Have you ever seen the movie How to Train Your Dragon? A young Viking boy, who has grown up fearing dragons, meets one in the wild. Over time, he develops an understanding of the creature and learns how to communicate with him, and eventually both overcome their fear of the other and work together to reset the course of their village’s future. Sounds a bit like scary technology reshaping our industry, doesn’t it?
Just like dragons, bots don’t train themselves. Artificial intelligence (AI) and Machine learning (ML) can do wonderful things for your CPA firm, but you need to understand what they are, and what they are not, before the bots can work their magic.
Debunking Myths
Truth: Bots complement an accountant’s job, not overtake it. We know from first-hand experience that when bots are successfully deployed at any size CPA firm, the accountants’ jobs become more valuable. Not only is the data more accurate, but the people have more time to focus on their clients and the big-picture, critical-thinking issues.
AI has been proven to cut overhead costs for accounting firms in half. And because it’s scalable, you can grow your firm 10 times the existing size for the same price. AI powers a more organized accounting firm, and it gives you back the only non-renewable resource that matters: time.
Bots in Action
Bots are a form of automation, a task that technology performs automatically. Many CPA firms are already using automation in bookkeeping, like creating rules in QuickBooks Online—these rules are a form of automation. The rules are specific to a client and account category, so if something changes within that client account, like the spelling of the name or an invoice number that’s out of order, the rule breaks, and the accountant has to look at what’s going on.
The limitation in that example is that the automation stops at that particular account.
Bots and ML take that same concept and apply it to the entire bookkeeping system. Instead of working within a single account, we’re now looking at them all for small variations and dynamic rules.
Here’s the thing, though. Bots don’t train themselves.
How to Train Your Bots
Bots require lines and lines of code and algorithms to work. Our engineers work with the CPAs and accountants to identify patterns in transactions and write code that implements those patterns into rules. There are a couple easy examples of this.
One is that every time you see an airline in a client’s account, you know it’s a travel expense. Or if you see the name of a hotel, it’s a lodging expense; if you see the electric bill, you know it’s a utility, and so on. You can give those rules to ML and the software will automatically recognize those expenses, thousands of times, without any human interaction.
We can teach machines how to read client history, find patterns and make predictions, just like an accountant or bookkeeper would. Only unlike an accountant or bookkeeper who can only retain and remember so much, Botkeeper and its ML are capable of remembering and retaining an infinite amount of data as well as seeing and identifying patterns in data sets that would be unrecognizable to humans.
The more data you give to bots, the smarter and more accurate they become.
Bots work by connecting to your firm’s accounting software, like QuickBooks Online or Xero, and tackling everything from payroll to categorizing transactions in client books and records. Bots can handle anything that’s a repetitive task based on patterns. Smart bots are given lots of data and many rules so they can easily interpret, analyze and categorize transactions. This happens at the beginning of every Botkeeper engagement, when the Botkeeper Growth team works with the CPA firm and asks dozens of questions so engineers can write the necessary code for the bots to work.
Expect lots of “yes” or “no” questions in the process of training bots. These fundamental questions model transaction similarity or dissimilarity and are the basis of Botkeeper’s platform. We use a process called ensemble learning to apply probabilistic mathematical models to ML, which then creates and trains the bots.
The result is a tailored workflow for each client, custom controls, and an automated platform for remotely managing all your client’s bookkeeping needs.
What about the nuances of individual clients?
Bots can do a lot of things, but we all know that each client is unique. We build performance models according to each client so that the bots know what to look for in different accounts. If a bot is underperforming and returning more gray dots—indicating a confidence level of less than 90%—then engineers retool it and add more non-ML tools to the predictive analysis until the automation is completely reliable. One of these tools is Zapier, which is a type of middleware integration software that automates repetitive tasks.
Gain Back Time
One of the expectations that accountants might have is that bots are ready to go the minute you deploy them, or that they’ll work perfectly the first time. This isn’t necessarily the case, and I don’t want you to be discouraged. Just like it takes time to hire and train a new team member, it takes time to train and retrain the bots. You’ll have a support team to help you through this entire process, and they’re just a phone call or email away, so you’re definitely not going to be out there by yourself trying to figure this stuff out.
The magical part happens, the part when you know the bots are working, is when you’re sitting around with extra time on your hands and you have no idea where it came from! This is our goal, and it could be your reality sooner than you think.
You can read more about training and deploying bots to scale your CPA firm in Chapter 2 of Botkeeper For Dummies.
