Generative AI, also known as gen AI, has the potential to provide enhanced insights, particularly by using plain language to delve into the details. However, there are challenges that we must overcome before this technology can work reliably, such as addressing data governance and gen AI's capacity to hallucinate.

We introduce these topics in part 1 and part 2 of this series. In brief, AI needs good data to do its job, yet there are legitimate concerns about protecting that data from misuse and leakage. Businesses are reluctant to harness AI because they don't want sensitive data to be shared inappropriately, such as being added to a database used to train AI models. Hallucinations, which is a gen AI quirk to confidently yet unintentionally fabricate answers, are another big concern. How can you trust what the AI says as accurate?

Such problems are considerably bigger in areas that require high accuracy and regulatory compliance, such as finances. Yet, the potential for gen AI to be a sparring partner that helps people analyse complex financial information, quickly understand spending patterns, and identify areas for cost optimisation—that is a revolutionary prospect.

It's been 1Nebula's quest to make these capabilities a reality for customers using our OneView financial management platform. Let's take a look at how we're achieving this goal. 

Tackling data governance

Data security, privacy, and sovereignty are thorny issues for companies. The risks associated with data leakage and mismanagement are manifold. Poor management can lead to regulatory fines, reputational damage, and reduced competitive advantages.

These problems are not new, but gen AI makes them impossible to ignore. There is a chance that employees could feed sensitive information into public AI services. Many businesses have already banned this practice. Yet, gen AI tools are very useful, so we sought a middle ground for data management and governance.

Our solution is to host a private gen AI model within the Microsoft Azure cloud service. This approach establishes boundaries or guardrails that ensure a customer's data stays within that private space. The AI can process the data without risk of leaking it to other databases, and 1Nebula ensures the customer data is never used to train the model. All our AI does is contextualise the data and provide accurate answers.

Another benefit is that we control the private area's governance and regulation requirements, ensuring it meets necessary industry standards.

Increasing accuracy 

Data governance, though, was the easy challenge. Getting gen AI to provide accurate and reliable information is a much tougher task. 1Nebula takes an approach here that may seem counterintuitive to how much of the market views gen AI.

To a layperson, gen AI seems intuitively intelligent. It can converse in plain language, grasp ideas, and respond with creative insights. Yet, this is an illusion. Instead, gen AI is simply a high-speed mimic that can confidently structure words into coherent sentences.

But financial data needs clarity and accuracy derived from understanding underlying financial principles and processes. So, we took a different tact. Our gen AI acts as the interface, which passes instructions to a library of plugins we custom-coded to match various financial requests. These plugins interact with the financial data, producing results that the gen AI then articulates to the user, complete with citations that users can track back to the source data.

To ensure a fluid and intuitive user experience and to stop the AI from branching into unrelated areas, we train it to specialise in financial queries and financial operations standards. We also have humans who continually analyse the AI to improve its feedback

Human in the loop

Does the AI replace people? No, not at all. The goal of our gen AI is to add a powerful companion to OneView that will help our customers get answers and insights faster.

Financial management is demanding. For example, IT professionals will tell you that forecasting and managing cloud costs is time-consuming and convoluted. Pouring over spreadsheets, even granular dashboards, is tedious and draining. But what if they can ask questions and start to see patterns emerge?

Gen AI has shortcomings. Yet, with the right guidance, guardrails, and supporting processes, it can indulge elaborate and deep queries that users express in plain language. The gen AI revolution is an interface revolution. It opens up the possibilities to delve into deep and complex data without needing a data science degree, programming skills, or huge amounts of time.

We envision this level of engagement for our OneView customers, reducing the time and friction between complex financial questions and answers. It's been a challenging development journey reaching this point, but our gen AI is currently in private preview with select clients, and we aim to release it soon as part of OneView.

But what is the real potential of our gen AI collaborator? In part 4 of this series, we'll look at use cases and practical examples, establishing that human-AI collaboration truly is the future of fast, efficient, and effective financial management. 

Read part 1 and part 2 of this series in case you missed them.