- decision
Design compliant conversations for US debt collection
- assumption: the belief that made me confident
Guardrails and natural conversation can coexist even where every sentence has legal exposure
- confidence at the time
- 70% · high
Compliant voice bots
Chatbots and voice bots for US debt collection: guardrails and empathy in the most hostile conversational domain there is.
Conversational Experience Designer at Skit AIRetrospective
the call I made
what it looks like
US debt collection
domain
heavily regulated, compliance-first
what I did
Before "agent" was a job title, this was conversation design under the hardest constraint set available: US debt collection, where a misphrased sentence isn't a UX bug but a legal event.
I worked across intent-entity recognition, flow mapping, dialogue management, and prompt engineering, coordinating with compliance and engineering stakeholders, and set the consistency standards (personality, tone, inclusivity) that kept conversations natural: clean turn-taking, low cognitive load, drawing on psychology and linguistics.
This is where my design background and the agent work meet: everything I now believe about guardrails I first learned somewhere guardrails were non-negotiable.
what happened
Bots stayed within legal bounds in a heavily regulated space while keeping turn-taking clean and cognitive load low
▸ the full record: evidence, alternatives, risks, what I traded away
- risks I named
- Compliance constraints could flatten the conversations into scripts nobody engages with
what I knowingly traded away
- Personality standards (tone, inclusivity, consistency) set as explicit rules rather than left to writer taste: less flair, fewer surprises
built with: Intent-entity recognition · Flow mapping · Dialogue management · Prompt engineering