Preparing forecasts. Putting together decks of analysis for C-suite presentations. A relentless stream of deadlines. Last-minute requests for revenue analysis.
Life as a member of a finance team can feel like you’re living out a Greek myth—like Sisyphus, eternally rolling a boulder up a hill, only to watch it roll back down.
While trying to crunch all the numbers and sift through the data, an FP&A professional also needs to make sure that everything is above board—that there are no mistakes, no miscues and no bad calculations that can cause delays.
Traditionally, that’s meant that finance teams need to tread carefully, not only when completing their tasks at hand, but also when adopting new tools such as AI.
Data from Deloitte points to Finance's growing role in ensuring the accuracy of AI outputs: One in five (19%) CFOs say they have the greatest responsibility for AI governance at their organization, ahead of the CEO, boards, and chief risk officers.
This tension between the need to accelerate to meet business demands and the need to ensure the right controls and governance are in place is one that we took seriously when building out Vena AI and its suite of agents.
Vena AI is a tool that finance teams can trust to produce accurate outputs and insights with just a simple natural language prompt, allowing them to maximize their productivity. Vena AI is built on Microsoft’s world-class generative AI technology, Azure Open AI, and is driven by your specific organization’s data.
Here’s how we built it—with finance teams in mind.
“Vena AI allows you and all of your business stakeholders to interact with all your financial and operational data in the most natural format ever, which is human language,” said Andrew Stanbridge, Vena’s VP of Product, when announcing these capabilities for the first time at Vena’s Excelerate Finance 2024 event in Nashville.
“It’s a conversational interface,” he explained, and anyone—even those outside of core finance teams—can interact with it in a very intuitive way. In effect, Vena AI is a tool that acts much like a translator of financial and operational data. It can help tell stories, uncover narratives, decipher the meaning behind financial data and trends and put this information directly into the hands of stakeholders and decision-makers.
Vena AI does this by:
Leveraging best-in-class Microsoft Azure Open AI models, specifically trained to understand hundreds of common FP&A problems.
Allowing you to customize your own AI models based on your business’s unique needs, getting “smarter” over time.
Working seamlessly with Microsoft 365—powering your work within Excel and PowerPoint through Vena's native integration, where you can also interact with your data through Microsoft 356 Copilot.
Keeping your data secure—Vena AI is built on Azure OpenAI, meaning that your data is never used in any way to feed public models—it stays safely within your organization.
In other words, Vena AI makes your finance team’s job a whole lot easier and more efficient, while also ensuring that your business is using generative AI tools securely and responsibly.
Vena AI's intuitive chat interface allows you to use simple prompts to sift through mountains of data, answer questions efficiently and generate reports.
These are tasks that, previously, may have required hours. Now, they can be done in minutes. How did we make it happen? We started with an FP&A-focused rule engine.
“We built a proprietary rules engine that is trained on hundreds of FP&A questions,” said Anton Medvedev, former Product Manager at Vena, when showing off Vena AI at Excelerate Finance 2024.
“But best of all, the rules engine allows you to customize Vena AI to your own needs,” he said. That allows you to add rules to the system, further train it to answer your exact queries or questions related to your line of business and fold in additional dimensions to the tool’s outputs.
For example, it can be helpful for stakeholders outside of the finance team to understand some of the “whys” behind the company’s data. Using Vena AI, it’s possible to add contextual rules to the engine, which can provide more clarity into why, say, seasonal salary adjustments alter a specific quarter’s revenue projections. An FP&A team member may know that immediately, but it may not be as intuitive to someone outside of the finance team.
Additionally, “you can build Vena AI to be intelligent against any type of dataset you’re storing in your Vena CubeFLEX™,” said Andrew. “It’s not limited to just your core financial data.”
Ultimately, our goal in creating Vena AI was to deliver an intelligent Complete Planning assistant that works for everyone in your business, and one that’s easy to implement and use.
Vena AI was designed with four major goals in mind:
Flexibility: The Vena platform’s flexibility is what sets us apart, and what our users have come to expect. As such, we built Vena AI to adapt to your unique use cases.
Adoption: We want everyone, especially those outside of an FP&A team, to be able to interact with their organization’s financial and operational data in a self-sufficient way.
Accessibility: We wanted to make sure that Vena AI could be used seamlessly as a part of almost any workflow. That means it can answer questions about data anywhere in your Vena database, an advantage over other AI-powered tools, which may only be able to answer questions about specific files or reports.
Security: The advent of large-scale generative AI tools has naturally brought up a lot of questions surrounding data security and integrity. Vena AI was designed with enterprise-ready security and compliance capabilities—and can still be implemented without a call to your IT desk.
With those foundations in mind, here’s a look under the hood at the nuts and bolts powering Vena AI’s engine.
First, we started with Azure OpenAI and trained it with hundreds of unique FP&A questions, based on those we see most frequently among the numerous FP&A professionals using Vena. This makes Vena AI able to produce responses that are highly tailored to your FP&A team’s needs and workflows.
The questions and training focused on essential FP&A topics and outputs, including variance analysis, trend analysis, revenue performance analysis and cost analysis. The model is trained to tap into Vena’s CubeFLEX database technology, retrieve your organization’s data, and perform an analysis on top of that data related to a user’s specific prompt.
We also made Vena AI highly trainable and customizable, so that your Vena power users can review past conversations that other users have had with the model to help further train it with additional rules. That’s important because every business’s operations are unique, and so you need to be able to trust that the answers you’re getting are relevant to how your company is orchestrated.
“This allows users to continuously train their model as they use it, over time,” said Anton.
Vena Modelers have complete transparency into how their business is using Vena AI. Pictured here is the view of chat history in Vena AI, helping finance teams validate the accuracy of its results to support the business.
To get Vena AI off the ground, we worked with Adastra, a company on the leading edge of the generative AI revolution—a relationship that was formed through our close partnership with Microsoft.
Vena and Adastra found mutual cultural alignment through Microsoft’s Unicorn program, which helps support startups as they build new, disruptive businesses. We leaned on the Adastra team’s expertise in AI to guide our developers and give the project some initial lift.
Adastra helped review and fine-tune Vena AI’s architecture and prompt engineering, in addition to the large language models. Since Vena AI is built specifically to support finance teams and FP&A professionals, Adastra helped us enhance our large language model (LLM) to help it flag relevant information in users’ prompts or queries, generate follow-up questions, and beef up its data retrieval abilities based on those prompts. They also assisted in integrating chatbot technology into Vena CubeFLEX.
To get even deeper into the weeds, Vena and Adastra used Semantic Kernel technology, which help LLMs run code that can help speed the AI tool’s processing abilities. Ultimately, this improved the performance of the chat interface, allowing users to get the answers they need without excessively long prompts, in turn minimizing operational costs for the project.
Adastra provided benchmarking, consistent measurement of performance and accuracy, said Jasper Chow, a Manager of Software Development at Vena. Adastra also “improved our questions bank for model training, designed and implemented large language model agents and provided best practices in prompt engineering,” Jasper said. “[They also] put in guardrails for doing relevance checks and ambiguity checks, and blocking any abusive language or questions.”
In all, we leaned on the expertise and solutions that our partners provided to help extend your FP&A team’s capabilities without adding extra headcount.
“Collaborating with Vena on this project, we've been able to push the boundaries of what's possible with generative AI, resulting in a tool that not only enhances accuracy and insight but does so with an intuitive ease that's revolutionary for finance teams," said Waleed Hilal, Principal Data Scientist at Adastra Canada
The same Deloitte research we referenced earlier found that balancing pressures to deploy AI quickly while managing potential risks is the top challenge holding CFOs back from deploying enterprise-wide AI (cited by 59%).
For finance teams to use AI to its full potential, they need to be sure it's built on a data foundation they can trust, while keeping their most sensitive information secure. That's exactly what informed our approach to building Vena AI: combining your financial and operational data with the business context behind every decision, while maintaining the security, permissions and control your organization relies on.
Request a demo of Vena AI today.