Subscribe to DSC Newsletter

Avoid Common Pitfalls in Launching AI Projects

By Pranay Agarwal, VP, Sales and Customer Success at Analytics2go 

 

     The theme of artificial intelligence adoption is becoming the competitive edge for many companies in different verticals—allowing for both disruption and innovation. Accenture’s research shows that two-thirds of organizations surveyed are planning investments in AI over the next year. That’s not surprising when you consider that, in some industries, these investments are expected to boost revenue by over 30% over the next four years.

     We are already seeing many AI solutions in the marketplace, from facial recognition and chatbots to machine-learning models applied to supply chain management, and even solutions embedded within Internet of Things (IoT) applications. According to Gartner, however, only 13% of AI projects have gone into production. Why is that? Below are some reasons I have observed when speaking to our customers at Analytics2Go: 

 

Lack of Strategy

     Selecting the best strategy and having a clear vision of where to start an AI project can dictate its outcome. Like any new initiative in a company, AI initiatives should be guided by business goals, although not all business goals can be addressed with AI. Companies need an AI plan to avoid wasting time and money by mapping out what projects to do first, second, third, and so on. Understandably, companies are often confused about where to begin because of all the breathless hype and pressure to do something quickly, as in these demands: “Hire data scientists!  Buy this shiny new software! Our platform will run your business!

     Your business will have different priorities and goals than even your closest competitors. To differentiate, it doesn’t make sense to simply buy what everyone else is buying and run the same solutions. So, define what your goals are and the outcomes that your organization wants to achieve with AI. This part of the process is arguably the most important and should involve a stakeholder from each level of your organization to ensure agreement on goals and adoption by your organization. If you want to improve your company’s supply chain accuracy, for instance, you need to involve your S&OP leader, your inventory planner, and your supply/demand planner, etc. They have their particular focus and perspective and understand the business challenge from their lens. Your own people are your best source of information on how to achieve your overall business goals. 

     To begin the AI strategy process, the most effective plan is to start by decomposing your current business processes to identify which recurring decisions would benefit from being automated. What are your recurring decisions that could benefit from automation and save your workforce’s time for more important responsibilities? Even if the recurring decisions typically rely on a human’s judgment as the final step, automation can still reduce the time required for your managers to make a more accurate, data-driven decision. These are your “quick wins” and are possible with automation and purpose-built AI solutions embedded into your existing workflows.  

 

Oversimplification of the Problem 

     Companies tend to underestimate the complexity of the problem they are trying to solve, especially in the current “rush to AI or die” environment. They can fail to identify the hidden complexities often found in business processes that span multiple functional areas and stakeholders. For instance, if a company is focused on creating a machine-learning algorithm to improve the forecast accuracy of sales, it is critical to understand all of the stakeholders’ objectives in the value chain. For a CPG company, for instance, that would require evaluating the entire value chain, including your retail sellers (large and small), internal planners, suppliers, and others. Each functional area is focused on improving its own objectives, which may not align cross-functionally, so it’s important to achieve alignment among these different stakeholders and strive for synchronization toward a common objective. Some companies create workshops on this objective alone. 

 

Data Challenges

     For an AI project to be successful, it is important to identify the data your company has available to solve the problem, but it’s equally important to get creative and bring in datasets that provide you with contextual data intelligence. For example, geopolitical events, location, weather, and competitor data can take your supply chain demand-forecasting model to the next level by shedding light on the context surrounding each point in your value chain.

     Perhaps your company could use imaging data of your rail routes or benefit from real-time social-sentiment-analysis data to understand product-demand indicators. The possibilities are growing every day with new datasets, so think outside the box. Every company operates in a different environment and has a different set of variables affecting daily, monthly, or quarterly decisions. I have observed a tendency to underestimate how much time/work it will take to optimize the data intelligence for successful AI initiatives. Importantly, a continual evaluation of available data should be put in place since the world is always changing and new data are  available every day. 

 

Overly Aggressive Project Plans

     Companies are often so focused on achieving results in a short timeframe that they lose sight of what is really possible. Any AI project will require a proof of concept (POC) phase that serves to prove that the model and data are capable of delivering value. Most POCs can be done in 4 – 8 weeks, but the next steps are the most difficult and often erroneously assumed to flow effortlessly once the model has been proven to work.

     We have observed that with a successful POC, acceptance of the AI solution grows within the company. Getting support from C-level to end users is critical to the success of any AI initiative in any company, regardless of size. Taking an AI solution from POC to deployment and enterprise-wide operationalization involves technical integration that is typically accounted for before the beginning of the project. However, getting AI solutions to operate throughout a company and have end users able and willing to adopt them can often be the failure point.

     Likewise, user experience and training can often be the differentiators for the success or failure of AI initiatives. A step-wise scaling of AI solutions within a company lowers the risk of its falling into that 87% that never deploy and operationalize their AI initiatives. Adoption takes time and should be planned along with the AI strategy. Stakeholders want to be part of the solution and the rollout. If they own it, they will use it.

     To prove value in a POC and in later phases of an AI project, “gates” should be put in place where the project is evaluated on the basis of measurable metrics to determine if it continues or if changes are needed. More often than not, the success metrics serve to help the business stakeholders and the data scientists to pause and make necessary changes to assure success. 

     We often call describing something from beginning to end in a brief way as “the 30,000-foot view.” For AI projects, this view could be used to outline the overall challenges faced and the benefits received from AI as related to the long-term goals of a business. For example, from a composite customer: “We partnered with an AI-as-a-Service company to guide us in our AI strategy. We were in need of more accurate and frequent demand-prediction forecasting of our global inventory to capture more revenue and reduce our exposure to ‘out of stock’ scenarios. Our partnership with A2Go led to multiple solutions that provided automation and sophisticated AI solutions that brought in a large amount of external data that we would not have had access to without our AI-as-a-Service partner. These capabilities have allowed us to reach our company goals with completely automated, real-time AI solutions that recalibrate and learn on their own. Now we can spend the recaptured time on our customers.”

     Based on my observations above, to get to this 30,000-foot view, a business can break it down into many sprints separated by “gates”—sprints that make sense for that business. Eventually, the sprints add up to the proverbial “30,000-foot view.” Consultants refer to this sprint approach as the Scrum Methodology.

 

Pranay Agarwal is Analytics2Go’s Vice President of Sales and Customer Success. He has over 20 years of experience, from leading B2B technology sales and management consulting projects for Fortune 500 companies to leadership roles at SAP Ariba (in its procurement and supply chain management practice) and IBM (heading its Emptoris Sales for Latin America). He has been a management consultant at both PwC-PRTM Technology Group and Deloitte Consulting Service. 

Views: 198

Comment

You need to be a member of Data Science Central to add comments!

Join Data Science Central

Videos

  • Add Videos
  • View All

© 2020   Data Science Central ®   Powered by

Badges  |  Report an Issue  |  Privacy Policy  |  Terms of Service