AI assistant - how it works – Ciphr

This guide explains how Ciphr's AI assistant processes a user's question and returns an answer. It is intended to explain (particularly for IT teams) what the AI assistant does, where data flows, and how security is handled.

Note: For more technical and security detail refer to the Ciphr Trust Centre.

Overview

The diagram below illustrates the four layers involved, from the user opening the AI assistant through to the response being returned

How the AI assistant works

1) AI assistant UI (in-product)

Users access the AI assistant via the Ask button in the top navigation bar of HR. The conversation window opens inside the HR system.

Key points

Nothing additional to install.
No separate authentication flow - users are already logged into HR
The AI assistant is not a standalone product; it’s part of HR, so no new system to learn

2) AI processing layer (request handling and orchestration)

When the user sends a message, it passes to the AI processing layer. This layer does three things:

Identifies the customer and the user making the request
Determines which features are enabled for that customer
Establishes what data the user is permitted to see based on their existing HR permissions

The message is then processed by a language model (to understand the question in plain English) working alongside agent orchestration that determines what data needs to be retrieved to answer it

3) HR data (live data retrieval)

The AI processing layer queries the relevant data from HR, and responses reflect what is currently held in the system (not a cached or static copy).

Current Beta data scope (5 areas)

People and organisation records
Pay and absence
Skills and training
Holiday entitlement
Payments and allowances

4) The response (returned to the user)

The answer is assembled from the retrieved data and returns it to the user in plain, conversational language within the AI assistant UI.

How security works

The AI assistant does not have its own permission layer. It uses the same permission model already in place in HR, the same controls that govern what each user can see when they navigate the system directly.

What this means in practice:

Line managers can see their team’s data (as per their role permissions)
Employees can see their own data
HR administrators see what their role permits

Nothing additional to configure
From a security perspective, there is no extra permission setup required. If a user asks for data they aren’t permitted to access, the AI assistant will tell them they don’t have permission to view it.

Note: For more technical and security detail refer to the Ciphr Trust Centre.

What the AI assistant does not do

It does not store or retain HR data outside of Ciphr’s existing data environment
It does not create or modify records in HR, except for actions explicitly taken by the user (for example, submitting a holiday booking)
It does not have access to data from other Ciphr products (Payroll, Benefits, LMS) - this will expand over time

Further information

Footer

Related to