test

Start with an existing AMC export or scope a fresh analysis in your authorized AMC instance through the Amazon Ads MCP. Before running a new query, confirm available tables, segment IDs and the selected event. Review the query's campaign, date and membership joins against the question you want to answer.

Keep each segment’s ID, marketplace, event, campaign scope, run dates and lookback with its reach, converter count, rates and cost per converter. Use these rules to record a decision for each row.

| Check | Decision |
| --- | --- |
| Rate above the baseline, reach meets your floor and cost per converter fits your ceiling | Shortlist the segment for a test |
| Any of those checks falls short | Hold the segment and record why |
| Baseline missing, cost undefined or segment membership ambiguous | Resolve the measurement before ranking |
| Several candidates overlap | Review shared reach before sizing the test |

To estimate the effect of a targeting change, define the outcome, budget and evaluation period up front. Use randomized groups where the account supports the test and keep other campaign settings comparable. Amazon documents randomized holdout groups for DSP bid adjustment tests. Verify the experimental setup available for your targeting change. [Amazon DSP testing guidance](https://advertising.amazon.com/API/docs/en-us/guides/dsp/bid-adjustments/testing)

## What happens next: choose a segment for testing

Record one candidate, the comparison group, the selected event and the test owner. Decide how much evidence you need before changing targeting, then compare the groups on the same outcome over the test period.

*Check the event definition and DSP order association before running the analysis. [Measure off-Amazon conversions with Events Manager](/guides/amc-off-amazon-conversion-events-manager/)*



## Sources

- [Amazon DSP segment impression tables](https://advertising.amazon.com/API/docs/en-us/guides/amazon-marketing-cloud/datasources/dsp_impressions_by_segments)

- [Conversions by audience segment with impressions](https://advertising.amazon.com/marketing-cloud/use-cases/fd48dc055ccf6ea343f8ff5af99cd937f4e723af3e1b743a22c9f9ea4672ea91)

- [Off-Amazon conversions in Amazon Marketing Cloud](https://advertising.amazon.com/API/docs/en-us/guides/amazon-marketing-cloud/playbooks/off-amazon-conversions)

- [Amazon DSP customer holdout testing for bid adjustments](https://advertising.amazon.com/API/docs/en-us/guides/dsp/bid-adjustments/testing)