More businesses using AI as tools emerge

Simpler software brings more affordable and much faster deployment of artificial intelligence applications.
By John Murawski

Artificial intelligence isn't just for experts anymore.

Technology vendors are racing to introduce streamlined AI solutions for companies that can't afford to hire pricey data scientists and machine learning specialists. These tools can reduce the time it takes to get an AI application up and running from weeks to days, or even hours.

The software automates data preparation and model selection without requiring competency in statistics or computer code. It guides users with conversational language, on-screen graphics, and drag-and-drop interfaces. It ingests business data and generates charts, heat maps, predictions, and business insights. 

"Automation of data science is a priority for the industry," said David Tareen, the head of global product marketing for AI at SAS Institute Inc., the Cary, N.C.-based data analytics company. "The more we can make these tools available to the people who understand the business element, the more insights they can get."

Often dubbed automated machine learning, the technology is bifurcating into two tiers. The more sophisticated versions are designed for data scientists, data engineers, business analysts, and others who are trained in AI or are proficient in data analytics. These platforms automate routine functions to free up the experts for more challenging tasks, such as experimenting with AI models to improve results.

Tools in the second tier are not as robust but offer the promise of making some aspects of automated machine learning available to organizations that don't have the AI or data analytics experts readily available.

The simplification process has moved quickly, producing tools that can be deployed by employees in accounting, audit, finance, sales, marketing, and other operations. These tools require minimal technical support for novices. The evolutionary arc of AI technology is sometimes compared with building websites, which once required knowing computer code but can now be created with templates and graphics.

Automated machine learning platforms are particularly good for making predictions such as risk estimation, inventory forecasting, predictive maintenance, staffing projections, fraud detection, revenue projections, customer attrition, and sales prospecting, among many other use cases, said Kjell Carlsson, Ph.D., a senior analyst at Forrester Research Inc. who has studied and compared the vendors and their products.

Potential impact in the CPA space

With fraud detection, to take one example, the automated machine learning tool can monitor millions of transactions in real time, rather than sampling a small number retroactively. Machine learning models, trained on historical transactions that involved fraud, can flag suspicious activity by spotting clues — such as unusual times, amounts, or locations — that fit a pattern. At this point, a CPA could then investigate the anomalies to determine if fraud indeed occurred, said Troy Fine, CPA/CITP, manager of risk advisory services at Schneider Downs & Co. Inc. in Pittsburgh.

"It's making sure a CPA can provide high-value audits and actually audit areas of high risk," Fine said. "If our opinion can be more accurate and we can be more comfortable with it, then at the end of the day we have a better-quality audit."

Fine warned that CPAs should pilot the technology and continually validate its effectiveness to make sure it produces reliable results across different industries. To remain professionally current, he said, CPAs should be reaching out to vendors in this space and asking if they have CPA firms as clients and requesting introductions to those clients to learn about their experience with the technology.

David Cieslak, CPA/CITP, CGMA, executive vice president and chief cloud officer at RKL eSolutions, said the typical small and midsized CPA practice has not adopted automated AI solutions because the return on investment is not yet apparent for these practices. But he said that "high-effort, low-value" accounting functions are ripe for automation as the profession continues evolving from executing routine services to providing an advisory, strategic role.

AI will bring "more opportunity for better data capture, more accuracy, quicker turnaround, and lower cost," Cieslak said. "This clearly is both an internal practice opportunity as well as an opportunity to monetize that and look for ways to bring further value to our customers."

In a May report, Carlsson wrote that "AutoML solutions are quickly becoming a must-have for every organization looking to scale ML use." He added that this market is expected "to grow substantially as products get better and awareness increases of how these tools fit in the broader data science, ML, and AI landscape."

Gartner Inc. predicted last spring that by 2020, more than 40% of data science tasks would be automated, "resulting in increased productivity and broader usage by citizen data scientists." Gartner applies the term "citizen data scientist" to users whose main job is not in statistics and analytics but who can use automated machine learning to extract predictive and prescriptive insights from data. Gartner further predicts that automation will become so widespread by 2024 that the chronic global shortage of data scientists will no longer hold back businesses from adopting data science.

Auto machine learning tool vendors of note

Many machine learning tools have entered the market, with more on the way. This section looks at a few that have garnered some attention.

In October alone, at least two leading AI vendors introduced automated machine learning tools for general business users. SAS said its updated platform automates functions such as data preparation and data management. The company said its platform explains recommendations and predictions in business terms for citizen data scientists and other users in accounting, HR, legal, and other departments. Inc., a 9-year-old Mountain View, Calif.-based vendor whose motto is "democratizing AI," calls its newest iteration Q. It targets nontechnologist business employees in accounting, finance, sales, marketing, manufacturing, and other departments that sit on troves of rich data ready to be tapped for analytic insights.'s flagship platform, Driverless AI, is designed for data scientists and other data experts, whereas Q comes with a Google-like search interface that lets users ask conversational questions to cull insights from their company's data.

"We want to make every Excel user, every Tableau user, to be able to use AI," said founder and CEO Sri Ambati.

Then there are vendors like Aible Inc., a Foster City, Calif.-based newcomer founded in 2018 to cater to nonspecialists. Aible released its product in March 2019, promising that "business users, regardless of training or data science knowledge, can set relative costs and benefits to create a custom AI in as little as two minutes."

Forrester rated Aible "the best choice for pure businesspeople" but said it's not the best choice for companies looking to maximize the productivity of data scientists who need more robust technology. In August, Aible put out Aible Advanced, which lets data scientists and developers use Python code to train machine learning models.

Aible serves industries such as banking, insurance, manufacturing, health care, and consumer packaged goods, said CEO and founder Arijit Sengupta. Use cases range from lead-scoring of sales prospects to predicting whether an invoice will be paid on time or disputed. Sengupta said the experience of using Aible is comparable to using a chatbot.

Among Aible's customers is Beddr, a Mountain View, Calif.-based health care company that treats chronic sleep problems such as insomnia, sleep apnea, and poor sleep habits. The technology is not set up and operated by novices but by four software engineers at Beddr, said CEO Michael Kisch.

Beddr, the business name for Hancock Medical Inc., has been using Aible for the past year to process patient data about their sleep patterns — such as sleep position, sleep disruptions, heart rate, oxygen saturation rate, and lifestyle factors — and generate recommendations for the patients.

Using sensors attached to the patient, Beddr evaluates sleep problems at a patient's home rather than at a sleep lab staffed by health professionals. The raw data is sent to the cloud, where it integrates with Aible, then bounces back down to the patient's iPhone in a series of recommendations.

"What Aible has allowed us to do is to take data and turn it into insights that drive actions," Kisch said. "Without Aible we'd eventually get there, but we'd have to hire data scientists. Aible helped us get there sooner."

John Murawski is a freelance writer based in North Carolina. To comment on this article or to suggest an idea for another article, contact Jeff Drew, a JofA senior editor, at

Where to find June’s flipbook issue

The Journal of Accountancy is now completely digital. 





Leases standard: Tackling implementation — and beyond

The new accounting standard provides greater transparency but requires wide-ranging data gathering. Learn more by downloading this comprehensive report.