Contributing to your CPD?
If this article is supporting your professional development, it can count towards your verifiable CPD hours. Use the pop up at the bottom right corner of your screen to add reading this article as an activity to your online CPD record.
ChatGPT exploded into our consciousness last year as the world absorbed its potential to relieve us of everyday mundane tasks. Since then, the use and number of so-called co-pilots – artificial intelligence-powered assistants to help with mundane tasks – have exploded. Trust in their capabilities today is such that many organisations have deployed them to all their staff.
Developed by OpenAI, ChatGPT is powered by a Large Language Model (LLM), which is trained on huge amounts of publicly available data, such as websites, books and research. Co-pilots sit on top of the GPT model to help users code, respond to emails or create presentations, among many other things. They rely on knowledge graphs and good-quality data to work.
Late last year, rather than deploying ChatGPT, Deloitte launched its own tailored version of a co-pilot called PairD (pronounced ‘paired’), which is based on OpenAI’s LLM.
“All the functionality around PairD related to security, data management and some of the additional functionality for creation of PowerPoints, for example, we have built ourselves, as we have significant capabilities and skills in this domain,” says Costi Perricos, Global Artificial Intelligence and Data Lead at Deloitte.
Perricos says the firm launched PairD last October with the idea of a rapid roll-out in the UK to about 25,000 of its workforce. It ran some pilots around Europe just before Christmas and now Deloitte is in the middle of its European-wide roll-out, with around 75,000 of its workforce due to have access to PairD in the next few weeks. In time it will be extended to include member firms in the Middle East.
“There’s been huge demand for it based on the initial feedback we’ve had from the rollout. We’ve been quite surprised just how overwhelmingly positive the feedback has been. People are loving it and they’re loving it on a number of levels,” Perricos says.
Unlike some organisations, Deloitte had never banned the use of ChatGPT, but had warned staff to use it with care until they had their own in-house tool.
“We recognised that we needed to have a secure version that we could put Deloitte data on. We needed to be confident for ourselves, regulators and clients that data was safe in the same way that our email system is,” Perricos says.
Other considerations in the development of PairD included user friendliness and ease of functionality.
Deloitte’s decision to develop PairD was less about the ability to save time – that was a given – and more about the desire for innovation and to be “at the forefront and understand how these technologies work”.
PairD is “a platform that allows you to add extra functionality and modules with relative ease” – so the firm is looking at additional modules across its various businesses, including its tax, legal and risk advisory services.
At present, Deloitte hasn’t fully sanctioned the use of PairD without human scrutiny of results, but Perricos says once it’s certified and they are happy that the risks are covered, the firm will be able to rely on it.
Development continues at pace, Perricos says. Currently the firm is looking at finding “intelligent ways of using a range of LLMs” in order to discover other use cases – such as the ability to have an architecture where you can plug in data and interrogate it with ease.
Accessibility is a key part of PairD, which is why the firm partnered with Scope, the disability charity, to “try and understand what functionality we can enable so that people with disabilities, both physical and mental ones, can use the amazing levelling power that this technology has”.
Other Big Four firms are also beginning to use LLMs from sources such as ChatGPT to create specific generative AI tools.
KPMG has developed Ava, an internal implementation of Gen AI chat, that allows employees to use it in a secure environment. Ava was developed using publicly accessible models and is suitable for use in a regulated setting. The tool can be tailored to specific focuses, including testing and legal.
“Colleagues are saving time using it for document analysis, research, and summarisation. It serves as the front door of Gen AI for the firm, with new features regularly added,” says James Osborn, KPMG UK's Chief Digital Officer. All 17,000 UK partners and staff at KPMG, and member firms, have access to the tool.
Meanwhile, EY last year deepened its “strategic collaboration with Microsoft” to develop EY Intelligent Payroll Chatbot as part of the Next Gen Payroll Platform. EY has built an LLM and generative AI chatbot POC using ChatGPT In Azure OpenAI service, to understand the “anatomy of an individual’s pay slip” and to link regulatory compliance elements with company policies and to answer “granular questions and generate highly personalised explanations for employees”.
PwC uses its own Generative AI chatbot, ChatPwc, which makes OpenAI’s technology available to staff securely. Around 100 employees had access to the technology by last July, but the firm had planned to roll out the tool to its 75,000-strong workforce by the end of last year.
Generative AI guide
Explore the possibilities of generative AI in accounting.