How to use HighMark Health and Google Cloud Gen Ai to simplify medical claims and improve care: 6 main lessons

Among the many educational and wonderful discussions of the team on the integration of institutions that offer industry leaders in Venturebeat 2025 conversion conference This week was one led Google cloud Will Vice Chairman and Chief Technology Employee (CTO) Grannis and Richard Clark, HighMark HealthFirst Vice President and Head of Data and Analysis.

That session, “New artificial intelligence staple in health care: architecture for multi -style environments, multimedia “,” It presented a pragmatic view of how the two organizations cooperated to spread artificial intelligence on a large scale across more than 14,000 employees of the great HighMark HighMark in the United States (based in West Pennsylvania).

In addition, cooperation on all of these employees has turned them into active users without losing the vision of complexity, organization, or doctors ’confidence.

So, how did Google Cloud and HighMark do so? Reading to find out that.

A partnership based on the prepared foundations

Highmark Health, an integrated defender system that serves more than 6 million members, uses models of IQ of Google Cloud and infrastructure to update old systems, enhance internal efficiency, and improve patient results.

What distinguishes this initiative from each other is its focus on platform engineering – treatment of artificial intelligence as a basic transformation in how to do work, not just another technical layer.

Richard Clark, chief data and analysis employee in Heimark, stressed the importance of building an early flexible infrastructure. “There is nothing more inherited than an encrypted employment platform in Copol,” Clark pointed out, but HighMark merged even those systems with artificial intelligence models. The result: up to 90 % of the work burden without a systematic disorder, allowing the most smooth transformations and real -time visions in the complex administrative processes.

Google Cloud Cto repeated that Grannis repeated that success begins with the ground. He said: “This may take three, four or five years, but if your data is ready, you can run experimentation and assessments that make Amnesty International widely useful,” he said.

From proving the concept to daily use

More than 14,000 HighMark employees use more than 40,000 employees regularly the company’s internal tools of the company, supported by AI and Gemini.

These tools are applied via a set of cases of use – from the generation of intended communications to the members to the documents recovery to process claims.

Clark highlights an example of the service provider that includes accreditation data and verification of the contract. Previously, one of the employees was manually looking for multiple systems to check the willingness of the provider.

Now, artificial intelligence collects these data and the requirements of the interrupted examination, and creates designed outputs with complete categories and recommendations.

What drives this high adoption rate? A mixture of structured libraries, active training and user notes rings. “We just drop the tools and hope that people will use them,” Clark explained. “We explain to them how to make their work easier, then the range based on what is happening.”

Architecture on Chatbots

One of the most appropriate topics of the session was the shift from chat -based interactions to multi -agent systems capable of completing tasks from one side to tip. Grannis described this as a move away from the fast chat models to respond to the tasks and automation.

Granis said: “Think about getting a chat interface and more than that about saying:“ Go to this, return it, and allow me to decide. ”These agents coordinate multiple models, and may infiltrate through different functions – from translation to search to implement the workflow.

HighMark currently experiment with a single use agents for a specific workflow, and the long -term goal is to include these rear systems to perform the procedures independently. This will reduce the need for multiple facades or connectors and allow central control with broader access.

The first task, not the first model

Both speakers emphasized a major mental transformation of institutions: stop starting the model. Instead, start with the task and identify or coordinate models accordingly.

For example, the HighMark Gemini 2.5 Pro uses intense intensive information and Gemini flash for rapid reactions in actual time. In some cases, classic inevitable models are used when the task is best suitable – such as translating patients with the patients into multiple languages. Grannis said, “Your business is your IP. Think about fulfilling the mission, and organizing models to do so.”

To support this flexibility, Google Cloud is investing in the possibilities of directing models and standards. The modern agent protocol initiative, which was presented with Linux, is designed to enhance inter -operating and stability in multiple agents.

Practical advice for institution leaders across sectors

For those looking to repeat the success of HighMark, the committee members provided concrete instructions:

1. Early settingInvest in data preparation and system integration now. Even if the publication of complete artificial intelligence is years away, the reward depends on the early basis.
2. Avoid building your constituent models: Unless your work He is Building models, it’s expensive. Focus on the synchronization and refinement of the specified use cases.
3. Adopt a platform mentalityCentral access to the model and tracking use. Creating a structure that supports the experiment without sacrificing the ruling.
4. Start with tasks, not tools: Determine the result first. Then match it with the form or the structure of the agent that suits the best.
5. Measurement and participationInner adoption grows when employees see practical results. Follow use, capture success stories, and update continuous libraries for approved claims and flows.
6. Work design, not just informationThe future of AI is the implementation of tasks, not fixed insight. Create agents that can lead to real world procedures safely and safely within your systems.

We look forward

Although the partnership between HighMark and Google Cloud is still developing, progress has so far provides a model for others in the field of health care – and beyond – who want to build a developed, responsible and highly used AI systems.

With Clark summary, “It is not a matter of flashy features; it’s related to what helps people already perform their functions better.”

The leaders of the institutions who have missed the session can relax in this: success in obstetric artificial intelligence is not reserved for those who have the largest budgets, but for those who have the clearest plans, flexible platforms, and patience to build strategically.