News Analysis
Microsoft continues to roll out more generative AI capabilities across its software offerings with the debut of Copilot for Finance. The company stated the AI service, which is launching as a public preview, will make it easier for workers to sort data and find errors in financial reports.
Microsoft Copilot Goes Vertical
Since Microsoft invested an estimated $13 billion OpenAI, it has undertaken a breathtaking development cycle using GPT to bring generative AI to as many of its products as possible, especially its business offerings, through Microsoft Copilots. Already, it is offering Copilots for Microsoft 365 for its Sales offering, for its Service offering Copilot in Viva, and of course Github Copilot.
But the debut of Copilot for Finance shows the company starting to target specific verticals with custom-built Copilots. While Copilot for Sales gave a taste of this approach, with its integrations between Copilot for Microsoft 365 and customer relationship management (CRM) platforms, this isn't necessarily for a single vertical.
Copilot for Finance, however, takes the logic of creating a Copilot for a single vertical and runs with it.
Launched in public preview on Feb. 29, Microsoft claims the latest Copilot can perform several common role-specific actions in Excel and Outlook. It also states it has resulted in cost-savings in its own accounts departments. Charles Lamanna, CVP for business applications and platform at Microsoft, described the specific finance-targeted features in the release in a blog post. These features include the ability to:
- Conduct variance analysis in Excel using natural language.
- Carry out reconciliation processes in Excel.
- Create summaries of relevant customer account details in Outlook.
- Turn Excel data into presentation-ready visuals and reports.
Lamanna wrote of the release: "The Copilot offerings designed for business functions help workers tackle a common problem: getting from insights to impact — with the relevant data and workflows specific to their roles."
The writing is on the wall – or in the blog as the case may be: Microsoft is steering Copilot towards specific verticals and roles.
Related Article: Microsoft 365 Copilot Is Now on General Release. Are Your Permissions in Order?
Microsoft's Ambitions for Copilot
Microsoft's move into specific verticals shouldn't come as a surprise. The company has been clear about its ambitions to push Copilot into every corner of the digital workspace. So what is happening here?
Siddharth Kashiramka, a generative AI product leader at Amazon, called the three areas Copilot for Finance currently focuses on only the beginning. Those areas — variance analysis, which looks at the differences between financial forecasts and performance; collections for resolving delinquent accounts; and invoice matching and reconciliation — are a start, but he wants to see how Copilot will manage fraud detection by analyzing transaction patterns to identify suspicious activities, as well as its ability to streamline compliance requirements, a major pain point for financial organizations.
“Like any other Gen AI product, it is still early days. There will be a lot of work and learning required by organizations to adopt a product like this to realize real value,” Kashiramka said. Customers should take several considerations into account here, including:
1. Checking the model
As envisioned by Copilot, the product will integrate with Excel, SAP or other systems to inform its results. But understanding how generative AI arrives at its conclusions is crucial for building trust within the industry and with regulators. What if the data is incomplete? Can a model explain the reason behind the recommendation and outputs?
2. Data privacy
This, he said, is relevant for financial institutions that manage sensitive data. How would Copilot ensure protection against potential vulnerabilities arising from Gen AI integration, such as manipulation of data used for training or generated outputs?
From a wider perspective, though, Microsoft is the trend-maker, Kashiramka said. It is setting the pace for other large tech firms to follow. However, it is too early to say, especially in regulated domains such as finance and or healthcare.
"The regulatory landscape surrounding Gen AI is still evolving, and companies might be wary of investing heavily in this technology until regulatory clarity emerges," he said. "We have seen the impact of Google's foray into Gen AI image already. I think this step of Microsoft could still play a role in increasing overall interest in Gen AI and influencing competition, but it is too early to say.”
Specialized generative AI solutions will coexist with general AI solutions — not to be confused with general AI — catering to different needs and applications, he continued. In fact, the quick evolution of the field means that sometime in the future, the lines between the two will be very blurred where the models might have the ability to adapt themselves to different domains.
Related Article: High Cost Is a Barrier for Generative AI Use, But Not for Long
Niche AI Tools for the Workplace: Pros and Cons
Elixirr Consulting's Iliya Rybchin sees Copilot for Finance as a continuation of a growing trend. He notes the number of specialized solutions already available that focus on narrow problem sets, as Copilot for Finance does. This will only continue, he added.
Rather than rely on custom solutions or figure out how to integrate OpenAi into their tools, companies like Microsoft provide solutions integrated into existing tools and workflows, he said. "They are starting with Excel and Finance, but as you can imagine there are other use cases that we are likely to see soon," Rybchin said.
"Beyond Microsoft, we are seeing examples of this at large software and SaaS providers like Salesforce as well as tons of startups that are offering very niche and focused solutions that solve a very narrow problem set.”
Specialization of this kind does bring some potential disadvantages too, particularly in industries like finance or law where sensitive financial or personal information is involved. The cost can also be a hurdle for many, which includes the cost of the software itself as well as the potential need to adapt infrastructure to integrate such tools.
Related Article: A Look at the Large Language Model Landscape
Business-Specific vs. Generic
While finance is fertile ground for LLMs, plenty of other tasks can benefit from generated code for data analysis, and from the types of common sense that LLM uses to provide results, said Frame.ai CEO and founder George Davis. However, it is not necessarily a recipe for success.
He said what each organization achieves will depend in part on how accessible their data is to the LLM and how well this new product can build trust and incorporate feedback.
So far, Microsoft has shown a strong understanding of enterprise privacy and security goals by building LLM products that fit the same data governance standards as the rest of Azure, Davis continued. However, he believes that the remaining trust standards boil down to the LLMs use. “Microsoft has so far straddled between proprietary LLMs (like OpenAI) and their own contributions to open AI research,” he said. “I think leaning into the latter will give customers more confidence in the strengths and weaknesses of the AI they use.”
Webinar
Sep
25
CMS Briefing: A Live Look at What’s Next in AI-Driven Platforms
Learn how leading organizations are using AI‑driven tools to publish faster, personalize smarter and stay secure.
Register
Webinar
On demand
Ready or Not: How Data-First Organizations Are Unlocking Agentforce Potential
Learn how to cut through the noise, activate Agentforce and build a Salesforce AI strategy that actually delivers.
Watch Now
Webinar
On demand
AI in Customer Service: Faster Resolutions, Happier Customers
Don’t let rising demand burn out your team. See how to build a smarter, more resilient support org.
Watch Now
Webinar
On demand
From Hype to High-Impact CX Strategies That Actually Scale
Turn buzzworthy AI and outsourcing trends into measurable CX wins with fresh 2025 data.
Watch Now
Webinar
On demand
Insights to Action Rethinking the Contact Center for Real Business Impact
Join our exclusive webinar to hear CX executives share their innovative strategies for transforming service delivery.
Watch Now
Webinar
On demand
Accelerating Healthcare Ops with AI
Sign up to see how your peers are putting AI to work — and making it stick.
Watch Now
While he sees specialization as a winning strategy in biology, business and software, Davis expects enterprises will benefit the most from specialization in their specific business. "Even the smartest accountant must learn a lot about each company they analyze," he said.
Compare a generative AI that knows the specific products, markets and policies of a business as opposed to a prefabricated AI that is a specialist in textbook accounting — which do you think would be more effective?
“The use cases described for Copilot are both cool and powerful, but they're reactive,” he said. “Once finance identifies a need, they can delegate it to the AI instead of an outsourcer. I think the biggest wins for generative AI will be in proactive use cases, where AI is able to identify opportunities or risks that would be otherwise missed.”
Specialization Playbook
Copilot for Finance is part of Microsoft's broader strategy to create tailored AI solutions for different verticals, a strategy other companies will likely adopt, said Mento VP of engineering Kevin Ball. "It is likely that we will see other tech giants and startups following suit, as the demand for industry-specific AI tools grows. The trend towards specialist Gen AI solutions is clear, and we can expect this to be a key direction for the future of generative AI."
He said a common playbook is being used across the industry. “I am seeing a trend in generative AI to move beyond broad novelty-based consumer products like chatGPT, companies are following a playbook that has three elements." Those elements are:
1. General purpose
Start from a general purpose "foundation model" (examples include GPT4 from OpenAI, Claude from Anthropic, Llama 2 from Meta, or Gemini from Google) that has a lot of high-level capabilities for manipulating language, photos and/or video.
2. Domain-specific knowledge
Layer in domain-specific knowledge via some combination of fine-tuning (training the model further on examples), prompt tuning (writing very specific prompts) and context loading (searching and surfacing information from a combination of proprietary and public documents).
3. Connections
Connect the model to domain-specific non-AI functionality, such as (in the Microsoft case) financial modeling tools, analysis tools or other types of software integrations.
Ball said the most successful examples to date have been coding assistants (Github Copilot as the trendsetter that also started the "copilot" branding for Microsoft) and meeting note taking and summarization tools, but this approach is being applied to a variety of domains including legal (example LexisNexis), customer support (example Intercom), and more.