static credit applications were stalling deals. buyers received an offer link, navigated a long form, and often abandoned or needed customer success to chase missing information.
i designed gynger's first ai-native product: a conversational agent that guides buyers through options and underwriting, while routing exceptions to the right human at the right time.
how it works
- guide the buyer
the agent qualifies intent, explains financing options, and keeps the application moving. - classify every message
qualifying, standard, exception, sensitive, and meta intents each follow explicit routes. - isolate context by party
buyers, vendors, and ops each see only the information needed for their decision.
early beta signal
completion time is significantly faster than the form-based flow, while customer-success involvement in deal closure has dropped.
designed for trust
- constraints are features
vendor escalations use approved templates; sensitive data moves into secure form fields, never chat. - behavior over visuals
the system prompt and conversation state machine became core design artifacts. - human control at the edges
complex credit decisions arrive as structured context packets for gynger ops.
the work shifted design from specifying screens to shaping a distribution of possible behaviors—clear enough for buyers and constrained enough for operators to trust.