Author NULL-safe, type-correct, performance-aware Formula resources and condition expressions in Flow: lazy re-evaluation, BLANKVALUE/ISBLANK guards, ISPICKVAL vs =, VALUE/TEXT/DATETIMEVALUE coercion, time-zone differences between TODAY/NOW, and the 5,000-character formula limit. NOT for record-level formula fields on objects — see admin/formula-fields. NOT for Validation Rule formulas (different runtime context) — see admin/validation-rules.
Install with Codex or Claude Copy this prompt, paste it into Codex, Claude, or another assistant, and let it review the skill page and install it for you.
A direct command skips the review prompt. Inspect the source before running it.
Author NULL-safe, type-correct, performance-aware Formula resources and condition expressions in Flow: lazy re-evaluation, BLANKVALUE/ISBLANK guards, ISPICKVAL vs =, VALUE/TEXT/DATETIMEVALUE coercion, time-zone differences between TODAY/NOW, and the 5,000-character formula limit. NOT for record-level formula fields on objects — see admin/formula-fields. NOT for Validation Rule formulas (different runtime context) — see admin/validation-rules.
category
flow
salesforce-version
Spring '25+
well-architected-pillars
["Reliability","Performance"]
triggers
["flow formula returns null unexpectedly","ISPICKVAL in flow formula","5000 character formula limit","formula inside loop slow","BLANKVALUE flow formula null handling","TODAY vs NOW time zone flow","VALUE TEXT DATETIMEVALUE flow type coercion","flow formula resource referenced many times performance"]
["Flow design + the expression / formula intent","Return-type expectation (Text, Number, Boolean, Date, DateTime, Currency)","Whether any input is nullable","Whether the formula will be referenced inside a Loop body"]
outputs
["Correctly typed, NULL-safe, performance-aware formula resource OR decision-condition expression","Cached Assignment alternative when formula is referenced repeatedly inside a loop","Composed formula chain when the single-formula 5,000-char limit is at risk"]
dependencies
[]
version
1.0.0
author
Pranav Nagrecha
updated
2026-04-27T00:00:00.000Z
Flow Formula And Expression Patterns
Activate when authoring or reviewing any Flow Formula resource, Decision condition expression, or Screen-component formula property. The skill enforces NULL-safe wrapping, correct type coercion, lazy-re-evaluation awareness, and the 5,000-character ceiling — failure modes that surface at runtime as silent NULLs, "Comparison value cannot be null" decision errors, and per-iteration performance cliffs inside loops.
Before Starting
Gather this context before writing or fixing any formula in Flow:
What type does the formula return? Boolean, Text, Number, Currency, Date, DateTime, Time. Flow rejects implicit return-type changes; coercion must be explicit (VALUE(), TEXT(), DATEVALUE(), DATETIMEVALUE()).
What types are the inputs and which are nullable? Any null input propagates to a null output for arithmetic, comparison, and logical operators (NULL + 1 = NULL; NULL && TRUE = NULL). Wrap nullable inputs with BLANKVALUE(field, default) or IF(ISBLANK(field), default, field).
Will the formula be referenced inside a Loop body or by 3+ elements? Each {!FormulaResourceName} reference re-evaluates the entire expression. A formula touched 10 times inside a 200-iteration loop runs 2,000 times. Cache the result in an Assignment if the formula is non-trivial.
Is the formula approaching 5,000 characters? Single-formula limit. Decompose into composed Formula resources (FormulaA references FormulaB references FormulaC). Each layer still re-evaluates lazily.
Is this a picklist comparison? Use ISPICKVAL(PicklistField__c, "Value") or INCLUDES(MultiSelectField__c, "Value"). Do NOT use = against a literal string — that compares the running locale's API name vs label and is a P1 source of silent false negatives.
Is this a date/time formula?TODAY() returns the running user's local date in their TZ. NOW() returns the org's default TZ. Cross-TZ teams hit edge cases at day boundaries.
Is the formula in a Decision element vs a Formula resource vs a Screen-component property? Same language, three evaluation contexts. A Decision condition that throws "Comparison value cannot be null" needs the same NULL-guard treatment as a Formula resource.
Core Concepts
Concept 1: Three places formulas appear (one language, three contexts)
The Flow formula language is a strict subset of the standard Salesforce formula language documented at Help → Formula Operators and Functions. The same syntax appears in three distinct evaluation contexts:
Formula resource — declared once under Resources, referenced by {!FormulaResourceName}. Lazy: evaluates each time it's referenced.
Decision condition expression — formula entered directly inside an outcome's condition row when "Formula Evaluates to True" is selected. Evaluates once when the Decision element runs.
Screen-component formula property — formulas inside Display Text, default values, validation, and reactive component bindings. Evaluates per-screen-render in flow runtime.
All three share the same operator/function library plus flow-specific globals like {!$Flow.CurrentDateTime}, {!$Flow.CurrentDate}, {!$Flow.FaultMessage}, {!$Flow.InterviewStartTime}, and {!$Flow.ActiveStages}. Validation Rule formula context is NOT identical — validation rules can call PRIORVALUE() and ISCHANGED(), Flow cannot. Workflow Field Update formula context is a third dialect — also not interchangeable.
Concept 2: Lazy re-evaluation (the loop-body performance trap)
Flow does NOT memoise Formula resource results. Every {!FormulaResourceName} reference triggers a fresh evaluation. The cost compounds:
A Formula resource referenced 1 time outside a loop: 1 evaluation. Free.
A Formula resource referenced 1 time inside a loop iterating 200 records: 200 evaluations. Usually fine.
A Formula resource referenced 10 times inside a loop iterating 200 records: 2,000 evaluations. Visible CPU.
A Formula resource that itself references 3 OTHER Formula resources, each referenced 5 times in a 200-iteration loop: 200 × 5 × 4 = 4,000 evaluations and possible CPU-time governor breaches.
The fix: when an expensive formula will be referenced repeatedly, evaluate it ONCE in an Assignment to a typed variable, then reference the variable from then on. Decision elements, Screen components, and downstream Assignments all read the cached variable — zero re-evaluation cost.
Concept 3: NULL propagation and explicit type coercion
Salesforce formula NULL semantics are SQL-like, not Java-like:
NULL + 1 → NULL, not 1.
NULL = NULL → unknown (often FALSE in practice — use ISBLANK() to test for null).
NULL && TRUE → NULL, not FALSE. A Decision condition referencing a NULL-tainted formula throws or evaluates the default outcome unexpectedly.
"" = NULL → TRUE for Text (empty string and null are interchangeable for Text fields). FALSE for Number (0 is not null).
Type coercion is explicit:
Text → Number: VALUE(TextVar). Throws at runtime if TextVar is non-numeric — wrap nullable input with IF(ISBLANK(TextVar), 0, VALUE(TextVar)).
Number → Text: TEXT(NumberVar). Returns no thousands separator and no localisation.
Date → Text: TEXT(DateVar) returns ISO YYYY-MM-DD. Concatenation "Date: " & DateVar returns the running user's locale format — surprising in cross-locale orgs.
Text → Date: DATEVALUE(TextVar) requires YYYY-MM-DD.
Text → DateTime: DATETIMEVALUE(TextVar) requires YYYY-MM-DD HH:MM:SS in GMT.
Picklist → Text: TEXT(PicklistField__c) returns the API name, NEVER the label. Comparing labels in formulas is impossible without a Custom Metadata lookup.
Common Patterns
Pattern 1: BLANKVALUE Default Wrap
When to use: Any time a formula consumes a nullable input (a record field that's not required, an optional Screen input, the output of a Get-Records that may be empty).
How it works:
// Risky: returns NULL if Discount__c is null, then propagates everywhere downstream.
{!recordVar.Amount} - {!recordVar.Discount__c}
// Safe: defaults Discount__c to 0 when null.
{!recordVar.Amount} - BLANKVALUE({!recordVar.Discount__c}, 0)
For Booleans where you want the missing-input fallback to be FALSE:
IF(ISBLANK({!flagVar}), FALSE, {!flagVar})
Why not the alternative: Skipping the wrap and "checking the input upstream" works until someone adds a new caller. The defensive wrap inside the formula makes the contract explicit and survives upstream refactors.
Pattern 2: Cache Expensive Formula in Assignment
When to use: A Formula resource (a) costs more than a single field reference (concatenation, REGEX, nested IF, CASE with 5+ branches, date math) AND (b) will be referenced more than 2 times, especially inside a Loop body.
How it works:
// Inside Loop body — BEFORE (re-evaluates 6 × 200 = 1,200 times):
// Decision condition uses {!isHighValueOpportunity}
// Assignment 1 sets stageDescription using {!isHighValueOpportunity}
// Assignment 2 sets nextAction using {!isHighValueOpportunity}
// Assignment 3 logs message using {!isHighValueOpportunity}
// Update sets owner if {!isHighValueOpportunity}
// Email body refers to {!isHighValueOpportunity}
// AFTER — single evaluation per loop iteration (200 evaluations total):
// Assignment "cacheHighValue": cachedHighValue (Boolean) = {!isHighValueOpportunity}
// All 6 references downstream now read {!cachedHighValue} (variable, not formula)
Why not the alternative: Leaving the formula referenced N times relies on developers to mentally track per-element cost. The Assignment makes the single-evaluation contract explicit, easy to audit, and survives loop-body edits.
Pattern 3: ISPICKVAL for Picklist Comparisons
When to use: Comparing a single-select picklist field (PicklistField__c) or a multi-select picklist (MultiSelectField__c) against a known value.
How it works:
// Single-select picklist — CORRECT:
ISPICKVAL({!recordVar.Stage__c}, "Closed Won")
// Single-select picklist — WRONG (comparing label vs API name fails silently in some locales):
{!recordVar.Stage__c} = "Closed Won"
// Multi-select picklist — CORRECT:
INCLUDES({!recordVar.Industries__c}, "Healthcare")
// Negation:
NOT(ISPICKVAL({!recordVar.Stage__c}, "Closed Lost"))
// Multiple values:
OR(
ISPICKVAL({!recordVar.Stage__c}, "Closed Won"),
ISPICKVAL({!recordVar.Stage__c}, "Closed Lost")
)
// Convert picklist to API-name Text for further string ops:
TEXT({!recordVar.Stage__c}) & " — recorded"
Why not the alternative:= works for English-only orgs against the active default value, then breaks the moment a translation pack is enabled or someone changes the picklist label without changing the API name. ISPICKVAL compares against the API name and is locale-immune.
Decision Guidance
Scenario
Recommended Approach
Reason
One-time computation referenced once, outside a loop
Formula resource
Lazy evaluation cost is 1; readability beats variable indirection.
Computation referenced 3+ times inside a loop body
Assignment to typed variable (cache the result)
Avoids re-evaluation per reference.
Branching on the value of a Boolean/Number/Date
Decision condition (formula or operator)
Decision is the explicit branching primitive; formulas inside a Decision condition are fine.
Branching on a single-select picklist
Decision condition with ISPICKVAL(...)
Locale-safe; = against a literal is a known P1 bug.
Multi-step transformation across many fields with Apex-like logic
Apex Invocable Action returning a typed output
Beyond ~3 layers of nested formulas, Apex is more testable, debuggable, and reusable.
Expression > 4,500 chars and growing
Multiple composed Formula resources OR an Invocable Apex action
The single-formula limit is 5,000 characters; composition keeps each layer under the cap.
String concatenation that includes a Date and runs in a multi-locale org
Boolean expression of 5+ ANDs/ORs with nullable inputs
Wrap each input with BLANKVALUE first, then combine
NULL propagation makes naïve AND(...) evaluate to NULL not FALSE.
"Has any of these picklist values" against a multi-select
INCLUDES(MultiPicklistField__c, "value")
INCLUDES is the only correct primitive for multi-select.
Recommended Workflow
State the contract. Write down the formula's expected return type, every input, which inputs are nullable, and where the formula will be referenced. If references include a Loop body, note the iteration count expectation.
Choose the context. Formula resource (reusable), Decision condition (one-time branching), Screen-component property (per-render). If the same expression is needed in 3+ places, choose the Formula resource and DRY the call sites.
Author with NULL guards first. Wrap every nullable input with BLANKVALUE or IF(ISBLANK(...), default, value) BEFORE composing the rest of the expression. NULL guards are far cheaper to add upfront than to retrofit after a runtime null surfaces in a downstream element.
Make all type coercion explicit. Replace any implicit coercion with VALUE(), TEXT(), DATEVALUE(), DATETIMEVALUE(). For picklists, replace any = with ISPICKVAL() and any "contains" with INCLUDES().
Audit for re-evaluation cost. Count how many places reference the Formula resource. If any of them are inside a Loop, count references × loop_iterations. If the product exceeds ~500 and the formula is non-trivial, refactor to an Assignment-cached variable.
Check the 5,000-char ceiling. If the formula exceeds ~4,500 characters, split into composed Formula resources before you hit the hard limit at deploy time.
Verify in Flow Debug. Run the flow with realistic null and edge inputs (empty Get-Records result, missing optional fields, picklist with 0 selections, cross-TZ date boundary). Confirm no NULL surfaces in a downstream Decision or Update.
Review Checklist
Every nullable input is wrapped with BLANKVALUE or IF(ISBLANK(...), default, value).
Every picklist comparison uses ISPICKVAL (single) or INCLUDES (multi), not =.
Every Text↔Number, Text↔Date, Text↔DateTime coercion uses VALUE / TEXT / DATEVALUE / DATETIMEVALUE explicitly.
Formula resources referenced inside a Loop body are either trivial (single field reference, single arithmetic op) or cached in an Assignment.
Each Formula resource is under 4,500 characters; composed chains used for any larger expression.
TODAY() vs NOW() choice is documented when used in a multi-TZ org — confirm the right TZ semantic is desired.
Decision condition formulas and Screen-component formulas have been reviewed with the same NULL-safety + coercion checklist as standalone Formula resources.
Composed formula chains do not exceed 3 layers of nesting (FormulaA → FormulaB → FormulaC); deeper chains should become Apex Invocable.
No Workflow-only or Validation-Rule-only formula functions are used (e.g. PRIORVALUE, ISCHANGED).
Salesforce-Specific Gotchas
NULL propagates through arithmetic and Boolean operators.NULL + 1 = NULL, NULL && TRUE = NULL, NULL > 0 = NULL. A Decision condition that depends on a NULL-tainted formula either throws "Comparison value cannot be null" or quietly takes the default outcome — both are P1 production failures.
= against a picklist literal is a silent locale bug.{!Account.Industry} = "Healthcare" compares the value the user sees (label) to the literal string. Translation packs, label edits, or API-name vs label drift all turn this comparison FALSE without warning. Always use ISPICKVAL.
Lazy re-evaluation inside Loop bodies multiplies CPU cost. Every {!FormulaResourceName} reference re-runs the formula. Six references × 200 iterations = 1,200 evaluations. Cache in an Assignment when the formula is non-trivial and referenced more than twice in a loop body.
5,000-character limit per formula. The platform rejects deploys with a formula above 5,000 characters. Compose multiple Formula resources (FormulaA references FormulaB references FormulaC) — each child counts independently. Going deeper than ~3 composition layers is a smell; promote to Apex Invocable.
TODAY() and NOW() use different time zones.TODAY() returns the running user's local date in the user's TZ. NOW() returns the org's default TZ. In a multi-TZ org, "TODAY at 1am Pacific" and "TODAY at 1am Eastern" are different dates — formulas comparing TODAY() to a stored Date can be off-by-one for users on the far side of the org's default TZ.
TEXT(picklist) returns the API name, not the label. Any user-facing string built from TEXT(PicklistField__c) will show the API name. Use a Get-Records on a Custom Metadata mapping or a hardcoded CASE to map to label.
Implicit Date-to-Text concatenation uses the running-user locale."Created on " & {!recordVar.CreatedDate} returns "10/27/2026" for US users and "27/10/2026" for UK users. Use TEXT({!recordVar.CreatedDate}) for deterministic ISO output, or format explicitly with LEFT/MID/RIGHT of TEXT().
Decision conditions evaluate referenced formulas EVERY time the Decision runs. A Decision inside a Loop body that references 3 Formula resources costs 3 evaluations per iteration — auditable in the same way as element-level references.
Output Artifacts
Artifact
Description
Formula resource definition
Frontmatter (name, return type, dependencies on input vars), expression body with NULL guards and explicit coercion.
Cached-variable Assignment refactor
When the formula was being re-evaluated inside a loop — the Assignment that materialises the result and the list of downstream references switched to read the cached variable.
Composed formula chain
When the original expression exceeded 4,500 chars — the parent + child Formula resources, each independently under the limit.
Decision-condition rewrite
When the original = against a picklist label was a bug — the ISPICKVAL or INCLUDES rewrite.
Review report
Itemised list of NULL-guards added, picklist comparisons rewritten, type coercions made explicit, and re-evaluation hot-spots fixed, with line-level pointers in the flow XML or screenshot evidence.
Related Skills
flow-resource-patterns — broader guidance on naming, scoping, and choosing among Variable / Constant / Formula / Choice / Stage resources. This skill handles the formula-specific subset.
flow-decision-element-patterns — when the formula is being authored as the condition of a Decision outcome, pair these two skills.
admin/formula-fields — for record-level formula fields on objects (a different runtime context — same language, but evaluated by the platform on read, with field-history and reportability concerns this skill does not cover).
flow-collection-processing — when caching a formula across loop iterations, the assignment pattern interacts with collection-iteration patterns documented there.
flow-bulkification — performance audits of formula re-evaluation overlap with broader bulkification work.
flow-loop-element-patterns — formula re-evaluation is one of the top performance traps inside Loop bodies; cross-reference for loop-body design.
flow-runtime-error-diagnosis — when "Comparison value cannot be null" or "The formula expression is invalid" surfaces at runtime, that skill handles the diagnostic flow; this skill handles the prevention.
