How generative AI can help French-speaking CPAs

For a while, to some extent, speaking French protected French-speaking Canadian CPAs from competing foreign accounting service providers who are not CPAs and who cannot communicate in French. But in light of the linguistic advances made in generative AI over the past two years, is there cause for concern? 

It’s a valid question when you consider that, for example, HeyGen, an AI-powered video creation platform, can clone your speaking voice and style to deliver content in 40 languages. Rask AI, meanwhile, allows users to automatically translate videos and videoconferences into over 130 languages. 

For Olivier Blais, co-founder of Moov AI, a Montreal-based company that provides AI solutions to businesses and recently received an award from the Office québécois de la langue française, recent advances represent an opportunity for all CPAs, whatever language they speak. 

Underrepresentation of francophones 

As the fifth most spoken language in the world, with 321 million speakers, French is an essential language in generative AI, since all the tools are multilingual. “We can have conversations in French,” confirms Blais, “but without any real sense of belonging, without the characteristics specific to Quebec or French-speaking Canada. The content remains very neutral. Only Microsoft Copilot 365 is linked to the language of the keyboard, such as French (Canada), and can provide more adapted solutions. Apart from that, most of the time, the solutions are still very ‘French from France,’ or even translations.” 

With only 10.7 million French speakers (three quarters of whom are in Quebec), Canada’s importance is relative in the eyes of the Big Five investing in generative AI. And while it’s easy to generate any accent, vocalizing cultural references, such as expressions, remains a challenge. 

“For the moment, Quebec and French-Canadian culture is not kept alive through the tools we use,” says Olivier Blais. “It’s a bit like advertising agencies taking English ads and translating them into French without adapting them—it doesn’t work. That’s why the Conseil d’innovation du Québec wants to create a Quebec database (who we are, our expressions, our ways of doing things...), which would serve to feed AI and better represent our culture and reality.” 

Chatbots 

That’s not to say that CPAs, whether French-speaking or not, can’t adapt existing tools to improve their services. Whereas in 2023, chatbots were trained on limited data (which produced a lot of hallucinations), today they provide accurate, transparent responses based on live research. You can even connect Microsoft Copilot 365 to your own databases or download open-source models from the French company Mistral AI. 

“An accounting firm could very well design a chatbot connected to the laws and regulations of Revenu Québec and the Canada Revenue Agency, in addition to its own documents,” explains Blais. “It’s not a complicated process, and it’s a highly effective and targeted tool. This approach is becoming increasingly popular because it works so well. What’s more, you don’t need to retrain your own chatbot; you can simply customize existing models by connecting your own databases to them. This creates an experience for users that’s unique to your organization.” 

Google’s Gemini makes it possible to fine tune models, which involves adapting the bot to respond to specific problems, such as detecting anomalies. “For example, a firm may want to adapt a chatbot into a tool that validates financial information on tax returns, in order to look for errors.” 

In short, chatbots have never been so powerful, and natural language processing (NLP) will make them even more user-friendly. 

Threats and questions 

Questions remain, however, especially for those using online versions. “Generative AI strategies are still very rare within organizations,” explains Olivier Blais, “so employees often use free solutions like ChatGPT, where information is stored on OpenAI servers, creating security vulnerabilities.” 

This practice is all the more worrying given that, according to a survey conducted by KPMG in November 2023, around 22 per cent of Canadians use generative AI for work. Meanwhile, 61 per cent of generative AI users use it several times a week for professional purposes, and 20 per cent do so on a daily basis. 

To achieve this high adoption rate, companies have been quick to humanize their agents, although legislation could force them to be more transparent. The aim is to avoid any risk of confusion—a risk that is very real and will only increase with future advancements, such as OpenAI’s voice-based projects. The more natural the exchanges appear, the more important it will be to know who—or what—you are talking to. 

For example, based on the answers provided, which we can expect will become increasingly sophisticated, what will a firm be able to charge? What will be the value of the advice provided by generative AI compared with that of a CPA? What will be the firm’s liability in the event of a problem? And what about the question of copyright and intellectual property in the responses provided? 

According to Blais, “2024 is the most uncomfortable year in our history in terms of AI, because we’re really on the brink of disruption. The capabilities are incredible, but the laws are slow to follow. And while some people are starting to become proficient in the use of bots, the overall level of incompetence in companies remains dangerous. It’s up to the French-speaking world to seize the opportunity to to make its mark in this field.”