Support Automation Where Being Wrong Is Expensive
Clinical, financial and education settings share a problem: most enquiries are routine, and the small remainder carry real consequences. These pages cover how that line is drawn and enforced in each setting.
Book a Free CallThe Routine and the Serious Arrive Together
A clinic’s inbox holds appointment changes, opening hours questions and, occasionally, somebody describing symptoms that need attention today. A university’s holds deadline questions and, occasionally, a student in difficulty.
The volume is overwhelmingly routine, which is exactly what makes the exceptions dangerous to automate carelessly. Any system here is judged on how reliably it recognises the enquiry it must not handle.
A Confident Answer to the Wrong Question
The failure that matters is not an agent saying it does not know. It is an agent answering something that resembled a routine question and was not, in a setting where the person on the other end acted on the reply.
Everything in these builds is arranged around making that outcome structurally difficult: a narrow permitted scope, explicit triggers for handover, and a retained record of what was said.
Narrow Scope, Loud Escalation
Capability is deliberately limited.
In these settings a smaller set of things done reliably is worth far more than broad coverage that occasionally misfires.
An explicit permitted scope
Written down and agreed before launch, so what the agent may address is a decision rather than an emergent property.
Triggers for immediate handover
Clinical, financial and welfare signals route to a person at once, ahead of any attempt to answer.
A retained record
Conversations logged so an interaction can be reviewed afterwards rather than reconstructed from memory.
Genuine out-of-hours cover
Routine questions answered overnight, with anything serious escalated through the route you already use.
Connected to real systems
Booking and records integration, so an answer about an appointment reflects what the system actually holds.
Data handled properly
Retention and access decided at design time, because these settings carry obligations that outlast the project.
Settings We Build Support Agents For
Each page covers what the agent is permitted to answer in that setting, what triggers a handover, and how records are kept.
Clinical and healthcare
Where symptom description must escalate rather than receive an answer.
Finance and education
Regulated guidance and student welfare, both handled by handover rather than reply.
What the agent is allowed to answer
The permitted scope is written as a list before anything is built, and it is deliberately short: opening hours, appointment booking and rescheduling, what to bring, where to park, how to access a portal, what a form requires. Everything outside that list is a handover rather than an attempt. Teams are often surprised at how much volume that narrow list absorbs, because the routine share of an inbox is nearly always larger than it feels when the difficult messages are the memorable ones. Widening the scope later is a decision to be made deliberately with evidence, rather than a setting to be relaxed because the agent seemed to be coping.
- ✓Permitted list agreed in writing before launch
- ✓Anything outside it escalates rather than improvises
- ✓Scope widens on evidence, never by drift
What triggers an immediate handover
Certain signals bypass any attempt at an answer: a description of symptoms, an expression of distress, a complaint, a request for regulated guidance, or any mention of urgency. These route to whoever is on duty through the channel already in use rather than into a new queue somebody has to remember to check, because an escalation that lands somewhere unmonitored is worse than no escalation at all. The design assumption is that the agent will occasionally escalate something that turned out to be routine, and that this is the correct trade in a setting where the reverse error carries real consequences.
Records, retention and review
Every conversation is retained so that any interaction can be reviewed as it happened rather than reconstructed afterwards. Retention periods and who can access the record are decided during design, alongside the people who already own those obligations in your organisation, because these settings carry duties that outlive the project and inherited defaults are not a defensible answer. The practical benefit is that the first month of transcripts is the most useful input for deciding whether the permitted scope should widen, narrow, or stay where it is.
Cautious by Construction
In these settings a narrow agent that escalates readily is worth more than a capable one that occasionally answers something it should not have.
Scope agreed in writing
What the agent may address is decided before launch and widened only on evidence.
Escalation to a monitored route
Handovers go where someone is actually watching, not into a new unchecked queue.
Reviewable records
Conversations retained so an interaction can be examined as it happened.
Obligations designed in
Retention and access settled with the people who already own those duties.
What our clients say
Our marketing operations are automated end to end. We brief the outcome and the workflow handles the rest.
They built the automation around how we actually work rather than making us change to fit a tool.
They automated the process work that was quietly eating our week. It runs now without anyone thinking about it, which is the only real test.
AI Support Agents: Your Questions Answered
Will it give medical or financial advice?
What share of enquiries can it handle?
Where do escalations go?
Are conversations recorded?
Can it book and change appointments?
Can we widen what it handles later?
Our setting is not listed. Can you still build it?
Want to Know What Is Safe to Automate?
Send us a week of anonymised enquiries and we will tell you what share is genuinely routine and where the line should sit.
Book a Free Call