Unleashing AI in Accounting: Cooking up Success with Data-Driven Programming

But how specifically is AI different than previous technology advances?

To explore the idea, we can use a simple, relatable metaphor—a cook preparing a meal. Classical programming is similar to a cook following a pre-determined recipe, while data-driven programming—which underlies current advancements in AI—resembles an innovative chef creating new dishes based on the ingredients available to them.

Breaking Down Classical Programming

Classical programming, a fundamental approach that has been the bedrock of software development for decades, can be thought of as a traditional cook who follows a recipe. This recipe provides a pre-determined set of ingredients and a step-by-step method to follow.

In this metaphor, the ingredients represent the input data and the recipe represents the program written by the developer. The developer designs a specific program or algorithm to solve a specific problem, just as the recipe instructs the cook on how to create a specific dish.

In many fields, including accounting, classical programming has been implemented to create critical software applications. These applications have provided solutions for various tasks, such as calculating taxes, preparing financial statements, or automating transactions between systems. Like the pre-determined recipes, these programs are very effective, but they may lack flexibility in handling new or unforeseen situations.

Understanding AI and Data-Driven Programming

Now, let's consider a different approach to cooking and, by extension, to programming. A master chef might look at what ingredients are available, think about what flavors and textures work well together, and devise a completely new recipe to suit those ingredients. In the world of programming, this is akin to the data-driven programming that makes AI possible.

Data-driven programming leverages methods that can be shaped by data, such as machine learning, neural networks, and predictive analytics. It uses available data (the ingredients) to inform the creation of an algorithm (the recipe). Just as a chef may tweak the recipe based on the ingredients at hand, the algorithm adjusts based on the data it processes.

Taking this analogy further, machine learning is like a chef who, after creating a new recipe based on available ingredients, tries that recipe with all different combinations of ingredients and tastes them to determine if the recipe is suited for broader applications. This is similar to the iterative process of training a machine learning model, where a preliminary algorithm is developed based on available data. The model is then tested with different datasets to evaluate its effectiveness. The model also can be further trained with different combinations of data to improve its performance, akin to adjusting a recipe based on the taste outcomes.

This approach can offer tremendous flexibility and the ability to make predictions, providing broader applications than classical programming. For example, as an AI model learns and adapts to data, it can identify patterns and make predictions that are beyond the capacity of traditional programming techniques.

The strengths of classical programming lie in its reliability and predictability, like a well-tried recipe. Generally, accountants don’t want guesswork involved in grouping a chart of accounts for financial statements. The correct classification should be deterministic, and classical programming solves that problem.

However, the rigidity and lack of adaptability of classical programming may be a disadvantage when dealing with varied and complex data, such as classifying previously unseen transactions on credit card statements or identifying new segments of customers. Data-driven programming brings flexibility and the ability to handle new data and tasks—as well as the ability to learn from new data and apply those learnings going forward.

How AI Will Help Unlock Your Organization’s Data

Applying AI and data-driven programming will unlock new value from data for organizations. Organizations typically deal with a vast amount of data, ranging from highly structured data to semi-structured and unstructured data, and the data are stored across of a variety of sources.

Extracting value from unstructured and semi-structured data sources across the organization is challenging because of the diversity of formats and sources, the lack of standard formats and structures, the lack of context around this data, the volume of data, and issues with quality of the data.

Imagine how much time and effort it would take you to attempt to organize and summarize all of the knowledge embedded in the following data at your organization:

• Email Communications: Emails contain a wealth of information, but they're largely unstructured. Important details about client interactions, project statuses, and even financial figures are frequently exchanged via email. They may be stored in individual mailboxes, on company servers, or in the cloud through providers like Microsoft Office 365 or Google Workspace.

• Financial Statements and Reports: These documents might be semi-structured, as they contain structured data (numbers, categories) but may be stored in a format that doesn't allow easy data extraction, like PDF or scanned images.

• Invoices and Receipts: These are often semi-structured, with some information (like item names, quantities, prices) in a predictable format, and other information (like vendor or customer details) less structured. They might be stored in paper form, as scanned images, or in various digital formats across different personal computers and servers.

• Meeting Minutes and Notes: These are typically unstructured text documents, which may include important financial details or decisions. They might be stored in shared drives, note-taking apps, or document management systems.

• Spreadsheets: Accountants use spreadsheets extensively, but the data within them can be quite unstructured. While the cells in a spreadsheet are organized in a grid, the meaning and relationships of the data can be unclear without context. Spreadsheets might be stored on individual workstations, shared drives, or in cloud-based applications like Google Sheets or Microsoft OneDrive.

• CRM and ERP System Data: These systems can generate a large amount of semi-structured data. While much of it is stored in structured databases, certain components like customer or supplier notes, document uploads, and system logs can be less structured.

With data-driven programming, accountants might more readily gain insights from their organization’s data and unlock its embedded knowledge.

AI could help to discover, organize, relate, and summarize your organization’s data, allowing professionals to query that data in natural language and receive meaningful responses. Solutions like Informatica’s IDMC, Alation’s Data Catalog, and IBM’s Watson Discovery target this problem, and their capabilities are improving with advancements in AI.

Conclusion

In conclusion, as data becomes an increasingly accessible asset, the ability to harness it effectively will become crucial. AI and data-driven programming provide an exciting pathway for the future of accounting.

By adopting these approaches, accountants can leverage the potential of their data, just like an innovative chef creating masterpieces from the ingredients available to them. This new era of accounting offers enhanced efficiency, accuracy, and strategic insight.