You've probably tried some form of automation before. Maybe a Zapier workflow, an email sequence, or a spreadsheet macro. And it probably worked — for a while.
Then something unexpected happened. A customer with a special arrangement. An invoice in a different currency. A support ticket that didn't fit any category. The automation broke, or worse, did the wrong thing.
Welcome to the edge case problem.
Why simple automation fails
Most automation tools work on simple rules: "If X happens, do Y." That's fine for straightforward processes. But real businesses are messy.
Consider an invoice approval workflow:
- Happy path: Invoice arrives → amount is under £5,000 → auto-approve → process payment
- Edge case 1: Invoice is from a new supplier not yet in the system
- Edge case 2: Amount is £4,950 but there's already an unpaid invoice from the same supplier
- Edge case 3: Invoice is in euros, not pounds
- Edge case 4: The line items don't match the purchase order
- Edge case 5: The approver is on holiday
A simple rule-based system handles the happy path. The edge cases either break it entirely or require so many nested rules that the system becomes unmaintainable.
This is why businesses build automation, celebrate the initial time savings, and then quietly go back to doing things manually when the exceptions pile up.
The 80/20 of business processes
In most business workflows, roughly 80% of cases follow the standard path. The other 20% are exceptions, special cases, and judgment calls.
The irony is that this 20% consumes the vast majority of your team's time and energy. The routine 80% is boring but straightforward. The exceptional 20% requires investigation, communication, and decisions.
Simple automation addresses the 80%. AI agents address the full 100%.
How AI agents handle edge cases
An AI agent approaches edge cases differently from a rule-based system because it can:
Understand context. When an invoice arrives from a new supplier, the agent doesn't just flag it as an error. It checks whether the supplier was recently added to the procurement system, whether a team member has corresponded with them via email, and whether the purchase order exists. Then it routes the invoice appropriately.
Make proportionate decisions. A £100 discrepancy on a £10,000 invoice from a trusted supplier is handled differently from a £100 discrepancy from a new vendor. The agent assesses the situation in context, not just against rigid thresholds.
Escalate intelligently. When the agent encounters something it can't resolve, it doesn't just dump it in a generic exception queue. It identifies the right person to handle it, provides full context, and follows up if it's not resolved within a timeframe.
Learn from patterns. Over time, the agent recognises recurring exceptions and can handle them automatically. If a particular client always sends invoices in euros, the agent learns to convert and process them without flagging every time.
Real examples of edge case handling
Support ticket triage: A customer submits a complaint that doesn't fit standard categories. The agent reads the content, identifies it as a billing issue (not a product issue), and routes it to finance instead of tech support. A simple rule would have sent it to the wrong team.
Lead scoring: A lead comes in from a company with 5 employees — normally a low priority. But the agent notices they're in a high-growth sector and their enquiry mentions expanding to 50 staff. It scores the lead as high priority.
Scheduling: A client requests a meeting, but the suggested time conflicts with an internal review. The agent proposes alternative times, considering the client's timezone (they're in Scotland, not London) and the priority of the meeting.
The human-in-the-loop principle
AI agents aren't meant to handle everything autonomously. The best implementations follow a clear principle:
- Automate the routine — the 80% that doesn't need human judgement
- Assist with the exceptions — provide context and recommendations for the 20%
- Escalate the truly novel — flag situations the agent has never seen for human decision
This creates a system where your team spends their time on the interesting, challenging work — the exceptions that genuinely require human intelligence — while the agent handles everything else.
Moving beyond simple automation
If you've been burned by automation that breaks on edge cases, you're not alone. But the solution isn't to avoid automation. It's to use automation that's designed for the messiness of real business.
Start by mapping a workflow that has frequent exceptions. Identify the common edge cases. Then deploy an AI agent that handles both the happy path and the exceptions, with clear escalation rules for anything truly unusual.
The result is automation that works in practice, not just in theory.
Get automation insights in your inbox
Practical tips for UK SMEs. 1–2 per month. No spam, unsubscribe any time.