Skip to main content

Career Choices Dewis Gyrfa Ltd (CCDG): Cari - 16-18 Triage Chatbot

A bilingual AI-powered triage and signposting chatbot for 16-18 learners and their parents/guardians.

1. Summary

1 - Name

Cari - Post-16 Triage Chatbot

2 - Description

Cari is a bilingual (Welsh / English) AI-powered chatbot designed to support 16-18 learners and their parents/guardians by triaging queries and signposting users to

  • Self-help resources on the Careers Wales website
  • A system for booking careers guidance appointments
  • Human support via the contact centre, where appropriate

3 - Website URL

N/A

4 - Contact email

post@careerswales.gov.wales

Tier 2 - Owner and Responsibility

1.1 - Organisation or department

Career Choices Dewis Gyrfa Ltd (CCDG)

1.2 - Team

ICT Systems Engineers

1.3 - Senior responsible owner

Cyfarwyddydd Adnoddau a Thrawsnewid Director of Resources and Transformation

1.4 - Third party involvement

Yes

1.4.1 - Third party

MOBILISE CLOUD SERVICES LIMITED

1.4.2 - Companies House Number

09082209

1.4.3 - Third party role

Working alongside the Careers Wales project team, Mobilise led the development of the tool and provided knowledge transfers and supporting documentation throughout the project build.

1.4.4 - Procurement procedure type

It was a Mini-Competition among existing framework suppliers for a specific requirement.

1.4.5 - Third party data access terms

N/A

Tier 2 - Description and Rationale

2.1 - Detailed description

Cari is a bilingual (Welsh/English) AI‑powered chatbot designed to support 16-18 learners and their parents/guardians by triaging queries and signposting users to:

  • Existing Careers Wales digital resources
  • Booking routes for careers guidance
  • Contact centre support where appropriate

The system does not provide personalised careers advice, does not make determinations about eligibility or outcomes, and does not replace human decision‑making or professional judgement. It does not support learners still in statutory education, learners in HE or adults aged 19+

2.2 - Benefits

The tool is designed to improve accessibility to careers support by responding to common, non‑complex queries and directing users to existing Careers Wales content or to appropriate contact channels where human intervention is required.

2.3 - Previous process

Users would conduct a manual search of the Careers Wales website, make requests for guidance without diagnosis or fail to seek guidance when it would have been appropriate.

2.4 - Alternatives considered

We use Microsoft products for our ICT systems, including Office, Dynamics and Azure. We therefore conducted our discovery in conjunction with a Microsoft-funded development partner and did not consider any alternatives to Copilot Studio. We wished to build the tool in Copilot Studio because the project also included an element of internal learning and capacity building for future projects and we wished to ensure maximum compatability with our exisitng Microsoft infrastructure as well as making use of the ability to ringfence our data within our Microsoft tenant.

Tier 2 - Deployment Context

3.1 - Integration into broader operational process

Cari is integrated into the Careers Wales service model as a front‑end triage and navigation tool within the organisation’s wider digital and human‑delivered careers support offer. It is located on the Careers Wales website on the “options at 16” and “options at 18” pages in place of the standard webchat, and thus operates at the earliest stage of customer interaction.

Cari informs and supports users to identify appropriate next steps by responding to general careers, learning, and progression queries. Cari provides information on jobs, apprenticeships, courses, funding, and employability. Cari’s outputs triage users towards appropriate Careers Wales digital resources and to human careers support where appropriate.

3.2 - Human review

Cari is a triage and signposting chatbot and does not make decisions, provide personalised advice, or determine outcomes for users. As a result, individual chatbot responses do not require routine case‑by‑case human approval before being presented to users. Human review occurred before deployment, during design, testing, and approval. Outputs, refusal behaviours, and safeguarding responses were reviewed by relevant staff to ensure alignment with approved Careers Wales content, service boundaries, and safeguarding expectations. Scenario based testing, including safeguarding and out of scope prompts, was used to confirm that the system behaves as intended and does not exceed its defined scope. Ongoing assurance is provided through Careers Wales’ AI governance framework, including documented safeguarding assurance, inclusion within a central AI register, and continued oversight in line with organisational AI policy.

3.3 - Frequency and scale of usage

Cari is available, free to use, on the Careers Wales website at all times. There is no data available on usage at the present time

3.4 - Required training

None

3.5 - Appeals and review

N/A

Tier 2 - Tool Specification

4.1.1 - System architecture

The service is hosted on both the Welsh and English versions of the Careers Wales website. The tool is built in the Microsoft Copilot Studio suite and accessed via a Copilot Studio chatbot assistant which is embedded in a publicly accessible web UI. The tool uses procedural questions to determine a user’s status and career confidence level, then generative AI based on flows which have been preprogrammed into the tool manually for each status. The tool only references information on the Careers Wales website and has guardrails in place to deal with safeguarding issues and non-careers related enquiries. It identifies appropriate information and apps on the Careers Wales website and surfaces these to customers as links which open in new browser windows. The tool does not request any personal data from users and chats are not stored by Careers Wales. The production site is available at https://careerswales.gov.wales/ and https://gyrfacymru.llyw.cymru/

4.1.2 - System-level input

Selections to procedural questions or questions and responses typed into the chatbot during the generative phase

4.1.3 - System-level output

  • Links to appropriate information and apps on the Careers Wales website
  • the contact number for the Careers Wales contact centre
  • the telephone number for support from Samaritans and Childine, when a potential safeguarding issue is identified

4.1.4 - Maintenance

The outputs of the tool will be monitored on an ongoing basis to ensure performance is within desired parameters.

4.1.5 - Models

The tool uses Microsoft Copilot Studio

Tier 2 - Model Specification

4.2.1. - Model name

Microsoft Copilot Studio

4.2.2 - Model version

Copilot GPT-5 Chat

4.2.3 - Model task

Interpreting customer requests and provision of answers from publicly available information on the Careers Wales website

4.2.4 - Model input

Responses to procedural questions or questions and responses typed into the chatbot during the generative phase

4.2.5 - Model output

Responses to customer questions including links to apps and resources on the Careers Wales website and/or contact links for human support.

4.2.6 - Model architecture

The model uses Natural Language Processing to “chat” with users within pre-defined parameters, answering only career-related questions. Responses are based solely on information on the Careers Wales website

4.2.7 - Model performance

The tool has been tested through defined demo and scenario‑based testing, including safeguarding‑related prompts, to verify that its behaviours are proportionate to its role as a triage and signposting tool for careers-related enquiries. It has also been tested against intentionally harmful or non‑careers or education‑related prompts to ensure it refuses such queries and does not generate unsafe or inappropriate responses.

4.2.8 - Datasets and their purposes

The model is restricted to using data and information from the publicly available Careers Wales website only.

2.4.3. Development Data

4.3.1 - Development data description

The model is restricted to using data and information from the publicly available Careers Wales website only. https://careerswales.gov.wales/ and https://gyrfacymru.llyw.cymru/

4.3.2 - Data modality

Text based instruction files referencing data from the publicly available Careers Wales website only. https://careerswales.gov.wales/ and https://gyrfacymru.llyw.cymru/

4.3.3 - Data quantities

N/A

4.3.4 - Sensitive attributes

N/A

4.3.5 - Data completeness and representativeness

N/A

4.3.6 - Data cleaning

N/A

4.3.7 - Data collection

N/A

4.3.8 - Data access and storage

N/A

4.3.9 - Data sharing agreements

N/A

Tier 2 - Operational Data Specification

4.4.1 - Data sources

User inputs and the Careers Wales website

4.4.2 - Sensitive attributes

No sensitive attributes are requested from the user and none are produced in the outputs. Once a chat is closed the tool resets and the next chat starts anew. Data is not retained or carried over from previous chats.

4.4.3 - Data processing methods

N/A

4.4.4 - Data access and storage

No

4.4.5 - Data sharing agreements

N/A

Tier 2 - Risks, Mitigations and Impact Assessments

5.1 - Impact assessments

An Equality Impact Assessment has been completed and a Data Protection Impact Assessment screening has been performed. Both of these were completed in June 2026.

The Equality Impact Assessment concluded that Cari has the potential to improve access to careers information and guidance by providing a bilingual, accessible, 24/7 digital service that complements human advisers and particularly benefits young people who face barriers to traditional support. The assessment identified positive impacts across most protected characteristics, highlighted risks relating to accessibility, digital exclusion and potential AI bias, and set out mitigations including WCAG-compliant design, human escalation routes, ongoing bias monitoring, bilingual parity testing and continued alternative support channels. Overall, the assessment recommended proceeding with the project subject to ongoing accessibility monitoring and maintenance, monitoring of AI bias and outputs and monitoring of Welsh language parity.

The DPIA concluded that Cari can be implemented with appropriate safeguards because it provides scalable access to careers information while minimising privacy risks through user-controlled disclosure, no retention of personal data, no external data sharing, and processing within the Careers Wales Microsoft environment. The assessment identifies key risks including inaccurate AI responses, unintended processing of sensitive information, user over-reliance on AI, bias, and excessive disclosure of personal data, but sets out mitigations such as restricted content sources, clear user guidance, privacy notices, human escalation routes, testing, monitoring, and inclusive design. Following review, the Data Protection Officer concluded that the solution could proceed to go live, noting that users remain in control of the information they provide, no third parties store data, and the DPIA should be reviewed if the service is further developed.

5.2 - Risks and mitigations

  • The tool only returns outputs based on information on the Careers Wales website. Minimising the risk of incorrect or misleading responses or hallucinations
  • The system does not provide personalised careers advice, does not make determinations about eligibility or outcomes, and does not replace human decision‑making or professional judgement.
  • The chatbot has been tested against intentionally harmful or non‑careers or education‑related prompts to ensure it refuses such queries and does not generate unsafe or inappropriate responses. This validates that the system cannot be repurposed for unsafe or harmful uses.
  • Safeguarding considerations have been embedded throughout the design and development of the Cari AI triage chatbot - a safeguarding by design approach
  • Appropriate guardrails exist to prevent misuse, manage safeguarding‑related interactions, and ensure human support remains available

Updates to this page

Published 19 August 2026