Not every request against your Azure estate needs a pipeline, and not every request should skip one. The two look similar from the requester's side — fill out a form, get it approved — but what happens after approval is genuinely different, and picking the wrong model for a given change either wastes time or skips process it actually needed.
Batch: for anything that provisions infrastructure
Standing up a new resource — a VNet, a SQL database, a Databricks workspace, a Key Vault — is a provisioning operation. It has a plan step, a set of resources Terraform needs to reconcile, and a result worth reviewing before it's final. This is minutes, not seconds, and that's appropriate: you want a Terraform plan to actually run, not a resource to appear before anyone could sanity- check what was about to be created.
Symetriz's Manual Batch Deployment flow is built for exactly this: a catalog item maps to a Terraform blueprint, an approved request triggers a real pipeline, and the ticket tracks that pipeline through to completion.
Real-time: for operating what already exists
Starting an AKS cluster, triggering an already-defined ADF pipeline run, toggling a Databricks cluster on or off, updating one NSG rule, adding someone to an Entra group — none of these create new infrastructure. They're operations against resources that already exist, and they're the kind of thing that's actively annoying to wait minutes for. If someone needs a cluster running to unblock a demo in five minutes, a Terraform plan/apply cycle is the wrong tool even though it would technically work.
Symetriz's Real-Time Deployment flow uses the same create-ticket-then-approve experience as batch, but on approval it calls the Azure Management SDK directly and reports back synchronously — generally under fifteen seconds. No pipeline to track, because there's no plan to reconcile: the operation either succeeded or it didn't, and you know which before the approval screen even closes.
The deciding question
When deciding which model a new catalog item belongs in, the question isn't "is this important" — both paths carry the same approval gate and audit trail. The question is: does this change what exists, or does it change the state of something that already exists? Creation goes through batch. Operation goes through real-time. Getting this split right is most of what makes a deployment platform feel fast where it should and careful where it should be.