Part 2: Addressing generative AI challenges
In our first post in this series, we introduced why generative artificial intelligence (gen AI) is a potential game changer for telecom expense management software in an enterprise setting. We also noted that the technology needs to be carefully implemented in TEM solutions to ensure accurate outputs and full compliance with data privacy and security requirements.
Among the major challenges that we identified in leveraging gen AI in an enterprise solution include the fact that gen AI is designed to process language rather than maths; the reality of ‘hallucinations’ (false, misleading, or nonsensical outputs); data quality challenges; and data privacy concerns.
The good news is that these challenges can be managed through a framework for ethical, responsible A: disciplined data governance and a rigorous focus on data quality; and appropriate human oversight. These are all important conceptual underpinnings we have taken into account in the design of our AI FinOps Advisor chatbot.
With that in mind, let’s consider some of the best practices TEM software companies and enterprises can follow to get the best results from gen AI.
Leverage specialist financial solutions
We mentioned in our previous blog post that large language models (LLMs) driving gen AI solutions like ChatGPT are known to make mathematical errors when doing financial sums. Our recommendation is that companies should not rely on gen AI to do the actual math in expense management automation applications where financial data integrity is vital.
Instead, use generative AI for what it’s good at—answering natural language queries—and use general financial tools for computations. An advanced expense management solution will allow you to use natural language prompts to surface information and insights, but at the same time, will use robust, verified algorithms specifically designed for financial calculations for the sums.
Look to custom data models
Technology expense management automation is a specialist field, so the results you get from an all-purpose LLM and gen AI solution might not be optimal. It’s thus preferable to work with a solution that can be fed with data relevant to your business, so that the algorithms are trained to produce relevant and useful outputs.
One challenge companies might face is that their data isn’t rich or high-volume enough to help the AI model make more accurate predictions across different contexts. We would recommend using a solution trained on a wealth of relevant industry data models—another reason to opt for a specialised platform.
Clean up your data act
The performance of LLMs and gen AI in expense management automation applications depends a great deal on the quality of the training data. It’s important to Implement rigorous data cleansing processes to ensure your AI-driven applications are using high-quality, accurate data. This includes removing duplicates, correcting errors, and filling in missing values.
You should use diverse datasets that represent various scenarios and conditions in telecom expenses. It’s also key to continuously monitor data quality and promptly address issues such as corruption or outdated information. Leading companies will make a habit of auditing their data to ensure it remains clean, relevant and reliable.
Understand how your platforms process, store and manage your proprietary data
One of the major risks around using gen AI in an enterprise setting relates to the confidentiality and security of your proprietary data. It’s essential to remain compliant with regulations such as the Protection of Personal Information Act and the Global Data Protection Regulation when you expose any of your employees’ or customers’ data.
As such, it’s important to understand how many tools you use will process any data you share with them as well as how long they will store and how they will use it, for example, to train their own models. It’s best to opt for a solution with secure APIs that will allow you to keep sensitive data private by anonymising it or processing it locally.
Make sure any systems you use and models you build are transparent and explainable. The system should be able to provides clear explanations for its recommendations and decisions, including how it arrived at its output. This builds trust and helps users understand the AI's reasoning process.
Keep humans in the loop
AI platforms are maturing all the time, with increasingly advanced monitoring, contextual analysis and anomaly detection. However, human oversight remains important and probably always will. Forward thinking companies will implement feedback loops to check AI predictions and recommendations against actual outcomes.
Human experts can review flagged data to ensure its integrity before it is used in decision-making processes in expense management automation. They also play a key role in assessing whether AI is used in a responsible way and flagging potential ethical and data security concerns.
Conclusion
Generative AI will elevate technology expense management automation to a whole new level. Not only will it help reduce manual tasks and streamline processes, it promises to put powerful new predictive analytics insights at your fingertips. We’re using gen AI to build powerful new expense management and FinOps capabilities into our software.
Our AI FinOps Advisor is a co-pilot that connects to the same data sources and infrastructure as our FinOps and telecom management software. We believe this technology will elevate automation as well as make it easier for enterprises to access personalised insights into their technology spending.
Our next post will take a closer look at our solution and some powerful use cases of the future, such as better budgeting, enhanced anomaly detection, customisable workflows, and more. In the interim, if you are looking for a partner in FinOps and expense management automation, get in touch. We can help you simplify operations and accelerate your cloud journey.
Read part 1 of this series