Healthpath
Healthpath: AI-Powered Customer Service Pipeline
Elemra built a five-agent agentic RAG system for Healthpath, integrated with HubSpot and their Laravel customer portal, handling 90% of customer queries in under 4 minutes and avoiding £80k in projected hiring costs.
Healthpath
Elemra built a five-agent agentic RAG system for Healthpath, integrated with HubSpot and their Laravel customer portal, handling 90% of customer queries in under 4 minutes and avoiding £80k in projected hiring costs.
About Healthpath
Customer Response Times From 2 Days to 4 Minutes
Healthpath's customer service team was at breaking point during peak periods. High volumes of repetitive emails were consuming capacity, with standard queries taking 6+ hours for a first response and complex cases stretching to 1–2 days.
Their existing HubSpot chatbot could only point customers to help articles — it had no integration with live customer data. CS reps had to manually piece together context across HubSpot CRM and a bespoke Laravel customer portal (orders, subscriptions, practitioner notes) for every query.
Without automation, maintaining service levels meant taking on £80k in projected hiring costs — headcount growth as a substitute for the productivity problem.
A Five-Agent Agentic RAG System
We designed and built a five-agent agentic RAG system, integrated into both HubSpot and Healthpath's Laravel customer portal. The architecture is the key differentiator: rather than a single prompt handling everything, five specialist agents collaborate on every inbound ticket.
- Email orchestrator agent — receives inbound HubSpot tickets, classifies intent, routes to specialist agents.
- General-knowledge agent — queries a Supabase pgvector index across help articles, blogs, podcasts, and practitioner content.
- Customer-context agent — pulls live data from HubSpot CRM and the Laravel customer portal: orders, subscriptions, test results, and practitioner notes.
- Compliance guard-rail agent — enforces medical-advice boundaries: no diagnosis, no treatment instructions.
- Response synthesis agent — composes the final draft response with citations, posts back to HubSpot for CS team review before sending.
A continuous improvement loop closes the feedback cycle: the CS team rates each draft via a HubSpot dropdown, ratings feed into a root-cause analysis taxonomy (prompt vs. data-access issues), and fixes are deployed against a prioritised backlog rather than through blind prompt tuning.
Tools We Integrated
Customer Response Times From 2 Days to 4 Minutes
Healthpath's customer service team was at breaking point during peak periods. High volumes of repetitive emails were consuming capacity, with standard queries taking 6+ hours for a first response and complex cases stretching to 1–2 days.
Their existing HubSpot chatbot could only point customers to help articles — it had no integration with live customer data. CS reps had to manually piece together context across HubSpot CRM and a bespoke Laravel customer portal (orders, subscriptions, practitioner notes) for every query.
Without automation, maintaining service levels meant taking on £80k in projected hiring costs — headcount growth as a substitute for the productivity problem.
A Five-Agent Agentic RAG System
We designed and built a five-agent agentic RAG system, integrated into both HubSpot and Healthpath's Laravel customer portal. The architecture is the key differentiator: rather than a single prompt handling everything, five specialist agents collaborate on every inbound ticket.
- Email orchestrator agent — receives inbound HubSpot tickets, classifies intent, routes to specialist agents.
- General-knowledge agent — queries a Supabase pgvector index across help articles, blogs, podcasts, and practitioner content.
- Customer-context agent — pulls live data from HubSpot CRM and the Laravel customer portal: orders, subscriptions, test results, and practitioner notes.
- Compliance guard-rail agent — enforces medical-advice boundaries: no diagnosis, no treatment instructions.
- Response synthesis agent — composes the final draft response with citations, posts back to HubSpot for CS team review before sending.
A continuous improvement loop closes the feedback cycle: the CS team rates each draft via a HubSpot dropdown, ratings feed into a root-cause analysis taxonomy (prompt vs. data-access issues), and fixes are deployed against a prioritised backlog rather than through blind prompt tuning.
Tools We Integrated
Results
The five-agent system handles 90% of Healthpath's customer queries autonomously, with draft responses posted back to HubSpot for CS review in under 4 minutes — down from 6+ hours for standard queries and 1–2 days for complex cases.
The team's capacity is now concentrated on exception cases requiring genuine clinical empathy, while the structured feedback loop delivers measurable improvements: a single webhook fix removed 29% of poor-rated draft causes in one release.
Queries Automated
90% of customer queries handled autonomously by the five-agent system.
Response Time
Average response time under 4 minutes, down from 6+ hours.
Hiring Costs Avoided
£80k projected hiring cost avoided through automation.
Results
The five-agent system handles 90% of Healthpath's customer queries autonomously, with draft responses posted back to HubSpot for CS review in under 4 minutes — down from 6+ hours for standard queries and 1–2 days for complex cases.
The team's capacity is now concentrated on exception cases requiring genuine clinical empathy, while the structured feedback loop delivers measurable improvements: a single webhook fix removed 29% of poor-rated draft causes in one release.
Queries Automated
90% of customer queries handled autonomously by the five-agent system.
Response Time
Average response time under 4 minutes, down from 6+ hours.
Hiring Costs Avoided
£80k projected hiring cost avoided through automation.