Flow Collection Processing

Flow Collection Processing

Use this skill when a Flow must do more than retrieve a flat list — when it needs to iterate, filter, sort, remap, or write back a group of records as a unit. This skill covers the full suite of Flow collection elements: Loop, Assignment inside a loop, Collection Filter, Collection Sort, and Transform.

The core design choice: declarative collection elements (Filter/Sort/Transform) vs Loop + Assignment. The declarative elements are clearer, fewer elements, easier to maintain, and less error-prone — but they don't handle every scenario. Knowing when each pattern fits is what this skill teaches.

Before Starting

Gather this context before working on anything in this domain:

• What SObject or data type is in the collection? SObject Collections and primitive collections (Text, Number, etc.) behave differently and support different elements.
• Is the goal to read-and-branch, filter-and-keep, remap fields to a different SObject type, or accumulate records for a single DML write?
• Is the Flow running in a record-triggered context (bulk-safe, up to 200 records per transaction) or an autolaunched context called directly (potentially single-record)?
• Is the downstream consumer a DML (needs a collection) or a Screen Flow table (needs sorted collection) or a subflow (needs a specific contract)?

Core Concepts

SObject Collection vs. Primitive Collection

Flow supports two collection types. An SObject Collection stores a list of full records (e.g., Account[]). A primitive collection stores a list of scalar values such as Text or Number. Most collection operations — Loop, Collection Filter, Collection Sort, and Transform — apply only to SObject Collections. Primitive collections can be looped over and appended to via Assignment, but they are not compatible with the newer declarative filter/sort/transform elements.

Loop Element

The Loop element iterates over a collection one record at a time, exposing a "current item" variable for that iteration. Inside the loop, Assignment elements can read from and write to the current item, accumulate records into a separate output collection, or set flags for downstream branching. A Loop always has two exit paths: "For Each" (for each iteration) and "After Last" (when the collection is exhausted). Building up a result collection inside a loop by appending the current item to a separate collection variable is the foundational collection-building pattern in Flow.

Collection Filter Element

Available from Spring '23 onward, the Collection Filter element removes records from a collection based on one or more conditions — entirely without a Loop. It accepts an SObject Collection as input and produces a filtered SObject Collection as output. It supports AND/OR condition logic and field-level comparisons against literal values or other Flow variables. This element should be the first choice whenever the goal is simply reducing a collection to a subset; using a Loop with a conditional assignment to achieve the same result is more verbose and harder to maintain.

Collection Sort Element

The Collection Sort element reorders an SObject Collection by one or more fields, ascending or descending. It operates in place on the input collection and does not require a Loop. Sorting before passing a collection to a Screen Flow table or to a downstream subflow is a common use case.

Transform Element

The Transform element maps fields from one SObject Collection to a different SObject Collection or to a primitive collection. A common use case is producing a list of Task or Case records from a list of Lead records, mapping field values declaratively. Transform is the correct tool when the goal is type conversion or field remapping across SObject types; doing this inside a Loop with manual field assignments is more error-prone.

DML on Collections

Update Records, Create Records, and Delete Records all accept either a single record or a collection variable. Using a collection variable with a single DML element processes all records in one operation and consumes one DML statement and one DML row per record — exactly the same limits as a list DML in Apex. Placing a DML element inside a Loop converts that cost into N DML statements, which fails at scale.

Common Patterns

Pattern 1: Loop-and-Accumulate to Build a Modified Collection

When to use: You need to modify a field on every record in a collection (e.g., set Status__c to "Processed") and then write the whole list back.

Structure:

1. Declare a second SObject Collection variable (the output collection, initially empty).
2. Add a Loop over the source collection.
3. Inside the loop, use an Assignment element to set the field on {!Loop.currentItem}, then add {!Loop.currentItem} to the output collection using the Add operator.
4. After the loop exits via "After Last", connect to a single Update Records element pointed at the output collection.

Why not the alternative: Placing Update Records inside the loop issues one DML statement per record and fails when the transaction processes more than the DML statement limit allows.

Pattern 2: Collection Filter for Subset Selection

When to use: You need to pass a subset of a retrieved collection to a downstream element — for example, only the Opportunities with StageName = 'Closed Won' from a larger query result.

Structure:

1. Add a Collection Filter element after Get Records.
2. Set the source collection to the query result variable.
3. Define filter conditions (field, operator, value).
4. Store the result in a new SObject Collection variable.
5. Pass that filtered collection to the next element.

Why not the alternative: A Loop with an if/then branch and conditional accumulation achieves the same result but takes four to six elements instead of one and is harder to read during review.

Pattern 3: Transform to Produce a Related-Record Collection

When to use: You have a collection of parent records and need to create child records — for example, creating follow-up Task records from a collection of Case records.

Structure:

1. Add a Transform element after the source collection is available.
2. Set the source collection (e.g., {!CaseCollection}) and the target SObject type (Task).
3. Map fields: WhatId from Case.Id, Subject from a template or literal, etc.
4. The output is a Task Collection variable.
5. Pass that collection to a single Create Records element.

Why not the alternative: Building the Task collection inside a Loop with manual Assignment and Add operations is verbose, difficult to maintain, and does not communicate intent as clearly as a dedicated Transform element.

Pattern 4: Chained Declarative Pipeline

When to use: The collection needs multiple operations — filter, then sort, then transform — all of which have declarative elements.

Structure:

[Get Records] → [Collection Filter: Status = 'Open']
              → [Collection Sort: by CreatedDate DESC]
              → [Transform: map to Notification SObject]
              → [Create Records: bulk insert notifications]

Why not the alternative: Four declarative elements in a row is faster to author, faster to review, and less error-prone than a single Loop with conditional logic + sort + mapping.

Pattern 5: Empty-Collection Safety

When to use: Any collection pipeline — empty collections are common and must not fault.

Structure:

• Put a Decision element BEFORE expensive downstream work checking COUNT({!collection}) > 0.
• If empty, exit early with a Process Observation (no work to do).
• If non-empty, proceed.

Why not the alternative: A Create Records on an empty collection does nothing silently — not a fault, but also not an observable "job completed" signal. Downstream notifications fire on zero records, confusing users.

Decision Guidance

Well-Architected Pillar Mapping

• Performance — using Collection Filter/Sort/Transform instead of Loop saves Flow CPU time (fewer elements executed).
• Scalability — DML-inside-Loop is the canonical scalability failure in this domain. Eliminating it is the primary Scalability win.
• Reliability — empty-collection safety (Pattern 5) prevents silent no-op failures that confuse users.

Review Checklist

• No Create Records, Update Records, or Delete Records element sits inside a Loop without an explicit justified exception documented in the flow description.
• Collection Filter is used instead of a Loop-with-conditional-accumulation wherever the only goal is subset selection.
• Collection Sort is used instead of a sorting loop for ordering.
• Transform is used instead of a manual-assignment loop for SObject type conversion.
• SObject Collection variables have a defined SObject type — untyped collections cannot be used with Filter, Sort, or Transform.
• The "After Last" path of every Loop reaches the next meaningful element; unreachable paths cause silent flow termination.
• DML elements reference collection variables (not single-record variables) when operating on multiple records.
• Flow tested with an empty collection input to confirm it does not fault on the Loop's "After Last" path.
• Pattern 5 empty-collection gate in place for any pipeline that can process zero records.

Recommended Workflow

Step-by-step instructions for an AI agent or practitioner activating this skill:

1. Gather context — confirm the org edition, relevant objects, and current configuration state
2. Review official sources — check the references in this skill's well-architected.md before making changes
3. Implement or advise — apply the patterns from Common Patterns above; prefer declarative over Loop where applicable
4. Validate — run the skill's checker script and verify against the Review Checklist above
5. Document — record any deviations from standard patterns and update the template if needed

Salesforce-Specific Gotchas

1. Collection Filter and Collection Sort require typed SObject Collections — if the collection variable does not have a specific SObject type set, the Filter and Sort elements cannot reference its fields and the configuration UI will be incomplete.
2. Adding the current item to a new collection inside a loop mutates the shared object reference — if you modify a field on {!Loop.currentItem} and then add it to the output collection, the modification is captured. But if you forget to modify before adding, you accumulate the original unmodified values silently.
3. Transform does not support formula expressions in its field mappings — you can map a source field to a target field, or a literal value, but you cannot write a formula inline. If logic is needed during the mapping, a Loop-based approach with an Assignment formula is required instead.
4. An empty collection passed to a Loop does not fault — it exits immediately via "After Last" — this is safe, but downstream elements must handle the resulting empty output collection correctly.
5. Collection Sort is stable but sorts the original collection variable in place — there is no copy; the source variable is modified. If you need both the original and the sorted order, copy the collection into a second variable before sorting.
6. Collection Filter's condition on a formula field may behave differently from the same condition in SOQL — Flow evaluates formula fields based on the record snapshot, not re-evaluating against live data. Test edge cases.
7. The collection element count limit is 50,000 rows per Flow interview — not 10,000 DML rows, but 50,000 in-memory records across all collections. A Get Records returning 50k+ will fault.
8. Transform can't write to system fields — CreatedDate, LastModifiedDate, SystemModstamp, Id (when creating new records). Omit them from the mapping.
9. Sorting by a formula field works but is slow — each comparison evaluates the formula. Materialize the sort key to a real field if sort performance matters.
10. Collection Filter inside a subflow may not see the parent's filter context — subflow input variables are copied, not referenced. Filter criteria must be self-contained.

Proactive Triggers

Surface these WITHOUT being asked:

• DML inside a Loop → Flag as Critical. Canonical scalability bug; refactor to Pattern 1 or 4.
• Loop used for subset filtering → Flag as High. Replace with Collection Filter (Pattern 2).
• Manual Assignment loop doing SObject type conversion → Flag as High. Replace with Transform (Pattern 3).
• Empty-collection path with no gate → Flag as Medium. Pattern 5 prevents silent no-ops.
• Untyped SObject Collection variable → Flag as High. Filter/Sort/Transform won't function correctly.
• Collection Sort on a formula field without performance test → Flag as Low. Slow at scale.
• Chained Loops processing the same collection → Flag as Medium. Can usually collapse via Pattern 4.
• Collection size approaching 50k rows → Flag as High. Approaching hard limit; move work to Batch Apex.

Output Artifacts

Related Skills

• flow/flow-bulkification — when governor-limit safety under volume is the primary concern.
• flow/record-triggered-flow-patterns — when the question is trigger event / entry criteria / save behavior.
• flow/subflows-and-reusability — when collection-processing should extract into a reusable subflow.
• flow/fault-handling — when collection pipelines need fault-routing on DML failures.