The forgotten step to AI project success

A few weeks ago, a WSJ article discussed the reality of artificial intelligence (AI) project failures or delays that rarely make the news. It’s no surprise that some AI projects fail. We know that not all projects go as planned and that we learn from the experience, adjust and apply those lessons to the next project.

Why AI fails to fulfill its promise

AI and machine learning have received a lot of attention. These technologies can greatly enhance business processes by analyzing data at a volume and speed that have been previously unavailable. While AI-supported data analytics can provide greater insights due to a wider base of data analyzed, AI-supported technology is also used to streamline some business processes including order fulfillment, online benefits registration, purchase order generation and other tasks that require an automated process that connects multiple applications.

The WSJ articles points to two reasons AI projects may not be successful: quality of data that feeds the process and the unanticipated cost, time and effort required to prepare data for use. This is a critical point because an automated, AI and machine-learning-assisted program is only as good as the data that it can access.

In a survey of technology leaders, 42% cited a lack of quality, unbiased data as their greatest barrier to AI adoption. These are experts in their fields who understand the potential value of AI and machine learning, but also understand that they don’t have the data quality that will provide a good return on investment for an AI implementation.

The ICD-10 coding system

Here’s a great example of the importance of quality data from the healthcare industry. A few years ago, the International Classification of Diseases, Ninth Revision (ICD-9) coding system used for medical claims was replaced by the more robust, more specific ICD-10 system. ICD-10 provides a higher level of specificity that includes diagnoses, symptoms, site, severity and treatments. Health providers were given the option to use simpler “unspecified” codes during a one-year period after implementation as they learned and become accustomed to the more complex system.

This one year of leniency resulted in many providers defaulting to the simpler, unspecified ICD-9 codes rather than using the more accurate ICD-10 codes – and their automated claims submissions programs reflected the less specific data. The result was an increase in claim denials, which meant more work for the providers who had to retroactively collect supporting documentation to appeal the denial or face loss of revenue. If the claims had initially been submitted with the more accurate – although a bit more complex – data, the extra work wouldn’t be necessary.

Although this is a healthcare example, the need for clean, accurate data upfront is the same for any AI-supported automation in any industry.

Automating business processes

Automating some business processes is simpler than others. At OpenText™, when we enable different companies to collaborate in a business process by exchanging data for a specific purpose, the overall goal is clearly defined and we have enough context about the data to define it as part of a transaction. For example, a purchase order must have specific information to meet industry standards or an insurance company application must include standard information to identify the covered party. These parameters do not differ greatly from business to business.

This becomes a bit more complex with application-to-application, enterprise application integration and or even medical research projects. In these cases, the data in the applications is organized for specialized purposes and is defined by the assumptions of that application or purpose.

As data is brought into business processes, the experts of that data share or simplify the assumptions. That’s a big part of data transformation – taking data with one set of assumptions, shaping it and making it usable for other purposes by giving it a recognizable “face” that allows OpenText™ Contivo’s AI-assisted automated process to pick and choose data needed for the task.

This process requires expertise in AI, integration and data management, yet when asked about the status of their staff resources for design and implementation of AI-related projects, 45% of technology leaders reported that their organizations lacked the personnel needed. Outside providers with proven experience provide time- and cost-savings by guiding the process and overseeing the implementation.

Make your AI project a success

In spite of the stories about barriers to AI adoption in business, adding AI and machine learning technology to your business processes can enable greater productivity, profitability and growth. A successful AI project begins with ensuring your data is right for the processes that you are automating and includes a team that has the AI expertise and integration expertise needed for a holistic understanding of how data, AI and your business processes must interact to meet your organization’s goals.

Where are you in the AI adoption journey?

For more information about OpenText services that can help your company overcome the data transformation barriers preventing successful AI adoption, visit our website.

Related Articles

Close