How to Overcome 4 Common AI Obstacles

There’s no question the impact of artificial intelligence (AI) will soon be felt in tax departments. However, AI still poses obstacles for those interested in harnessing its potential benefits. 

Here’s one challenge: Let’s say you’re a bank using a form of deep learning AI to extract value from huge volumes of data related to mortgage lending. The system does a great job, building its own models and improving your decision-making. But it’s almost impossible to understand how the AI created those models. So, what happens when a regulatory body needs to know how you arrived at your decisions? How can that regulator peer inside your AI “black box?”

It’s not difficult to imagine how applications of AI in tax could encounter this type of scenario, which is described in a Harvard Business Review article Artificial Intelligence for the Real World. The authors — Thomas H. Davenport, professor at Babson College, research fellow at MIT and senior adviser at Deloitte Analytics; and Rajeev Ronanki, principal at Deloitte Consulting — set out a four-part framework to help companies overcome common obstacles while achieving their AI objectives:

1. Understanding the technologies. To avoid wasting time and money on the wrong technology for a given task, companies will need to build data science or analytics capabilities in-house, or build an ecosystem of external providers in the short term. 
2. Creating a portfolio of projects. Davenport and Ronanki suggest companies should conduct assessments in three broad areas: Identifying the opportunities; determining the use cases; and selecting the technology.
3. Launching pilots. The proof-of-concept stage is crucial and companies should be wary of attempts to bypass it in the rush to “do something cognitive.” Consider creating a cognitive center of excellence and pay close attention to business process redesign.
4. Scaling up. Rolling out AI throughout the organization can be challenging. Integrating with current systems and processes is the most commonly cited hurdle in the cognitive technology initiatives the authors studied. A careful change management design is crucial.

While these speedbumps are real, the authors are bullish on AI’s future. They conclude that “companies that are adopting AI in moderation now — and have aggressive implementation plans for the future — will find themselves as well positioned to reap benefits as those that embraced analytics early on.”