Axiomora AIAxiomora.
AI Assistants

Assistants that answer from approved information — and stop where they should.

An assistant should never invent a policy, a price or a commitment. We design the scope, the sources and the handover before we design the conversation.

Knowledge Assistant DesignBook a consultation
Support team reviewing approved reference material on screen while designing assistant answer boundaries
Two colleagues reviewing a printed policy document and marking approvals with a pen
The principle

Scope is the product. The chat window is just the surface.

Most disappointing assistant projects fail for the same reason: nobody agreed what the assistant was allowed to say. It then answers confidently on topics the business never approved, and trust disappears in a week.

We work the other way round. The approved source set comes first, the escalation rules second, and only then do we tune how it responds. It is slower to start and considerably calmer afterwards.

We do not claim an assistant understands every question or produces error-free output. We claim it stays inside a boundary you defined and tells you when it cannot help.

Assistant types

Five patterns, each with a stated boundary.

Each pattern is scoped to a department and an approved source set. Combining them is a later decision, not a starting point.

Customer-Facing Assistant

Answers published service questions, collects enquiry details in a consistent format and escalates anything account-specific to a named team.

Handles
  • · Opening hours, services, process questions
  • · Structured enquiry capture
  • · Out-of-hours acknowledgement
Hands over
  • · Pricing negotiation
  • · Contractual commitments
  • · Complaint resolution

Employee Knowledge Assistant

Lets authorised staff search approved policies, procedures and internal guides, with the source document shown alongside the answer.

Handles
  • · Policy and SOP lookup
  • · New-joiner questions
  • · Form and template location
Hands over
  • · HR decisions
  • · Interpretation of contracts
  • · Anything outside the approved source set

Sales Assistant

Prepares lead summaries, drafts follow-up messages for owner approval and keeps CRM records tidy.

Handles
  • · Lead summarisation
  • · Draft follow-ups
  • · CRM field hygiene
Hands over
  • · Sending outbound offers unattended
  • · Discount approval
  • · Forecast commitments

Operations Assistant

Supports process instructions, document handling, task preparation and internal request routing across departments.

Handles
  • · Procedure lookup
  • · Request intake and routing
  • · Handover summaries
Hands over
  • · Safety-critical instructions
  • · Supplier commitments
  • · Exception sign-off

Management Assistant

Prepares summaries, recurring reports and meeting packs from operational records, with figures kept visible for checking.

Handles
  • · Recurring report preparation
  • · Meeting pack assembly
  • · Structured theme summaries
Hands over
  • · Final reporting to regulators
  • · Performance judgements
  • · Unreviewed circulation
Support colleagues reviewing escalated conversations and monitoring notes together at a shared desk
Escalation, not abandonment

Handover is a designed step, not a failure state.

When a question falls outside scope, the conversation, the context and the customer's original wording go to a named team.

How an assistant gets built

Five steps, in this order.

The sequence matters. Tuning tone before agreeing sources produces an articulate assistant that is wrong.

  1. 01
    Source set agreed

    We list the documents the business is willing to stand behind, and exclude drafts, personal notes and anything unapproved.

  2. 02
    Answer boundaries written

    Topics the assistant may answer, topics it must hand over, and the wording it uses to do so.

  3. 03
    Escalation designed

    Where a handover goes, in what format, and what the customer sees while they wait.

  4. 04
    Tested on awkward cases

    We test on the questions your team finds hardest, not the easy demo ones.

  5. 05
    Reviewed after launch

    Logged questions are sampled so gaps in the source set are found and filled.

Operational documents and forms organised on a desk ready for structured review
Controls

Documented before go-live, reviewed during optimisation.

These eight controls appear in every assistant deployment we run. They are recorded in the scope of work, so there is no ambiguity about who owns which one.

  • Approved knowledge sources

    An agreed, versioned set of documents the assistant may answer from.

  • User access controls

    Who can use the assistant, from where, and with what visibility.

  • Escalation rules

    Named conditions that route a conversation to a person, with context preserved.

  • Feedback and correction

    A route for users to flag a wrong answer and have the source corrected.

  • Audit history

    A record of questions asked, answers given and sources used.

  • Data-handling policy

    What is stored, for how long, and what is excluded from the assistant entirely.

  • Human validation checkpoints

    Cases that must be reviewed before a response leaves the business.

  • Unsupported request handling

    A clear, non-guessing response when a question is out of scope.

What we will not claim

Assistants do not understand every question, do not guarantee accuracy and do not remove the need for trained staff. Confidence checks, escalation rules and human review exist precisely because that is true.

How human review worksCatalogue and pricing