Feature request: Add support for Connect Outbound Campaigns

[Dilski]

Use case

Amazon Connect Outbound campaigns supports invoking a lambda function as part of a journey flow.

This sends a batch of customer profiles for processing

I would like support for this in powertools

Solution/User Experience

I love powerools, pydantic, and types - so would be great to have:

• event source data classes for the request and the response
• batch processer support (stretch goal, would love to have).

I am happy to be involved in implementing this.

Details on the lambda function invocation: https://docs.aws.amazon.com/connect/latest/adminguide/lambda-invoke-functions.html It looks like the customer data is a json string that would be good to parse - would be good to get a concrete json schema for this.

Alternative solutions

Acknowledgment

• This feature request meets Powertools for AWS Lambda (Python) Tenets
• Should this be considered in other Powertools for AWS Lambda languages? i.e. Java, TypeScript, and .NET

[dreamorosi]

Hi @Dilski, thank you for opening the feature request and for being a fan of Powertools.

For the data classes, I don't see any concerns. We already have some Amazon Connect-related data classes here, but if the one you're suggesting is not included, I think it makes sense to add it.

Would you be interested in contributing a pull request to add them? If yes, we'd be happy to help you get it merged.

---

For the Batch Processor, I'd like to understand better what you had in mind. The Powertools for AWS Batch Processor is built around the concept of partial batch responses upon failures.

This works because these services that trigger Lambda allow the function to return a response that includes the IDs of the items that failed, informing them to keep these failed items behind and make the others succeed.

I don't know much of how Amazon Connect calls Lambda in the context of outbound campaigns, so I'd like to understand if that integration supports a similar pattern.

[Dilski]

Howdy,
I'll open up a PR to add the dataclass once I have proper confirmation of the event shape (I suspect that the docs may be incorrect).

As for the batch processing, connect passes a batch of customer profiles to lambda.

{
  "InvocationMetadata": {
    "CampaignContext": {
      "CampaignId": "campaign-12345",
      "RunId": "run-67890",
      "ActionId": "activity-abc123",
      "CampaignName": "Welcome Campaign"
    }
  },
  "Items": {
    "CustomerProfiles": [
      {
        "ProfileId": "customer-001",
        "CustomerData": "{\"firstName\":\"John\",\"lastName\":\"Doe\",\"email\":\"john.doe@example.com\"}",
        "IdempotencyToken": "token-xyz789"
      },
      {
        "ProfileId": "customer-002",
        "CustomerData": "{\"firstName\":\"Jane\",\"lastName\":\"Smith\",\"email\":\"jane.smith@example.com\"}",
        "IdempotencyToken": "token-abc456"
      }
    ]
  }
}

The lambda should then process each profile in the batch (independently) and return an array of results (one for each element in the batch, with success or fail).

{
  "Items": {
    "CustomerProfiles": [
      {
        "Id": "customer-001",
        "Result": "SUCCESS",
        "ResultData": {
          "recommendedProduct": "Premium Plan",
          "score": 85,
          "nextAction": "send_email"
        }
      },
      {
        "Id": "customer-002",
        "Result": "FAILURE",
        "ResultData": {
          "error": "Invalid customer data",
          "errorCode": "VALIDATION_ERROR"
        }
      }
    ]
  }
}

I think this fits well with the powerools batch processor model (call a function for each element, catch exceptions, collect and report the result) - it might be too niche to be included in the powerools.

[dreamorosi]

Oh interesting, I didn't know, thank you for explaining!

I kind of agree now that conceptually speaking, it seems like a good fit. Regarding the demand, I think I'd love to give it some time and see if there're other customers interested in this kind of feature.

In the meanwhile, it's possible to extend the base class of the Batch Processor to create your on partial batch processor (as described here).

Either way, it could be interesting to see if the current base class is flexible enough to allow you to write a Connect-specific partial batch processor. If you do try and see some minor changes that we need to do in that regard, I think it'd be a much easier lift to implement before we go full on.

[Dilski]

@dreamorosi it looks like the individual item / partial batch failure mechanism has disappeared from the documentation - so I'm going to forget about that part for now ;)

I've got a branch for the data models now that I'll share soon. Before I do is there a way for us to get a concrete schema from the service team?

[dreamorosi]

Before I do is there a way for us to get a concrete schema from the service team?

If you have a way to test and log the payloads and build the schema from there, I genuinely believe it'll be faster.