Book a demo

Site Operations

Service Management

Capital Planning & Project Delivery

Data centre engineer holding a tablet beside server racks, Clarisync solutions overviewSolutions overview. Nine capabilities across Site Ops, Service, and Capital Delivery.

How the AI works, and how it's governed.

A reference for the people who have to sign it off – CTOs, security teams and the board. It sets out how the AI works, what it does with your data, where the boundaries sit, and who stays accountable.

See the principlesRequest the governance pack
A glowing AI cube set into a circuit-board landscape

Six commitments that
govern the AI.

These hold everywhere the AI operates – across every module and every deployed domain. They are the frame within which the more detailed claims in this page sit.

See the Trust Centre
  1. Colleagues analysing charts on a tablet and laptop, illustrating AI that never trains on customer data
    01

    No training on your data

    Your operational data is used to answer your questions – never to train foundation models, and never shared to improve anyone else’s.

  2. Colleagues reviewing a tablet together, illustrating human-in-the-loop AI decisions
    02

    Human-in-the-loop

    Consequential decisions always require a person to review and approve. The AI proposes; accountable humans decide.

  3. Colleagues reviewing dashboards and charts on screens, illustrating explainable AI
    03

    Explainable, with evidence

    Outputs cite the operational context and records they drew on, so a reviewer can trace why – not just accept a black-box answer.

  4. Hand holding a glowing Audit symbol with AI monitoring icons, illustrating governed and auditable AI
    04

    Governed & auditable

    Every AI action is written to a tamper-evident audit trail – who asked, what it did, what it drew on and who approved it.

  5. Pen pointing at donut charts on a printed report, illustrating customer data isolation
    05

    Data isolation & residency

    Customer data is isolated per tenant, with residency options by region.

  6. Hand placing an AI chip on a circuit board, illustrating bounded AI autonomy
    06

    Bounded autonomy

    The AI acts only within explicitly defined limits. Anything outside those limits is escalated to a human – never actioned silently.

Reasoning over your context
— not training on your data.

The AI doesn’t learn your business by ingesting it into a model. It retrieves from your live operational context at the moment you ask, and reasons over that. Your data stays your data.

The approach

Retrieval, not training

Streams of light connecting points across a dark network
  • The AI reasons over your operational context via the Operational Ontology and the connected-operations knowledge graph – the platform’s shared model of your estate.
  • It uses retrieval (RAG) at query time to ground answers in your live records, rather than training on your data.
  • Because it runs on the ontology, it can reason across domains – the relationships a siloed estate can’t traverse.
The models

Foundation models, under control

Rows of server racks in a data centre
  • Answers are produced using foundation models used under enterprise agreements, with contractual data-handling controls.
  • Under those agreements, your prompts and data are not used to train third-party models.
  • The models themselves may be trained or tuned by their providers – what stays constant is that your data is never part of that training, ours or anyone else’s.
  • The approach is deliberately vendor-neutral – we don’t tie your governance posture to a single model provider. Specifics are confirmed in the governance pack.

From assisted, to
automated, to autonomous.

The same maturity model shown on the Platform page – governed and transparent at every stage. Assisted is live today, Automated is in beta, and Autonomous is planned; autonomy is earned, within guardrails, not assumed.

  1. Assisted

    AI surfaces insight and evidence; people decide.

    Live

  2. Automated

    Routine work runs itself, within guardrails.

    Beta

  3. Autonomous

    Self-optimising within defined guardrails.

    Planned

Governed and transparent at every stage – human accountability is retained throughout.

See MIRI on the Platform page

What the AI does on its own —
and what needs a human.

Autonomy is scoped, not blanket. Below is where the AI acts within guardrails and where a person must sign off. All of it is logged, and any output can be challenged.

AI · within guardrails

The AI can do this on its own.

  • Surface insight & recommendations

    Analysis, summaries and suggested next actions, with the evidence they rest on.

  • Draft work & communications

    Prepare draft work orders, notes and messages for a person to review before they take effect.

  • Routine, low-risk actions

    Execute pre-approved, reversible actions that fall inside defined guardrails.

Human sign-off

A person must approve this.

  • Consequential or irreversible actions

    Spend, safety-relevant, contractual or customer-facing commitments require explicit human approval.

  • Changes to rules or the ontology

    Altering governance rules, guardrails or the shared model is never done by the AI alone.

Accountability

Accountability stays with people.

The AI is a capability your teams direct — it does not hold accountability. Consequential decisions carry a named human approver, recorded in the audit trail.

The audit trail

Every AI action is logged

  • Who asked, what the AI did, and the records it drew on.
  • Which guardrails applied and who approved a consequential action.
  • Tamper-evident and retained – exportable for review and audit.

Challenge & override

How to push back on an output

  • Any AI output can be rejected or overridden by an accountable user.
  • The cited evidence lets a reviewer check the reasoning, not just the answer.
  • Overrides are captured so patterns can be reviewed and guardrails tuned.

Need the AI governance pack?

The AI governance & DPIA pack is shared on request for a detailed review.

Request the AI governance pack

Security or technical review? See the Trust Centre.

Continue your review.