In the early days of the generative artificial intelligence (GenAI) revolution, users had only a couple of ways to use new services such as ChatGPT. Most often, they had to log on to a specific website or open a dedicated mobile app.

Now, that model is changing. Instead of offering AI as a stand-alone service, tech companies are weaving the new technology into existing applications and services. And the company making some of the biggest changes is the biggest provider of software for finance professionals: Microsoft.

Microsoft is investing heavily in Copilot, a suite of AI-powered features for applications across its portfolio. The upgrades are already rolling out to business users of productivity apps including Excel, Office, and Outlook, as well as for business tools like Power Platform. Early reviews indicate Copilot could be a powerful new force for accountants and other knowledge workers.

“Is it hype or reality? It’s definitely reality. It’s such a game-changer,” said Virginia-based Cherie Gartner, KPMG’s global lead partner for Microsoft, who has helped activate Copilot features for her firm and some of its clients.

The scope and scale of Microsoft’s Copilot effort is a “shocking signal” of the potential the company sees in large language models, the machine-learning technology that underpins GenAI, said Jason Staats, CPA, a former accounting firm head in Portland, Ore., who founded and runs the accounting technology community Realize.

“Long term, there is no aspect of our work that I don’t see being impacted by this, in the very best way,” he said.

NEW AI FLIGHT PATH FOR MICROSOFT

Copilot generally appears as a chat interface within existing apps. But behind that simple interface is the power of a large language model. Here’s how it works:

• The user inputs a question or a request in plain language.
• The Copilot feature analyzes the user’s input based on its general knowledge base and Microsoft cloud data from the user or the user’s organization.
• Copilot responds with information or instructions for the user, or it can execute actions within Microsoft apps to achieve the desired outcome.

Microsoft has built other forms of AI into its apps for years, but Copilot differs because it is based on the same technology that powers ChatGPT, greatly improving its ability to understand natural language and respond appropriately.

Microsoft and its users have described an impressive array of use cases that are already available through Copilot:

• Summarizing transcripts of Teams meetings and other sources. “It can very quickly summarize what we discussed and provide lead-in questions as to what we need to do next,” Gartner said.
• Rearranging the contents of a Word document, such as by building a table that summarizes information from the text.
• Creating new content, including first drafts of emails, presentations, and memos, while drawing on the user’s own data from Microsoft’s cloud for some of the details.
• Stylizing presentations in PowerPoint.
• Looking for trends and analyzing data in Excel.

Business users of Microsoft 365 now have access to Copilot services at an additional cost of $30 per user per month. For a base Microsoft 365 enterprise license, that’s an 83% increase, raising the price from $36 to $66 per user per month.

Microsoft expanded Copilot for Microsoft 365 to businesses of all sizes on Jan. 15. Before that, access was limited to organizations with at least 300 individual users. Among those was KPMG, which started with a few hundred licenses before expanding to a larger deployment of undisclosed scope, Gartner said.

The general public can sample the new features with Windows Copilot, a version of the technology built into Windows 11.

The new AI features could be worth some $10 billion per year for the company by 2026, analysts estimated.

A WORK IN PROGRESS, WITH PROMISE

Microsoft is releasing a steady stream of updates as it fine-tunes Copilot for particular apps, with a long-term goal of allowing the technology to work seamlessly across the company’s ecosystem.

“The functions and the power that Copilot has is really at its infancy. You can see that it can do much more,” said Daniel Rey, chief learning officer for Smarter Consulting, which specializes in business and technology solutions, including training organizations on Microsoft apps.

He’s recently used Copilot features to:

• Create PowerPoint presentations based on the AI’s analysis of his Word documents;
• Condense lengthy training manuals into succinct agendas; and
• Generate a customized template for a computer-assets audit document based on the AI’s contextual knowledge of the topic. In other words, he gave Copilot very basic instructions and conditions, and Copilot built out a custom audit document template.

While some of those abilities are already available through ChatGPT and other freestanding services, Copilot “puts it into the applications I use every day,” Rey said.

The version of the software he has been testing still has some warts — for example, it can’t always directly execute actions within apps. Instead, users sometimes must manually execute the actions suggested by Copilot.

But Rey and early adopters in the financial world say the early results are promising and that organizations large and small should be thinking about whether, when, and how to embrace Microsoft’s AI investment.

TURNING PRACTITIONERS INTO PROBLEM-SOLVERS

“There’s a lot of hunger for this technology because it promises, and it delivers, if you know how to use it,” said Ashley Francis, CPA, who has used Copilot to start automating aspects of her Washington state-based firm, The Francis Group.

Francis is a sole practitioner in tax consulting and planning for ultra-high-net-worth individuals — a niche that isn’t very technologically advanced, she said.

“We’re using technology that was created probably 20 years ago and has only been minimally updated,” she said. “Basically, the planning software that I use now looks exactly the same as it did when I started in the profession.”

In recent years, she had grown interested in the potential for automation to speed up her work. But when she attended a meetup on the topic, she was intimidated by the seeming level of complexity.

“Usually, I feel like I can hold my own in conversations,” she said. “And I was so overwhelmed by how advanced everybody was that I decided that I would never ever, ever attend that session again.”

Still, she decided to try once more with Microsoft Power Automate, a powerful, complex piece of software that had recently gained Copilot functionality.

For her first project, she wanted to use Power Automate to take details from Excel forms, determine the template to use for the engagement letter, populate the template with the appropriate information, adjust parts of the text as necessary, and export a PDF into a project folder — a process that would normally take her about an hour.

It was an “incredibly difficult” project because she could find little information on the web about how to use Power Automate with Word templates.

But she was able to ask Copilot a steady stream of questions that were answered with helpful suggestions about how to order the flows.

“It was like having a tool there that would never judge me,” Francis said.

Francis completed her project after about a month, allowing her to complete engagement letter drafts in mere moments. Then, her social media posts about her success caught Microsoft’s attention, leading the company to feature her in a blog post and to even invite her to Las Vegas, where she spoke on a panel alongside representatives of national brands, including AT&T.

That contrast — a solo practitioner among huge corporations — speaks to the potential of GenAI, she said.

“What it said to me is that this technology is meant for everybody,” said Francis, who was recently nominated for Microsoft’s prestigious MVP designation based on her automation work and who has also launched Kitchen Table Automations, which offers classes and other content on AI and automation for busy professionals new to AI. Rather than relying on developers and programmers for automation upgrades, she said, generative AI “brings this technology down to a level where it’s hitting all of the needs that a group or a company has.”

WEIGHING THE COSTS AND POTENTIAL OF AI UPGRADES

The race to incorporate large language models into business tools is just beginning, and Microsoft’s rivals are hoping to match its advances.

Google has introduced Duet AI, an assistant feature for Gmail, Drive, Slides, and other Workspace apps. Duet shares some of Copilot’s features, such as generating and revising text, creating slideshows, and working with spreadsheets.

Google’s version, like Copilot, will reportedly cost $30 per month per user for large organizations. The competitive landscape between the two titans will become clearer as they introduce and iterate features in the months ahead.

But organizations of all sizes can still start planning for whether — and how — they might use the tools as they become more prevalent. Here are some things to consider when evaluating GenAI applications, including Copilot:

Financial costs and productivity returns

Especially for large organizations, $30 per user per month can add up quickly, but it also can be seen as a token investment for access to cuttingedge technology.

“If I were able to pay $30 a month for an assistant that never slept, was getting smarter all the time, and never asked for a raise, that feels like a bargain,” Rey said. Francis similarly said that, based on her experience with Copilot and Copilot for Microsoft 365, the Outlook tools alone are worth $30 a month.

Calculating the return on that investment may prove difficult, especially as the features are constantly evolving. But it’s worth asking which specific features might be most useful, how much time saved or gain in productivity would be needed to offset the cost, and how the cost of the AI features might increase in the future.

Several recent studies point to the potential upsides. Researchers at MIT found that GPT-3.5, an older version of the GPT model, helped people significantly speed up writing tasks while improving quality. The Nielsen Norman Group found productivity increases for customer support, writing, and programming roles. And a study by the Harvard Business School pointed to improved productivity and quality from consultants working on “realistic consulting tasks.”

On the other hand, it remains to be seen how Microsoft Copilot performs in practice for many users, Staats said. In terms of usefulness and adaptability, the Windows Copilot variant was toward “the back of the pack” of mainstream chat assistants that he had tried, he said in early January.

However, Windows Copilot is one of many implementations across the Microsoft ecosystem. They are in a constant state of improvement, Francis stressed, with growing individual and networked capabilities. To her, the investment seems like an “easy win” for firms.

Data security

Microsoft and Google both insist that they’ve built data and privacy protections into their products. The companies say that they are not using data from the services to train their general AI services — only to power the organization-specific features.

That’s meant to answer common concerns that data provided to AI applications could be leaked to other users, but any emergent technology comes with the risk of data leaks and breaches. A thorough legal and cybersecurity review may help to mitigate the risk, as could guidelines for how employees are to use the new features.

Risk of errors

Despite their promise, large language models are notorious for making factual errors and mathematical mistakes and for ignoring the parameters of a request. If organizations make these tools more accessible through Copilot and Duet, they may increase the risk of AI errors slipping into work products.

Beyond the cost considerations and technical questions, the deployment of technologies such as Copilot and Duet are a sign that GenAI is here to stay. Large language models are gaining a foothold in the most mainstream tools of digital work, and companies need to prepare their workforces, Staats said.

“People are already really overwhelmed at the rate of change and the rate of software change,” he said. “We’re throwing a lot at our staff all at the same time. It’s a big opportunity, but probably the biggest problem, as it has been for a while, is change management.”

For those companies ready to make the leap with Microsoft’s Copilot or with a competitor, the key is to encourage employees to experiment, learn, and share their findings — and to stress that the tools are meant to support them, not replace them. “You need to encourage a culture of curiosity.

That probably means creating some new feedback loops where people are sharing their positive experiences. That’s going to multiply everyone’s learning, just by encouraging a level of transparency,” Staats said. “There absolutely will be new, really weird stuff to navigate, but it’s why you have to keep learning and keep engaging with it.”