Case Study - Delivering a secure multiagent AI assistant on AWS Bedrock

Background

Cochlear partnered with Vivanti to deliver “Clark”, a production multiagent conversational AI system built on AWS Bedrock and powered by Anthropic's Claude models, giving its workforce a secure assistant for drafting, analysis and everyday knowledge work.

Vivanti designed and deployed Clark with security as the first principle, recognising the governance obligations that come with operating in a highly regulated industry. The result is an enterprise assistant that Cochlear's people trust with proprietary information, and one that has become embedded in how work gets done day to day.

Challenge

Across Cochlear's workforce, a significant share of knowledge work was highly manual. Drafting documents, building presentations, writing email and analysing internal and external sources consumed time that skilled people would rather spend on higher-value work. Any solution had to meet Cochlear's data residency and data sovereignty requirements, and give the business confidence that proprietary data could be connected safely and governed properly. These constraints ruled out general-purpose consumer tools and drove the need for an enterprise-grade platform built inside Cochlear's own security boundary.

Key elements of our solution include:

Vivanti built and deployed Clark on AWS Bedrock, working alongside Cochlear colleagues, with Anthropic's Claude models providing the underlying intelligence.

• Deployment in the AWS Sydney region, satisfying Cochlear's data residency and data sovereignty requirements.
• Availability of Anthropic's Claude models in-region, allowing Cochlear to connect proprietary data to the system safely and govern it under existing controls.
• A multiagent architecture supporting document drafting, presentation creation, email and analysis across the enterprise.
• Security-first design appropriate to a highly regulated industry, embedded from architecture through to deployment.

This approach allows Cochlear to extend AI assistance across the workforce without compromising on data governance.

Results

Jason Yeo, Director of Analytics, Data and AI Practice at Cochlear, said: "Clark has had a huge positive impact on productivity, empowerment and curiosity for our colleagues across Cochlear."

Clark has seen broad adoption across the enterprise, replacing manual effort in everyday knowledge work and becoming part of the standard workflow rather than an occasional tool.

• Exceptional stickiness: 52% of weekly users are also using Clark on any given day, an unusually high rate of daily engagement for an enterprise tool.
• Manual effort removed: Drafting, presentation building and research that previously consumed significant time are now assisted end to end.
• Sovereign and governed: Running in the AWS Sydney region on Anthropic's Claude models, Cochlear retains full control over where its data sits and how it is used.

Business Impact and Outcome

These directly translate into the following key business outcome for Cochlear based on our learnings from Capita’s deployment:

In a central case, Clark releases around 58,700 hours of annual capacity, approximately A$5M in fully-loaded workforce value, reinvested into higher-value work rather than headcount reduction.

Clark has made Cochlear approximately 18% more efficient across four key role families, while AI-specific gains are estimated to release an aggregate 5,900 person-days roughly 27 full-time-equivalents of capacity annually.