The AI Adoption Barrier: Key Challenges and How To Overcome Them

Artificial Intelligence (AI) has become quite important in recent years to drive up business performance.

Enterprises across various industries have started adopting AI to increase efficiency and productivity.  

However, several hurdles lie in our path to full AI adoption. 

In this article, we will discuss some of the several hurdles that lie in our path to full AI adoption.

Challenge 1: Data Quality and Availability  

One of the most significant challenges that enterprises face while implementing AI is the quality and availability of data. The models require large amounts of data to be trained. However, the data must be clean, accurate, and relevant to the problem at hand. Poor data quality and availability can significantly impact the effectiveness of AI models.  

For example, one of our consumer clients was facing a lot of challenges in processing incoming data from their clients. Every one of their clients operated slightly differently. The traditional approach to staging the data, mapping the data, and then running ETLs is a significant manual effort. It often takes several weeks to onboard new or updated data feeds. 

So, as a complement to the traditional approaches, we decided to create an AI-based solution to this problem. We had positive early success, but one of the problems we ran into quickly was the lack of sufficient data to train the AI model. We also encountered data privacy and security issues that needed us to mask the data, which required multiple iterations and took up valuable time. Second, there wasn’t enough data to make sure the model would handle exceptional scenarios properly. While we were able to handle this scenario by generating reasonably high-quality mock data for other more operational business processes, that may not have been an optimal solution. 

It was clear that AI was an ideal solution, but also apparent why impactful AI initiatives fail to see the light of day. 

Some ways to mitigate this challenge are: 

• Ensure that AI is top of mind so that we are conscious of keeping data available even though there isn’t an immediate need. There are tradeoffs here to consider 
• Maintain a high-level enterprise data model that enables us to brainstorm and implement the right data augmentation techniques quickly 
• Collaborate with partners to maintain clean and masked data repositories — an industry consortium, in this case, would have helped a lot 

Challenge 2: Lack of Understanding and Alignment  

Another significant challenge faced by enterprises is the lack of understanding and alignment between the business and technical teams.  

For example, as Google sunsets universal analytics and moves to GA4 in June, one of the biggest challenges is in making the historical data available in the new GA4 dashboards for continuity of analysis. The reason for this challenge is the basic mismatch in the way activity is tracked. Google, like many others, is moving away from session-based to event-driven tracking. 

If they were not early adopters, enterprises would now have to maintain 2 different dashboards for their GA3 and GA4 data.  However, since we understood the marketing data in detail, we were able to devise a solution that effectively allowed the GA3 and GA4 data to be available in a single dashboard.  

We were also able to integrate predictive analytics and what-if scenario modeling into the Google dashboards for clients.   

However, a sufficient gap exists today between technology and business teams that most enterprises will not deploy such a solution that presents itself quite naturally to us.  

Some ways to address this challenge are: 

• Provide training and education to technology teams on how the business works...especially when there are external SaaS applications that the business uses.  
• Provide an understanding to the business in simpler terms about what AI and other technologies mean for them 
• Do design thinking-led brainstorming for a project so that processes and requirements can be understood without getting muddied by tech and business jargon.  

Challenge 3: Operationalization of AI Insights (Last Mile Problem) 

While AI models generate valuable insights for enterprises, operationalizing these insights into business processes and tech applications is challenging.  

This problem is often known as the last-mile problem for AI. As a result, since the cost of implementing AI can be significant, enterprises may struggle to justify the investment without a clear path to generating the ROI. 

For example, we frequently face challenges as we present solutions around building Customer Data Platforms (CDP) to clients. Building a CDP is no easy undertaking. It needs integration with an enterprise data repository and then needs building even more supplementary data integrations because the enterprise repositories are often incomplete. It also takes time to build the right business reporting and dashboards — an effort that can very easily snowball into a complex program of its own.  

The approach we generally take to resolve this is an AI-first approach that is tightly linked with technology. Rather than limiting ourselves to traditional business intelligence dashboards, we start with AI scenarios and then also immediately bake in the API integration of insights into the client's business processes. As an illustration, for a retail client, we recently pitched integrating pricing analytics with their online offer management system as one program. Doing that right off the gate helped everyone see the value that we will leave on the table by not doing the AI program rather than the other way around. 

Some approaches that enterprises can take to address this challenge are: 

• Ensure that AI is not just viewed as a data initiative. It must be tightly integrated with tech implementation to provide a path for the insights to reach the business processes.  
• Integrate your API gateway with the AI models being developed. This makes it easier for developers to integrate insights into their applications. 
• Don’t hesitate to use robotic process automation to fill technology gaps. Sometimes technology is not the optimal or feasible solution.  
• Make AI a part of the governance process so that it becomes a prerequisite for every tech and operational project 
• Bring cross-functional and cross-business teams together so that the brainstorming canvas can get comprehensive inputs that span the entire set of business processes. 

Conclusion 

To summarize, while the adoption of AI can greatly enhance performance, there are several challenges to full adoption. These include data quality and availability, lack of understanding and alignment, and operationalizing AI insights into business processes, also known as the last mile problem.  

To mitigate these challenges, enterprises can consider approaches such as maintaining clean data repositories, aligning and cross-training technology and business teams, and closely integrating AI programs with technology implementations.  

Taking these steps will help resolve the common AI adoption hurdles and provide a clear path to ROI.