Bookkeeping and AI: Why So Little Progress

In the fall of 2020, I was introduced by Dr. Bill McCarthy to Tom Hall as someone who was giving a lot of thought to how to make the bookkeeping and accounting processes more efficient, and useful. Tom had written a paper based upon his very practical experience with accounting and bookkeeping evaluating how could the information we attempt to gather be made and used more effectively.

The introduction and paper resulted in a series of three discussions exploring the topic. We recorded our conversations. A summary of each is presented shown here:

Session 1

In this episode Tom explains the basic importance of an Entity, whether that is a person, or partnership, a corporation or something else. The intent of the financial recording process is to make information–accurate information–about the entity accessible.

Tom uses a set of graphics to explain the concepts embedded in his paper. The discussion points out the issues of:

  • Determining beforehand what information is desired of the model;
  • How those decisions can be off from the kinds of questions we should be asking the data,
  • What information the historical accounting model does not capture that are commonly needed in today’s world,
  • How do we appropriately summarize activity over time, to efficiently process and understand it.

Here is the full video lightly edited for understandability.

Session 2

In Session 2 Tom goes farther into points covered in the prior session. Kip talked about the concepts of Metric Engine, and how it would compare to a search engine. Search engines show individual instances, but accounting systems attempt to give a status over time, present the aggregated results of the activity.

Tom points out that the premise of recording data can affect the information that can be retrieved from the data. For example, entries which simply true up balances can obscure a lot of individual activity, which then reduces the value of the data to represent what really happened; in other words journal entries can be constructed in such a way that nearly unrelated activity are reflected as happening at the same time.

Kip points out how use of a key identifier in sub-ledgers, and how the loss of those views in the enterprise view, reduces the usefulness of our overall accounting data environments. Maintaining balances at a customer or vendor/contract levels would increase the data available from the cost of the accounting system without increasing the overall costs of the system.

Session 3

Session 3 begins with discussion of why more vendors of accounting software not focused on this problem; what are the impediments to use of machine learning and AI in accounting systems. In some respects the early automation of financial systems, which provided huge savings from doing things manually, meant that the cost savings are less in replacement of existing systems which have been more sensitive over time, not less.

Tom gives examples of types of things that at times are not captured by the historical accounting system, and how it would allow us to understand the world more effectively. They also discuss the future, and how to progress the discussion further, how transparency and ethics are critical to making changes.

The session ends with Tom going over each of his formulas he developed in his paper, showing what the variables are attempting to capture and measure and expressing the relationships between them which starts to create a framework by which automation could be used to enhance the historical bookkeeping processes.

Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out /  Change )

Google photo

You are commenting using your Google account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s