PwC is piloting an internal generative AI tool as part of the company’s focus on leveling up its AI capabilities and offerings.
ChatPwC serves as a conversational AI assistant for employees, functioning similarly to popular available models such as OpenAI’s ChatGPT or Anthropic’s Claude. Through secure, private access to OpenAI’s technology, the company fine-tuned its model with data around popular tax questions and regulations.
The professional services company has started rolling out the tool to employees in a phased approach, according to Scott Likens, global AI and innovation technology leader at the company. Currently, around 1,000 employees have access to the tool and over the summer the company plans to focus on expanding access more broadly.
“We don’t have to wait on the public models or rely on external providers to add that functionality; we have the ability to do that,” Likens said.
The generative AI model is part of the $1 billion, 3-year AI roadmap PwC’s U.S. firm laid out in April. PwC also plans to bolster AI offerings for customers, strengthen partnerships with tech companies and upskill 65,000 employees with the investment.
“We made the initial wave of investment in ChatPwC, which is our interface to leveraging the large language model from OpenAI and then others as needed,” said Likens. “To do that, we had to make it valuable to the work we do every day, so that means bringing in some of our own [intellectual property].”
The goal is for all of PwC’s 75,000 U.S. employees to have access to the generative AI tool, and then the company will turn its sights on rolling it out globally.
In action
PwC has invested in AI for years and in generative AI from a research and development perspective since the transformer architecture was invented in 2017, Likens said. When the current wave of enthusiasm hit, the company wanted to treat itself as client zero, which meant taking a hands-on approach to fine-tuning tools, frameworks and training programs.
The company plans to tailor its generative AI upskilling efforts to different business departments and front and back office roles.
Most employees won’t need to understand how to configure plugins and prompt with code, but the company wants to broadly raise the base level of knowledge. Understanding how to ask the right questions to get the best answers and spot hallucinations or inaccuracies is key for everyone using the tool, Likens said.
In the engineering, products and technology business, training will go deeper into data science, Likens said.
“We feel passionately about this being additive, not replacing what our people do,” Likens said. “Of course, there are opportunities to be more efficient, have higher quality, but we wanted everyone to kind of understand what this meant and invest in our people and do it right.”
Most businesses looking to leverage generative AI are partnering with providers to skip the initial costs tied to development. But even when businesses opt to use an existing model, costs can still be high.
Even for a company like PwC, which operates globally and saw revenues increase 13.2% to $50.3 billion in FY 22, the cost to support generative AI workloads is top of mind.
“I know everyone goes online and they can use some free versions, but in the enterprise, there’s still a cost to us,” Likens said. “We want to be financially responsible, so that’s why we’re giving this training and guidance, but we trust our people to do the right thing from a business perspective.”
Other consulting firms are pursuing AI strategies as well. EY, which began building on top of OpenAI’s GPT models several years ago, has strengthened partnerships with Microsoft and OpenAI. Deloitte is tapping Google Cloud’s generative AI capabilities to build industry-tailored solutions. Accenture is investing $3 billion in its AI and data practice and plans to add around 40,000 AI-skilled employees through hiring, acquisitions and training.
