Counterintuitive, but true: the biggest automation wins are almost never in the happy flow. They sit in the messy edge cases where people lose time, focus and patience.
1. Happy flow automation is easy
Most organisations start automation where it feels safe:
- Invoices that match PO and GR
- Orders that follow a fixed template
- Forms that are fully filled in
These flows are predictable, well documented and often already partially digitised.
They are also the part of the process where humans spend the least cognitive effort.
Result: quick wins, nice demos, but limited structural ROI.
2. Exceptions are where humans really lose time
Now look at what actually eats up time:
- Missing references
- Price mismatches
- Partial deliveries
- Free-text emails that do not match any template
- Customers who do things “almost right”
Each exception looks small on its own, but together they:
- Interrupt people constantly
- Trigger manual checks across systems
- Require context switching and judgement
- Are hard to standardise in work instructions
In practice, 20 percent of the cases cause 80 percent of the handling time.
This is where automation impact explodes.
3. Why classic RPA alone is not enough
Pure RPA works best when:
- Rules are deterministic
- Inputs are structured
- Decisions are binary
Exceptions break all three assumptions.
That does not mean they cannot be automated. It means you need a stacked approach:
- Business rules for what is explicit and stable
- AI or IDP to interpret unstructured input and variations
- RPA to orchestrate actions across systems
- Human-in-the-loop only where confidence is low
The goal is not zero human involvement.
The goal is that humans only touch the cases that actually need judgement.
4. What exception-focused automation looks like
In mature setups, you typically see:
- Automatic classification of exception types
- Confidence scores instead of hard yes or no decisions
- Pre-filled proposals for the user instead of blank screens
- Feedback loops where human corrections improve the model
This turns exception handling from a time sink into a controlled, optimised process.
How to identify your top 5 exception types
You do not need process mining to start. A pragmatic approach works:
- Pick one high-volume process, for example AP or order processing
- Ask the team to list reasons why cases cannot be processed straight-through
- Group them into categories, not individual stories
- Estimate time spent per category, not number of cases
- Rank by total time impact, not by annoyance level
Those top 5 exception types are almost always your best automation candidates.
In practice, automation value starts appearing the moment you stop optimising the happy flow and start designing for reality.
