Safety Management Study Reports in 2 to 3 Days Instead of a Week

The situation

One safety study report took about a week.

The firm saw the opportunity in AI, but its data was siloed, IT capacity and AI-delivery bandwidth were constrained, and there was a real risk of shadow AI as staff reached for public tools. Off-the-shelf assistants had not met expectations for drafting and grounded data access, and anything adopted had to keep sensitive data on Australian soil. Two workflows hurt most. A Safety Management Study report for a pipeline client was a manual, client-specific document-assembly effort that took roughly a week. Bid, proposal and tender preparation was slow and inconsistent, with a stated goal of cutting bid-preparation time by 30 percent while proving ROI to stakeholders.

What we built

One agentic AI platform, in the firm’s own Azure tenant.

We established a reusable agentic AI platform in the firm's own Azure tenant. Source content from SharePoint and user uploads is secured in the firm's Azure databases and vector stores; agents reach it through secure Azure APIs and Azure OpenAI; drafted content is reviewed and augmented in Azure-hosted interfaces; final documents are assembled through Power Automate and SharePoint templates; and a Power Platform analytics layer provides insight. Every agent runs on this one platform, so new use cases ship with limited technical intervention: a secure internal assistant grounded on the firm's SharePoint, quick-proposal and complex-tender drafting, SharePoint-grounded answers that reference approved, up-to-date policies, project-experience and CV generation, and workflow automation that captures digital sign-offs for ISO and governance audits. The flagship Safety Management Study report portal takes the report type, the client and project details, a review chain from originator to approver, and supporting documents in 8 categories. An LLM on Azure keeps each pipeline client's standard boilerplate intact and inserts or updates only project-specific detail. Engineers review the report section by section, revise any single section with a prompt, and export to DOCX or PDF in the pipeline client's preferred template, with an audit log of who changed what and when.

The proposal suite is built toward the firm's stated goal: 30 percent less bid-preparation time, across about 200 quick proposals and 20 to 30 complex tenders a month.

How it works

The architecture, stage by stage.

1. Inputs

   • SharePoint content and user uploads

2. Secure store

   • Azure databases and vector stores in the firm's own tenant

3. Agents

   • Secure internal assistant and purpose-built agents on Azure OpenAI, accessing the stored content through secure Azure APIs

4. Assemble and review

   • Review and augmentation in Azure-hosted UIs, assembly through Power Automate and SharePoint templates

5. Deliver

   • DOCX or PDF export and client portals

6. Insight

   • Power Platform analytics layer

What changed

The outcomes, as the client measures them.

• Safety Management Study report turnaround fell from about a week to 2 to 3 days, including full review and final report generation.
• Staff have a secure internal assistant, resident in Australia, paid for on use.
• The firm owns a reusable platform on which the next agent ships without starting from scratch.