A UK HR SaaS company (~£3.2M ARR, 280 customers) saw support ticket volume grow 5.2x while team headcount only doubled. Average first response time crept past 24 hours; CSAT dropped from 88% to 72%. SpiderHunts built an AI customer support layer combining Anthropic Claude with a Retrieval-Augmented Generation (RAG) pipeline over their internal docs. The AI now resolves 67% of all tickets without human involvement, response times collapsed from 24 hours to 12 seconds, CSAT climbed back to 94%, and the build paid back in under 3 months by avoiding ~£42,000/year of a senior support hire.
Project Snapshot
Bloom HR (anonymised) provides cloud-based HR & payroll software for SMEs across the UK. Between 2024 and 2026 the company grew from 80 to 280 customers — but the support team did not scale at the same rate, and ticket volume outpaced hiring by more than 2x.
Bloom HR (anonymised) provides cloud-based HR & payroll software for SMEs across the UK. Between 2024 and 2026 the company grew from 80 to 280 customers — but the support team did not scale at the same rate, and ticket volume outpaced hiring by more than 2x.
Before SpiderHunts
After SpiderHunts
SpiderHunts built an AI customer support layer sitting between Intercom and the support team. The AI autonomously handles all tier-1 tickets and routes only complex cases to humans — with full context already gathered, including customer plan, recent activity, and prior ticket history.
01
Every new ticket fires a webhook into the SpiderHunts AI layer with the full message thread and customer ID.
02
The AI fetches the customer's plan, payment status, recent activity, and prior tickets via Stripe, HubSpot, and the internal product API.
03
Semantic search over 340+ help docs, release notes, and internal runbooks using pg_vector and OpenAI embeddings — surfacing the 3 most relevant sources per query.
04
Anthropic Claude 3.5 Sonnet drafts a customer-ready answer citing the retrieved sources, in the brand's established voice.
05
High-confidence answers post automatically; low-confidence drafts are queued as suggestions for human agents to approve or edit.
06
Bug reports auto-create Linear tickets with reproduction steps extracted from the conversation; billing issues route to finance with full account context.
07
Agent thumbs-up/down feedback feeds a weekly prompt refinement cycle; failure modes are added to a regression test suite.
08
Resolution rate, response time, CSAT, top failure categories, and cost-per-resolved-ticket — all visible to the support lead in real time.
Production-grade components selected for reliability, observability, and ease of handover.
Numbers measured 3 months post-launch versus the same period the previous year.
We are handling 3x the ticket volume with the same team — and customers actually prefer the AI for most issues because it is instant. The £16k build paid back in under 5 months.
— Head of Customer Success, UK HR SaaS (anonymised)
Want to add an AI layer to your support, sales, or internal ops? SpiderHunts builds production AI agents with RAG, integrations, and observability. Book a free discovery call.
