Hey everyone, I've been trying to wrap my head around a specific workflow and was wondering if anyone here has any real-world experience with it.
We're starting to use Salesforce Data Cloud (formerly CDP) more heavily, and my marketing team really wants to activate those audiences in The Trade Desk for programmatic campaigns. The idea is to move beyond just email/SMS and get those segments into the display/video bidding world.
I've been reading docs from both sides, and it *seems* possible, maybe through a cloud storage handoff (like Snowflake or BigQuery) or a direct integration partner? But the implementation path feels a bit fuzzy.
* What's the actual technical flow? Is it a scheduled export/import, or is there a more real-time method?
* Are there specific identity resolution hurdles to watch out for (like matching Salesforce IDs to TTD's identity graph)?
* Any major latency issues from the time the audience is updated in Salesforce to it being usable in TTD?
I'm grateful for any pointers or war stories. I'm a bit overwhelmed trying to map this out from theory to practice, and a screenshot of a failed setup is probably in my near future 😅. Even knowing which integration pattern is most common would be a huge help.
null
You've hit on the core architectural challenge. The direct, real-time API integration you might hope for doesn't exist. The flow is fundamentally a scheduled cloud storage handoff, and latency is its defining characteristic.
Your hunch about Snowflake or BigQuery is correct - that's the prescribed path. Salesforce Data Cloud exports segment membership tables to a linked cloud storage instance, which The Trade Desk then ingests on a polling schedule. This is usually a daily process. The major latency issue isn't just the transfer, but the subsequent identity resolution within TTD's own graph. You're looking at a minimum 24-48 hour delay from segment update in Salesforce to campaign activation.
The identity resolution hurdle is the most critical piece. You must map your PII (like email or phone) to a Unified ID 2.0 (UID2) or TTD's own IDL *before* the file lands in TTD's bucket. This hashing step is non-negotiable; you cannot send raw PII. Many teams use a middleware translation layer or a partner like LiveRamp for this key translation, as getting the privacy mechanics wrong here will break the entire pipeline.
— Harper
Oh, that's such a timely question. I'm currently staring down the barrel of this exact task and it feels more complicated than I expected.
Following up on the cloud storage handoff point, how are you setting up the actual export from Salesforce? I've seen references to using "Audience Studio" for the export job, but then getting it into the right file format and location for TTD to pick up seems like another whole step. Are you handling that with something like a scheduled Airflow job, or is there a simpler config I'm missing?
And about the latency issue... that 24-48 hour window you mentioned, does that effectively mean you can't do anything remotely "real-time" with this pipeline? Like, a segment based on website abandon cart activity wouldn't be usable for a retargeting campaign for two days? That seems tough to sell to marketing.
You're right to focus on the export mechanics, because that's where the operational complexity lives. The term "Audience Studio" is a red herring, that's the old DMP branding. In Data Cloud, you configure a "Segment Export" job to a connected cloud storage location (GCS, S3, ADLS). This doesn't produce a TTD-ready file. It dumps a table with internal keys and segment membership flags.
The missing piece is the translation layer. You absolutely need an orchestration job (Airflow, a Lambda, etc.) to:
1. Ingest that export.
2. Map the internal keys to a PII identifier like hashed email.
3. Format it to TTD's exact spec (CSV, naming convention, no headers).
4. Place it in the specific TTD-managed cloud bucket they provision for you.
There is no simpler config, this is custom pipeline work.
Regarding real-time use, you've understood the limitation perfectly. A cart abandon segment via this pipeline is useless for timely retargeting. The 24-48 hour window is a floor, not a ceiling; any misconfiguration or backlog in the pipeline or TTD's graph resolution pushes it further. Marketing should consider this pipeline only for large, persistent audience segments like "loyalty program members," not ephemeral behavioral events.
Trust but verify.