Crown Resorts / Applied AI

Give the draft a source.
Give the reviewer a head start.

I built a content assistant that brings organisational guidance into the moment someone writes, inside Microsoft Teams.

The working system

Interface: Microsoft Teams
Agent: Microsoft Copilot Studio
Knowledge: SharePoint policy and brand guidance
My contribution: Built the content assistant and connected its knowledge sources.

01 / The challenge

The guidance exists.
It needs to reach the draft.

Preparing content meant writing an agency brief, waiting for a draft and bringing it back for internal review. Handoffs and competing priorities could stretch that process across several weeks. Brand and policy checks still had to happen before publication.

My approach was to bring policy and brand knowledge into a tool people already use. A brief or existing copy goes into Teams; the assistant uses SharePoint knowledge to help prepare a policy-aligned draft.

The useful AI task is grounded drafting. The approval decision stays with a person.

Built

A knowledge-grounded content assistant

Copilot Studio connected to SharePoint guidance and made available in Teams. This is retrieval-augmented generation, or RAG: retrieve relevant reference material, then use it to inform the response.

Designed extension

A more focused legal handoff

A proposed reviewer packet and routing workflow, with unresolved claims, source references and questions for a human reviewer. This extension remains proposed.

02 / What I did

Bring the guidance
into the drafting process.

Start where people work

Teams puts the assistant alongside the drafting conversation. SharePoint supplies organisational context instead of relying only on a model’s general knowledge.

Make uncertainty reviewable

The legal-workflow design separates missing information from unsupported claims. A reviewer should see the question and the relevant evidence, rather than reconstruct the whole conversation.

Separate assistance from authority

A retrieved document can inform wording. It cannot, by itself, prove that an offer is accurate or that a draft is legally approved. The content owner and relevant reviewers retain those decisions.

03 / The assistant in action

From brief or draft
to content ready for review.

Start with an existing draft or a set of facts. See how the assistant retrieves guidance, revises the content and flags missing information or unsupported claims. These simulated examples use real Crown Melbourne venues with illustrative requests and guidance.

Real Crown Melbourne venue; simulated request and response. Conservatory venue context

Start with the person’s copy or brief, separate from the reference knowledge.

Input / Dining email draft

What the person supplies

Start your weekend with Melbourne’s BEST buffet breakfast at Conservatory. Everyone is welcome, whether you’re staying with us or just dropping by!

Supporting facts

The factual starting point

Conservatory serves buffet breakfast at Crown Melbourne. Crown’s public page restricts breakfast to Crown Towers guests on weekends, public holidays and school holidays. No evidence supports the draft’s comparative claim.

The copy and facts enter the conversation. Reference guidance is retrieved separately.

Step 1 of 4

Under the hood: knowledge, instructions and response

Implementation explained / not an internal configuration export

1. Receive in Teams

A person supplies copy or a brief. This is task context, not the reference library.

2. Retrieve from SharePoint

Copilot Studio searches selected knowledge sources for relevant content within the user’s access permissions.

3. Apply instructions

The agent combines the input and retrieved context with instructions. Knowledge supplies guidance; instructions specify how to use it.

4. Return to the person

A draft, revision or clarification supports the next action. Approval stays with the responsible people.

Illustrative instruction for this scenario

Review against retrieved guidance. Preserve confirmed facts, correct eligibility and explain material edits. Do not invent prices or inclusions.

The built foundation is Copilot Studio, Teams and SharePoint. The excerpt display and change log explain the approach; they do not imply a custom vector database, a separately trained model or a verified automated compliance gate.

Microsoft: how SharePoint knowledge works

Explore the proposed legal-review handoff

Designed extension / not deployed

Give the reviewer the question, not just the copy.

AI-prepared summary: Revised draft for review. The dining content owner confirms the current eligibility and reservation link before release. The assistant has corrected the draft, not approved it.
Draft or response: Staying at Crown Towers this weekend? Begin your morning with buffet breakfast at Conservatory. Weekend breakfast is available exclusively to Crown Towers guests. View dining details and make a reservation.
References: Brand voice; Claims and conditions; Conservatory breakfast information
Decision owner: The content owner confirms facts. Legal or the appropriate specialist reviews questions within their remit.

A proposed Power Automate flow could route this packet and record the human response. This demonstration sends nothing and approves nothing.

04 / Results

Better-prepared content.
A faster path through review.

Earlier turnaround

Several weeks

Briefing an external agency, waiting for content and completing internal review created a long preparation cycle.

Observed turnaround with the assistant

1 to 2 days

For content preparation and internal review, based on my experience of the change. People could draft or revise directly in Teams instead of waiting for the agency cycle. Legal review, when needed, remains separate.

What changed

Drafting moved in-house

The content team could start with facts or existing copy, use organisational guidance during drafting and spend its review time improving a prepared response.

The next step: a focused legal handoff

I designed a proposed extension that would give legal reviewers an AI summary, source references and unresolved questions alongside the draft. Its purpose is to reduce the work of reconstructing the context. Legal routing has not been implemented.

How the technology works

Copilot Studio retrieves guidance from SharePoint and makes the assistant available in Teams.

A related challenge?

Make the knowledge useful
in the work.

For conversations about applied AI, measurement and workflows with clear human ownership.

Start a conversation

Next case: Victorian Restorations