Direct answer
Automation executes steps; an operational application makes state, ownership, approval, and exceptions visible and manageable. If the team only needs to move data under stable rules, start with an integration or deterministic automation. If it needs a shared workspace with states, queues, and correction, an application may be justified. If standard software covers the need, configure it before building custom.
Who it is for and when it does not apply
Use this decision when a spreadsheet has become the interface for several roles and it is unclear whether you need an automated job or a system of work. Do not treat a custom application as an inevitable upgrade. Without a stable workflow, owner, and acceptance criteria, either option can merely formalise confusion.
Decision table
Swipe or scroll to compare the columns.
| Dominant need | Initial choice | Why | Required control |
|---|---|---|---|
| Copying between systems, stable rules | Integration / deterministic automation | Reduces re-entry while keeping ownership clear | Idempotency, logging, reconciliation |
| Shared form and reporting | Standard software | The function may already exist | Permissions, configuration, export, and owner |
| Variable data, classification, or extraction | Bounded AI + rules | Helps with interpretation, not authority | Sources, thresholds, evaluation, and review |
| Queues, states, approvals, and custom exceptions | Custom operational application | Provides a coherent work surface | Roles, audit, recovery, and operations model |
| Sensitive or ambiguous decision | Human control | Context and accountability are not implicitly delegated | Stop condition, context, owner, and deadline |
| Rare or undefined process | No automation | Clarification comes before construction | Documentation and review |
Method and definitions
Draw three layers: data (where the value lives), execution (who or what moves state), and decision (who approves an exception). An “integration” connects existing systems while keeping responsibilities explicit; “deterministic automation” applies repeatable rules; an “operational application” gives users an interface for state, actions, and exceptions. They can coexist, but they are not interchangeable.
Ask:
- Can an operator explain the same transition every time?
- Is there a place showing who owns the case now?
- Do cases outside the rule need manual review or correction?
- Does standard software already provide the required roles, history, and export?
- What happens when a connector, model, or user does not respond?
Reusable decision tree
- If there is no owner and accepted outcome, stop construction and document the process.
- If standard software solves the need without creating a competing system of record, configure it.
- If transfer is the problem, define the data contract and integrate.
- If steps have stable rules, automate deterministically and test duplicates and resumption.
- If inputs require interpretation, add bounded AI within a verifiable boundary.
- If the team needs its own queues, roles, approvals, and exceptions, design the application.
- Retain human approval for important outputs, regardless of tool.
Limitation: A well-designed interface does not prove that data is correct, and an automated workflow does not prove that authority is properly bounded.
Synthetic example
For a fictional request table feeding a register, an integration can transport validated fields. A deterministic job can deduplicate and change state. Bounded AI can propose a category from text, while the owner approves uncertain cases. If operators need a queue, filters, case assignment, and an escalation reason, the application is the work surface. None of these choices implies a measured result.
Limitations
The suitable option depends on data, policy, existing systems, workflow maturity, and operating capacity. There is no universal hierarchy between standard software, automation, and custom applications. Examples are synthetic; public AutomateFlow references describe architectures or published implementations with explicit limits, not guaranteed metrics.
Related AutomateFlow pages
Material history
Initial public version or material revision.
Suggest a correction
Include the page and the source supporting the proposed change.