AI AGENT, B2B EMAIL
AI AGENT, B2B EMAIL
An AI agent handling B2B customer-care email — quotation and order requests, 92% of volume. Phase 1: a validated design spec, one quarter, sole designer.
Client
DIRECTOR OF CUSTOMER OPERATIONS
Year
2026
Category
Application Design
PROJECT OVERVIEW
PROJECT OVERVIEW
An AI agent that reads incoming B2B customer emails, extracts quote and order details, resolves what it can verify, and escalates only what actually needs a person's judgment.
An AI agent that reads incoming B2B customer emails, extracts quote and order details, resolves what it can verify, and escalates only what actually needs a person's judgment.
PROJECT OVERVIEW
An AI agent that reads incoming B2B customer emails, extracts quote and order details, resolves what it can verify, and escalates only what actually needs a person's judgment.
GAP
GAP
139,000 emails/year, 91.7% already self-resolved by reps — but nothing distinguished simple lookups from cases genuinely needing judgment. Every case got equal, unnecessary attention.
139,000 emails/year, 91.7% already self-resolved by reps — but nothing distinguished simple lookups from cases genuinely needing judgment. Every case got equal, unnecessary attention.
GAP
139,000 emails/year, 91.7% already self-resolved by reps — but nothing distinguished simple lookups from cases genuinely needing judgment. Every case got equal, unnecessary attention.
MY ROLE
MY ROLE
Analyzed a year of unstructured customer-care case data to identify recurring request patterns and map which case types were resolved from memory versus verified against a system.
Reframed the brief from response automation to authority design, defining where an AI agent could act alone and where it must defer to a human.
Designed the end-to-end agent flow: field extraction, clarification logic with a three-attempt escalation cap, mandatory ERP validation, and full-context handoff to reps.
Designed and built out the internal ops console end-to-end , regional routing, knowledge-base approval queue, business-rule configuration, and per-account accuracy reporting, with no existing design system to inherit.
Defined the guardrails the system depends on, including a rule that the agent never asserts unverified pricing or lead-time data.
Held the human-in-the-loop position against an initial ask for full automation, and built the evidence case that made it defensible to stakeholders.
Worked directly with ops and engineering leads to align the design against technical feasibility and rollout constraints across a 30+ country scope.
Supported handoff by documenting decision logic and escalation rules so business rules could be tuned without engineering involvement.
Analyzed a year of unstructured customer-care case data to identify recurring request patterns and map which case types were resolved from memory versus verified against a system.
Reframed the brief from response automation to authority design, defining where an AI agent could act alone and where it must defer to a human.
Designed the end-to-end agent flow: field extraction, clarification logic with a three-attempt escalation cap, mandatory ERP validation, and full-context handoff to reps.
Designed and built out the internal ops console end-to-end , regional routing, knowledge-base approval queue, business-rule configuration, and per-account accuracy reporting, with no existing design system to inherit.
Defined the guardrails the system depends on, including a rule that the agent never asserts unverified pricing or lead-time data.
Held the human-in-the-loop position against an initial ask for full automation, and built the evidence case that made it defensible to stakeholders.
Worked directly with ops and engineering leads to align the design against technical feasibility and rollout constraints across a 30+ country scope.
Supported handoff by documenting decision logic and escalation rules so business rules could be tuned without engineering involvement.
MY ROLE
Analyzed a year of unstructured customer-care case data to identify recurring request patterns and map which case types were resolved from memory versus verified against a system.
Reframed the brief from response automation to authority design, defining where an AI agent could act alone and where it must defer to a human.
Designed the end-to-end agent flow: field extraction, clarification logic with a three-attempt escalation cap, mandatory ERP validation, and full-context handoff to reps.
Designed and built out the internal ops console end-to-end , regional routing, knowledge-base approval queue, business-rule configuration, and per-account accuracy reporting, with no existing design system to inherit.
Defined the guardrails the system depends on, including a rule that the agent never asserts unverified pricing or lead-time data.
Held the human-in-the-loop position against an initial ask for full automation, and built the evidence case that made it defensible to stakeholders.
Worked directly with ops and engineering leads to align the design against technical feasibility and rollout constraints across a 30+ country scope.
Supported handoff by documenting decision logic and escalation rules so business rules could be tuned without engineering involvement.
Interested in knowing more about the project?
Interested in knowing more about the project?
Interested in knowing more about the project?
More Works More Works
More Works More Works
Fin.
Fin.
Fin.
Fin.
©2024 AMAN ASWANI
GO BACK TO TOP
©2024 AMAN ASWANI
GO BACK TO TOP