CLOSE
aigocy@gmail.com

CUP

TWITTER (X) DRIBBBLE LINKEDIN

Configuration

COLORS
CUSTOM CURSOR
Skip to main content

espialtech

Menu
Start a Project

HelpDeskAI

Home
/
Blog
espial
July 27, 2026
No Comments
HelpDeskAI
Support Agent

Answers from your own knowledge, actions taken through your own tools,
and a handoff to a human — with full context — the moment it should not be guessing.

Project Details

Support volume had outgrown headcount, response times were slipping, and a previous generic chatbot had done real damage: it hallucinated answers, customers caught it, and trust in automated support collapsed. HelpDeskAI was built around the opposite principle — no source, no claim. Approved help content, macros and policies are chunked and embedded into a hybrid vector-plus-keyword store; a re-ranking pass feeds the top results to an LLM with strict grounding constraints.

Every reply carries citations back to the document it came from. Where an answer requires action rather than information, the agent has scoped, audited tool access — issuing refunds, editing orders, resetting accounts, pulling live account state. Confidence and policy gates decide what it resolves on its own and what escalates, and an escalation arrives at a human with a full summary, sentiment read and a suggested reply already drafted.

DELIVERABLES
RAG architecture Tool actions Omnichannel Escalation design
INDUSTRY
SaaS / Retail Support

Project Research

Deflection rate is the metric everyone asks for and the wrong one to optimise alone — a bot can deflect 90% of tickets by frustrating people into giving up.

We instrumented for grounding and CSAT alongside deflection, and tuned the escalation threshold against both. The system is designed to lose a deflection rather than risk an ungrounded answer.

Project Results

HelpDeskAI deflects 72% of tickets with a 38-second average resolution time and a CSAT of 4.7 out of 5 — and, critically, zero ungrounded replies. Agents now spend their time on the cases that genuinely need judgement, arriving with the context already assembled.

RevenueIQ
Sales Intelligence
View Project

RevenueIQ
Sales Intelligence

DESCRIPTION
ICP building, evidence-grounded lead enrichment, prospect scoring and drafted outreach, synced to CRM — 3× pipeline velocity and five hours back per rep, per week.
DELIVERABLES
Enrichment engine, Lead scoring, CRM sync, Forecast dashboard
INDUSTRY
B2B Sales / GTM