Short answer: Use workflow automation for predictable steps between systems, AI for interpreting variable language or documents, and RPA only when a needed system cannot be connected more directly and its screen-based workflow is stable.
The labels are often mixed together, but the choice changes reliability and maintenance. A useful design can combine all three, though the simplest dependable option should be tested first.
The differences
| Approach | Best fit | Main limitation |
|---|---|---|
| Rules-based workflow | A known trigger, conditions, and actions, such as routing an approved request | It cannot safely infer missing business rules |
| AI-assisted step | Language, classification, summarisation, or extraction where inputs vary | Outputs can be wrong or inconsistent and need evaluation |
| RPA | Repeating a user action in a stable interface when direct integration is unavailable | Screen or layout changes can break the automation |
Example: routing a purchase request
If a form has a department, amount, and named approver, ordinary rules can route it. If requests arrive as emails with inconsistent wording, AI might identify the request type or extract a proposed vendor, but the amount threshold and approver should remain explicit rules. If the only available action is a legacy desktop screen without an API, RPA may be considered after checking the screen’s stability and failure path.
How to choose
- Describe the desired result without naming a technology.
- Check whether inputs and decisions are consistent enough for rules.
- Identify the specific ambiguity that AI would resolve.
- Check if systems provide APIs, webhooks, exports, or supported connectors before considering screen automation.
- Compare error visibility, auditability, maintenance effort, and human fallback.
Do not choose AI because it is fashionable. If a deterministic rule can do the job, it is usually easier to test and explain. Do not choose screen automation simply because it appears fast to prototype: the cost of maintaining it matters too.
Formula Works helps scope workflow automation, software integrations, and AI-assisted operations around the actual process.
Further reading
- How to map a business process before automating
- NBER: Generative AI at Work, a customer-support field study; its results should not be applied as a universal productivity promise.
Written by
Formula Works
Published
Fri Oct 09 2026