Workflow automation has the best-looking business case in enterprise software and one of the worst realisation rates. The reason is rarely the technology. It is that automation is applied to processes that exist because of an organisational accident, and encoding that accident into a platform makes it permanent and much harder to remove.
01Delete before you automate
The cheapest automation is the process that stops existing. Before scoping any automation, trace what each step is for. A surprising proportion of enterprise workflow steps are approvals added after a single incident years ago, reconciliation between systems that should be integrated, or data re-entry that exists because two teams could not agree on a schema.
Automating a redundant approval makes it instantaneous and invisible, which sounds like a win and is actually worse - the control now provides no oversight while still being a dependency in the flow. Removing it is the better outcome and costs nothing to run.
02Where platforms earn their licence cost
Commercial automation platforms are worth their price when the work is genuinely long-running and human-in-the-loop: multi-day approvals, tasks that route by role, processes that need an auditable trail of who did what and when, and flows that business users must be able to change without a deployment.
They are poor value when the work is machine-to-machine. Moving data between two systems on a schedule does not need a workflow engine and a per-seat licence; it needs a scheduled job with error handling and alerting, which costs almost nothing to run and is far easier to debug at 3am.
| Workload shape | Right substrate | Why |
|---|---|---|
| Human approvals, days-long, audit trail required | Workflow platform | State persistence and audit are the product |
| System-to-system data movement on a schedule | Scheduled job or pipeline | No human state to hold; licence buys nothing |
| Event-driven reaction within seconds | Queue and consumer | Workflow engines add latency and indirection |
| Rules that business users change weekly | Workflow platform or rules engine | Change without deployment is the requirement |
| Screen-scraping a system with no API | RPA, as an explicit stopgap | Brittle by nature - budget for its replacement |
03Ownership is the thing that decays
Automation platforms make it easy for anyone to build a flow, which is the selling point and the long-term problem. Two years in, a typical estate has dozens of automations, a meaningful fraction built by people who have left, and no reliable way to tell which ones still matter.
The failure this produces is specific: an automation breaks silently, nobody notices because its output was only checked when it was built, and the downstream data has been quietly wrong for weeks. This is not a hypothetical - it is the single most common automation incident we see.
- Every automation needs a named owner and a review date, enforced at creation
- Every automation needs an alert on failure that reaches a human who can act
- Automations with no runs in 90 days should be disabled by default, not left running
- Output correctness needs a check, not just execution success - a flow that runs and writes wrong data reports as healthy
04Where AI agents change the calculation, and where they do not
Agents extend automation to work that could not be encoded as rules - reading unstructured documents, handling requests phrased in a hundred different ways, deciding which of several paths applies. That is a real expansion of what is automatable, and it covers a large share of the processes that resisted traditional automation.
What does not change is the need for determinism where determinism matters. A workflow that moves money, grants access or files a regulatory return should have a deterministic path and a human check, with the model used for classification and extraction rather than for the decision itself. The reliability requirement belongs to the process, not to the technology used to implement it.
Topics
Aarav Patel
Principal Engineer · SyncTrix
Writes about the engineering decisions behind production systems - architecture, delivery and the trade-offs that only show up at scale.
Building something like this?
SyncTrix engineers AI, SaaS, platform and cloud systems for enterprises and high-growth teams. Tell us what you're shipping and we'll scope it with you.