Skip to main content

Temporal Decision Framework

info

This page is part of the Temporal Platform Hub.

This guide helps you quickly determine whether Temporal is the right solution for your use case.

Temporal decision framework​

note

Tailor these questions to match your organization's technical landscape.

To decide whether Temporal is a suitable solution for your use case, ask yourself 3 questions:

  1. Does your digital process have multiple steps that can fail independently?
  2. Do you need the process to survive failures?
  3. Does your process span multiple services, APIs, or long time periods (i.e. >10 seconds)?

If you answered "yes" to 2 or more questions, Temporal is likely a good fit. Continue reading.

If you answered "no" to all three questions, consider alternatives first. Skip to Bad use cases for Temporal to explore alternative solutions.

Temporal benefits​

note

Highlight benefits that address your developers' pain points.

  1. Durable Execution - your code will always complete.
    • Automatic retry, recovery from infrastructure failures, durable state persistence, and exactly-once execution semantics—all without custom code.
  2. Developer velocity - ship faster with less code to maintain.
    • Write business logic in familiar languages, collaborate with developers across language barriers, eliminate boilerplate infrastructure code, and leverage built-in testing for rapid iteration.
  3. Audit trail - complete visibility in your digital process.
    • Immutable execution history, self-documenting Workflow execution, and operational transparency.
  4. Priority and Fairness - enterprise-grade multi-tenancy.
    • Priority-based execution, and fair distribution of Workflow Executions across your customer base or tenant.
  5. Workflow fabric - break down development silo.
    • Cross-team Workflow orchestration with reusable operations, cross-namespace coordination, and service registry for discoverability.

Good use cases for Temporal​

note

Replace with use cases from your domain. See Customer Stories for inspiration.

Business transactions​

  1. Payment processing
    • Why Temporal is perfect: Multi-party coordination with compensation logic, audit requirements, idempotency guarantees, timeout handling for authorizations that expire, and scalability to support more than billions of transactions per day.
  2. Order management
    • Why Temporal is perfect: Long-running state machines spanning hours to days with complex state transitions, human intervention, parallel operations, different order priority, variable timing per order, and support for more than millions of orders per hour.
  3. Mortgage underwriting
    • Why Temporal is perfect: Weeks-long processes with complex decision trees, multiple external integrations, human approvals, strict compliance requirements, and durable state persistence.

Customer experience​

  1. Marketing campaign
    • Why Temporal is perfect: Multi-channel orchestration with time-based sequencing and long campaign durations with dynamic personalization.
  2. Customer onboarding
    • Why Temporal is perfect: Great for long-running, multi-step, and sometimes human-in-the-loop processes that onboarding often requires.

Data engineering​

  1. Document processing
    • Why Temporal is perfect: Multi-stage pipelines with variable processing times, external service dependencies, rate limit requirements, and coordinated large-scale processing.
  2. Data pipeline
    • Why Temporal is perfect: Data orchestration with complex dependencies, incremental processing, backfill coordination, cross-system dependencies, SLA monitoring, and idempotent execution.
  3. Video processing
    • Why Temporal is perfect: Long-running compute, resource-intensive GPU activities, complex pipelines with parallel variant generation, failure isolation, and cost-optimized scheduling.

AI/ML​

tip

See the AI Engineering section for production patterns, reference architecture, and security guidance specific to AI workloads.

  1. ML inference
    • Why Temporal is perfect: Multi-model pipelines often involve 3–5 sequential steps: embedding, retrieval, reranking, generation, and evaluation. Each step can fail independently. Temporal wraps each LLM or model call in a durable Activity, so a failure at step 4 does not re-run steps 1–3. Fallback logic (try GPT-4o, fall back to Claude if unavailable) is expressed as ordinary conditional code. The Activity history gives model governance teams a built-in record of which model made which decision on which input — without any custom logging infrastructure.
  2. RAG
    • Why Temporal is perfect: Retrieval-augmented generation typically fans out across multiple sources — vector stores, keyword indices, internal APIs — then assembles context within a token budget before calling the LLM. Temporal's parallel Activity dispatch handles the fan-out and convergence naturally. Rate limit backoff, context window management, and evaluation-triggered retry loops are all first-class concerns in an Activity-based design, replacing fragile custom orchestration glue.
  3. AI agents
    • Why Temporal is perfect: Autonomous agents run multi-step loops — plan, act, observe, plan again — that can last minutes (a coding task) or days (a research project). Most agent frameworks keep state in memory and lose everything when the process restarts. Temporal persists the full agent state (message history, tool results, step count) durably, so a crashed Worker resumes the agent exactly where it left off. Human-in-the-loop gates — pausing for reviewer approval before a high-risk action — use Temporal's native Signal and Update primitives rather than polling tables or webhook plumbing. See AI Engineering → Reference Architecture for the canonical pattern.

Operational​

  1. Infrastructure management
    • Why Temporal is perfect: Multi-step provisioning with automatic rollback on failure, idempotent cloud operations, change management, and complete auditability.
  2. CI/CD
    • Why Temporal is perfect: Complex pipeline stages with environment promotion gates, parallel test execution, conditional deployment strategies, automatic rollback monitoring, and approval gates.

Multi-tenant systems​

  1. Tenant-aware task processing
    • Why Temporal is perfect: Task Queue Priority and Fairness provides built-in controls to prevent high-volume tenants from starving smaller ones, all without the need to build custom scheduling logic.

Bad use cases for Temporal​

note

Add anti-patterns specific to your organization's domain and technology stack.

  1. Simple Request-Response APIs
    • No failure recovery needed
    • Better alternative: REST / gRPC server
  2. Real-time stream processing
    • High throughput (>1M events/sec)
    • Ultra-low latency requirements (<100ms)
    • No durable state needed
    • Better alternative: Flink, Amazon Kinesis, Google Cloud Dataflow
  3. Database triggers & stored procedures
    • Logic tightly coupled to database
    • Needs transactional guarantees within single DB
    • No external service calls
    • Better alternative: database native features
  4. Pure Compute Workloads
    • CPU/GPU intensive calculations
    • No I/O or service calls
    • No state management needed
    • Better alternative: AWS Lambda, Spark, Ray

Next steps​

note

Add relevant links (i.e. support channel) for your developers to explore next.

To learn more: