KPMG is both a partner and a customer for Google Cloud and that dual role is honing methodologies, use cases and approaches for AI agent deployments.
On a webinar for analysts, Google Cloud and KPMG walked through the early lessons learned from deploying Gemini Enterprise.
At Google Cloud Next in April, KPMG said it would expand its AI partnership with Google Cloud. KPMG said it would use Google Cloud to scale its multi-agent platforms to transform business processes and integrate Gemini Enterprise to boost internal productivity.
Specifically, KPMG is leveraging Gemini Enterprise and Vertex AI with other services. Google Cloud is also being used to build AI capabilities and agents for KPMG Law US.
Stephen Chase, Global Head of AI & Digital Innovation at KPMG, said the firm adopted Gemini Enterprise across the workforce with 90% of employees accessing the system within two weeks of launch. "We believe this is the fastest adopted technology our firm has had and we are in a regulated industry," said Chase. "We went into it with the idea this was going to be part of our overall transformation. It was never about individual use cases. It was about sparking innovation."
- Virgin Voyages: Lessons learned from scaling Gemini Enterprise AI agents
- Google Cloud launches Gemini Enterprise, eyes agentic AI orchestration
- Google Cloud Q3 revenue surges 34% as backlog hits $155 billion
- Google Public Sector: AI agents and the future of government
KPMG and Google Cloud teamed up on the Gemini Enterprise deployment to hone best practices for regulated deployments.
Hayete Gallot, President of Customer Experience for Google Cloud's global, multi-billion-dollar commercial business, said scaling AI agents is about building repeatable processes and methodologies to scale.
"Beyond the models, it's really about how you're going to build those multi-agent systems," said Gallot. "We've done a lot of work to help our customers from the learnings we've had in building those multi agents. We've packaged that through our ADK (Agent Development Kit) so they can build their own agents."
Gallot added that Google Cloud is investing in the ecosystem and partners so customers can scale agentic AI. She said that Gemini Enterprise is an example or providing pre-built agents for coding and research while giving the leeway to build and connect to other AI agents.
"The more the ecosystem is on a common set of tools and protocols, the better it is to build those multi agent experiences," said Gallot.
KPMG's internal Gemini Enterprise deployment
Chase walked through the early lessons from the KPMG internal adoption of Gemini Enterprise. Among the key points:
Understand the data and regulatory issues and take a measured approach. Chase said KPMG had a good data foundation and understanding of the regulatory issues. "We took a measured approach to rolling it out and testing," said Chase. "We were doing the evaluations on what we were seeing versus what we thought we might see. Were we getting the right data and responses? Was the connector delivering back what we expected with the right controls? We spent a lot of time testing upfront."
Co-innovation. Chase said Google Cloud and KPMG engineers worked together on operating agents in the consulting firm's security environment. "We were helping actually shape how agents are built in Google Enterprise and used that to build trust in AI in our transformation program," said Chase.
Use cases. Chase said the first problem KPMG was trying to solve--and it's critical in a services firm--was enterprise search. "I need good answers and I need to get them right now," said Chase. "Solving that problem was one of the reasons people gravitated to the system."
NotebookLM as a go-to tool. Chase said NotebookLM got a lot of play internally and about 11,000 notebooks have been shared after a month and a half. Gemini Enterprise's Deep Research AI agent is also getting a lot of usage.
Data quality is everything. Chase said KPMG also worked through data quality issues to make sure responses returned were correct and kept client confidential information private.
Beware of AI sprawl. Chase said one of the things plaguing AI deployments is that they're installing more AI than people can consume.
Client facing deployments
KPMG is also deploying Gemini Enterprise at its enterprise accounts.
Chase said KPMG is looking to take its best practices and make them available broadly to clients.
"Ultimately, clients will share the agents they build with each other. And some of those will be industrialized," said Chase.
Once agents are industrialized they can be distributed and "spark innovation at the edge and core and everything else we're doing," said Chase. "That's what our clients are really interested in."
The other key item in Gemini Enterprise deployments is that it's a horizontal system that "fits really nicely in a heterogeneous environment," said Chase.
"Gemini Enterprise doesn't have to be in a monolithic environment," added Chase.
Gallot said that Google Cloud has revamped its technical teams to be hands on and focus on methodology. "We're building a lot of consultative capability in our front end so our people can spark ideas with customers. We have developed a methodology to help our customers to go from idea to production," she said. "It's technology, methodology, catalog and people."
Enterprises are currently looking for knowledge in agentic AI deployments. Chase said the key issues for clients are:
- Data security and broader cybersecurity.
- Data management.
- Use cases. "We have a dedicated process that we go through to pull use cases from both client work and what we're doing internally," said Chase.
For KPMG, the next step after collecting use cases for processes such as finance, procurement and various operating areas, say consumer lending at a bank, is to create reusable starter kits.
"We're all headed toward orchestrating agents and what we're working on now is the building blocks to get us there," said Chase.
These building blocks are then shared across KPMG's tax, audit and advisory service lines. Every client will have different circumstances, but KPMG's goal is to have common areas that can be adapted. Sharing those lessons will make it easier to generate returns.
"We get a lot of questions in the enterprise and if they're going to invest we need to help demystify AI agents and share lessons," said Chase.
