Why does a single user with the same CRM ID (cgrid) see both control and variant experiences in Adobe Target?

[ShoebSh]

Hi everyone,

I wanted to clarify a scenario related to cross-device/profile merge behavior in Adobe Target, specifically in an A/B test setup using A4T and CRM ID (cgrid) for identity stitching.

Here’s the scenario I tested:

On Desktop A, I logged in as a user (with CRM ID = xyz) and received the Variant experience. The ECID was 123.

Then, on Desktop B, I visited the site anonymously (not logged in) and was assigned to the Control experience with ECID = 456.

After browsing as anonymous on Desktop B, I logged in with the same CRM ID xyz.

Adobe Target then switched the experience to Variant (aligning with the login history of Desktop A), but the ECID remained as 456 on Desktop B.

This effectively means that the same user (CRM ID = xyz):

Saw Control when anonymous (Desktop B)

Saw Variant after login (Desktop B)

Already had Variant history from another device (Desktop A)

My Questions:

Is this expected behavior in Adobe Target?

Can this cause contamination in A/B test results, since one user is technically exposed to both experiences before Target reassigns based on CRM ID?

In reporting (especially A4T via Adobe Analytics), how should we handle this — considering the ECID does not change post-login?

Does this behavior contribute to Sample Ratio Mismatch (SRM) or inflated visitor/order counts across experiences?

[GigiCotruta]

Hi, @ShoebSh,

1. Is this expected behavior in Adobe Target? -> Yes, it is an expected behavior. The user got qualified for the Variant, so it will display for this user as soon as this happens. 
2. No. When the user is shown the variant, all his metrics will go to the variant reporting. All the data before the Variant, will go to the Control reporting, so no contamination in here.
3. In AA you should measure by activity and experience hits. I don't know your exact implementation, but in my case using WEB SDK, we are measuring by these parameters: 
   
4. As mentioned in the 2nd point, there should be no contaminated data, thus, no inflated numbers.

[ShoebSh]

Thanks for your response.

Just to add our specific case:

We’re running a test where Adobe Target is implemented server-side (not using Web SDK), and we pass a third-party customer ID (CRM ID or cgrid) for identification across devices.

In this test:

We set up a 50/50 split between Control and Variant in Target.

But when analyzing results in Adobe Analytics (A4T), we noticed:

The total sum of unique visitors across Control + Variant exceeds 100% of total unique visitors for the activity.

The split is not balanced — for example, Control might show 50.7% and Variant 50.3%.

To investigate further, we broke down the data by ECID and Target Experience, and found that:

Some ECIDs appeared in both Control and Variant, suggesting that a single user may have qualified for both experiences under different ECIDs.

This likely occurred when a user visited anonymously first (assigned Control), then logged in (CRM ID stitching kicked in), and Target reassigned them to Variant based on prior experience on another device or session.

We’re now exploring whether this cross-experience exposure is skewing test results — especially because:

Users who qualified for both experiences had a much higher conversion rate

After excluding these users, the lift flipped from positive to negative

So while we understand that Target handles alignment via profile merge, in our server-side implementation, the reporting in Analytics still reflects overlap, unless segmented properly.

Appreciate your insights and sharing best practices — it’s been helpful as we debug this more deeply!