Skip to content

service · conversational AI

AI support & conversation agents

Support work, held as a conversation — grounded in your answers, handed over when it should be.

Turning your support and enquiry handling into a conversation an agent can hold — grounded in your own answers, on your own channels, with a clear handover to a person.

  • Claude / LLMs
  • RAG / pgvector
  • MCP
  • WhatsApp Business API
  • Twilio
  • LangSmith

overview

What it is

Most support volume is the same small set of questions asked in a hundred different ways. An agent can hold that conversation — but only usefully when it answers from YOUR documented answers rather than generic web knowledge, and only safely when it knows the edge of what it is allowed to say.

We build the agent around three things: what it is grounded in, what it is allowed to do, and when it must stop and fetch a person. Every conversation is logged and reviewable, so the gaps it reveals become the next answers you write.

what's included

Capabilities

Channel integration

The agent meets people where they already write to you — web chat, WhatsApp, email, or inside your product — rather than behind one more login.

Grounded in your answers

Responses are drawn from your documentation, policies, and resolved tickets, with the source attached, so an answer can be checked rather than trusted blindly.

Scope & refusal design

We decide in advance what the agent must never answer — pricing exceptions, legal or medical questions, anything account-changing — and it declines and escalates instead of improvising.

Handover to a person

Explicit handover rules on sentiment, repetition, risk, and uncertainty, with the full conversation handed to your colleague as context rather than dumped as a fresh ticket.

Conversation review loop

Every conversation is logged and sampled for review, so wrong or thin answers get corrected at the source instead of silently repeating.

Gap reporting

The questions the agent could not answer are the most valuable output of the first month — we report them as a ranked list of the answers worth writing next.

why it matters

Benefits

  • The repetitive share of enquiries gets answered immediately, at any hour, in the channel the person already used.
  • Your team spends its attention on the conversations that genuinely need judgement.
  • Because answers are grounded and cited, a wrong answer is traceable to a document you can fix — not to an unexplainable model mood.
  • The gaps the agent hits become a concrete, prioritised list of what your documentation is missing.

in practice

Use cases

First-line product and account questions, with anything account-changing handed to a person.
Pre-sales enquiry handling — qualifying a question and booking the conversation a human should have.
Internal helpdesk for your own staff, answering from policy and process documents.
Out-of-hours cover that answers what it safely can and queues the rest with full context.

how we work

The engagement

  1. Answer inventory

    We gather what you already answer — docs, macros, resolved tickets — and find the questions that actually carry the volume.

  2. Scope & handover design

    We agree what the agent may answer, what it must refuse, and exactly when it fetches a person.

  3. Supervised pilot

    The agent runs on one channel with a person reviewing its conversations, so it earns its scope instead of being granted it.

  4. Widen & hand over

    We extend to more channels and questions as the review loop stays clean, and hand over a monitored system your team owns.

get in touch

Let's talk

Have an operation you'd like an agent to run, or want to see how AI support & conversation agents could fit your work? Tell us about it and we'll come back with a tailored plan.