PwC scales GenAI for enterprise with Microsoft Azure AI

PwC, operating in 151 countries and serving a client base that includes 87% of the Fortune Global 500 companies, has consistently prioritized equipping its employees with the most effective solutions available. “We all saw Generative AI take the world by storm,” says Robin Stein, Partner at PwC US. “Right away, we were all brainstorming how we could harness its potential to deliver more for our workforce and clients.”

Stein, a leader in the firm's products and technology organization, is responsible for PwC's global upskilling program. Recognizing the importance of creating a secure and scalable environment for the firm's Generative AI (GenAI) solutions, PwC collaborated closely with Microsoft on the development of ChatPwC, a private GenAI assistant enabled by proprietary GenAI plugins, to help solve a wide array of domain use cases. ChatPwC and these proprietary plugins utilize Microsoft Azure OpenAI Service, Azure AI Document Intelligence, and Azure AI Search for retrieval-augmented generation (RAG). To seamlessly integrate with PwC's proprietary plugin architecture, ChatPwC leverages Azure Kubernetes Service to integrate these services as APIs. Stein highlights the invaluable support received from Microsoft, alongside the quality assurance provided by PwC’s own governance organizations, in enabling the timely development of ChatPwC and GenAI plugins at scale.

Since its deployment in the spring of 2023, ChatPwC has achieved remarkable levels of adoption. Starting with a small pilot program, ChatPwC quickly gained traction and is now used by 75,000 professionals in the United States and a further 125,000 of their colleagues worldwide. Stein emphasizes the significance of this achievement, stating, "Given the substantial amount of data and information that PwC employees handle, providing them with a secure GenAI solution that enables them to perform their work more efficiently with high quality was a top priority for us. Equally important was giving them the essential skills to effectively utilize this solution."

Empowering 200,000 people at once

To highlight the significance of ChatPwC, Stein and her team built a holistic adoption journey for PwC employees. The journey began with a brief pilot program, led by Nathan Kobayashi, Director of Low Code Technology and Generative AI at PwC US, which lasted just over four weeks. This pilot program generated valuable user feedback that played a crucial role in refining the adoption process. Kobayashi emphasizes the importance of brand building, stating, "We recognized the value in establishing the ChatPwC brand, which has become an integral part of our organizational culture."

Generating astounding results

Since the beginning of the ChatPwC development and adoption process, PwC and Microsoft have worked together to enable not only the features that the firm requires, but the security controls as well. For instance, all data ingested into ChatPwC is stored behind secure PwC firewalls and is not used to train large language models.

Since implementation, Jacob Wilson, Principal for Products & Technology at PwC US, and leader of PwC's proprietary GenAI architecture, notes that the large language model and RAG ecosystem behind its GenAI plugins have performed extremely well, even as they address a wide variety of scenarios across industry domains. He has been especially impressed by the text extraction and optical character recognition capabilities of Azure AI Document Intelligence as well as the semantic ranking capabilities of Azure AI Search, both of which have helped ChatPwC and other PwC enterprise applications surface transformative GenAI capabilities and allowed PwC professionals to work differently using GenAI. “We've had thousands of internal and client use cases, and to date we have not had to do any fine tuning of large language models,” he says. Jacob and his team are also expanding upon these services with the use of Microsoft AutoGen to enable agentic systems to further drive workforce transformation through the power of GenAI.

The adoption process has gone so well that PwC leaders, including Stein, have begun similar initiatives for Copilot for Microsoft 365, GitHub Copilot, and the GenAI features recently introduced to Microsoft Power Platform. Through My AI, PwC’s AI upskilling initiative, employees were encouraged to get hands-on with the solutions and apply their prompting skills through events like prompting parties. These learning events have helped drive 95% engagement with My AI. During its Microsoft GenAI pilot program, 88% of users said it saved time, 86% of users were highly satisfied, and 69% of users said it improved accuracy and quality. Amongst the greater employee base, email coaching and the creation of first drafts in Microsoft Word and Outlook have proven especially popular.

GitHub Copilot benefitted from a similar adoption program among the firm’s engineering teams, with 70% of relevant employees having since adopted the solution. “These were not the KPIs we set out with,” says Stein. “Our adoption rates across ChatPwC and each of our Copilot programs simply blew our expectations out of the water.”

PwC has successfully engaged employees from various departments to utilize its GenAI solutions on a daily basis, and the remarkable achievements resulting from their usage are just as noteworthy as the high adoption rate. “There are lots of ways to measure the ROI of GenAI,” says Stein. “Our top ChatPwC users are not only realizing value faster but are completing a variety of tasks up to eight times quicker than previously possible.” In the coming months, PwC anticipates creating even more business value. Stein believes the goal of offering employees opportunities to develop in-demand skills and capabilities has been successfully achieved. “We have provided our people something powerful, and what they have done and continue to do with it has been the most amazing outcome of all.”

Find out more about PwC on Facebook, Twitter, YouTube, and LinkedIn.