Advanced Computing in the Age of AI | Monday, August 8, 2022

How Aligning AI Project Teams Can Ensure AI Success 

While AI deployments are picking up steam, challenges to success, both technical and cultural, abound. One of the biggest challenges is the lack of alignment among the three pillars of enterprise AI success: business users who are closest to the data, data engineers charged with keeping the data pipes open, and data scientists who make AI work. This lack of alignment typically means that data science teams find themselves “boiling the ocean” without a clear scope, while data engineers don’t know what data sets to focus on – which can lead to very disappointing results for AI projects. Only by aligning these groups around a standard data science methodology can consistent AI success be achieved.

The lack of alignment

The lack of alignment isn’t surprising. Business leaders who came out of school 15 or more years ago were never trained to take advantage of big data. While data scientists always “start with the data” – and the more data the better – business users, whether in finance, marketing or customer support, typically have a limited set of questions they want answered based on their accustomed approach to their discipline, so they frequently ask the data science team for a report based on a specific set of “how many” or “how often” attributes for a given criteria.

Fortunately, the disconnect between business users and data scientists is disappearing. Today’s business majors understand the power of big data, and it’s even making its way into elementary schools. Flash cards for memorizing addition and multiplication have been replaced with exercises that foster an understanding of categorization, organization and relationships.

In the meantime, we must still overcome the old-school mindset and allow data science to take the lead. That is, we must challenge the HiPPO (highest paid person's opinion). While the HiPPO’s experience and gut instinct should always be respected, relying solely on this approach will significantly limit the ability to apply data science to any problem.

Benefits of alignment and standardization

Ensuring alignment and a standard approach to AI will unify the entire enterprise around a critical new paradigm that more data is always potential energy to propel the business forward. Alignment and a spirit of collaboration will also eventually empower business leaders to ask the data science team to “use data to identify my best next action and automate it.” This is critical for launching high-volume, low-complexity projects that can quickly have a significant impact on the business.

Consider a virtual assistant designed to answer common customer questions. Today, customers calling into a help center often ask the same questions over and over, requiring support engineers to repeat the same answer. Instead, what if the support team works with the data science team and IT to use those known questions and standard answers to train a 24/7 virtual assistant, or chat bot, which can then provide customers with immediate answers – no waiting on hold, no support ticket, no wading through a website for FAQs or articles. This can transform support operations, increasing customer satisfaction while allowing support engineers to focus on new and more complex challenges that customers may encounter.

Another example is adding real-time system health monitoring to customer applications to allow support engineers to review system data leading up to a system degradation or crash. By aggregating and analyzing data from an ever-growing number of customer systems, it becomes possible to predict when an individual customer’s system is at risk and automatically notify that customer of any required preventive action.

This is essentially the goal of AIOps, using data combined with AI to automate next best actions based on what the data reveals, and then doing this continuously: more data, refine the next best action, automate the new action, and so on.

How to align your organization for AI transformation

  1. Build effective teams. To whatever extent possible, ensure the key members of AI project teams are highly skilled, familiar with AI, and work well together.
  2. Start with a standardized methodology. This ensures every group is on the same page. At NetApp, we standardized on the CrispDM methodology for data mining. However, other methodologies may be right for your organization, such as TDSP.
  3. Plan for Consumption. In every stage of the data science and business exercise, know how the end results will be consumed.  Make sure that your consumption plan is understood across the teams so that everyone is including information, features, automation, or whatever is necessary to meet the business result in consuming the new found insights and answers.
  4. Pick a scope you can learn from. Start with business teams that appear to have the best understanding of their challenges and are the most open to new approaches to solving them, and include subject-matter experts closest to the data. This approach offers the best chance for impressive results, which will encourage more business leaders and data experts to get involved
  5. Get buy-in for alignment. Solve a stubborn problem that management did not think was solvable. Every business group has at least one of these. Challenge the business team: “What would you like to do better?” If there’s data around it, then let AI solve it. Inventory forecasting might be a great target in your organization. Or it might be helping marketing to stop investing in content that isn’t supporting the current customer journey. These are difficult problems because they require analyzing huge amounts of data in a way that is beyond human capability.
  6. Ask why. Encourage business leaders to open up about the deeper problems and challenges they face. Probe by always asking “Why?” And respond to every answer with another “Why?” Ask what data exists that might be related to the problem, but this is only a starting point. It’s up to the data engineering team to find and include every potentially related data set and then refine the data based on the problem to be solved.
  7. Agree on the desired result The business sponsor of an AI project probably works with others who have a stake in the outcome. Work with all stakeholders to ensure they agree with the goals and the results.
  8. Be willing to fail fast. Quickly recognize when you don’t have the right data. Then either change the data or change the expectations.
  9. Be flexible. Use multiple AI techniques and models. Rarely is just one model sufficient for all problems. In addition, include cloud and on-premises solutions to balance capacity and costs. The efficient use of resources is another critical factor in spurring adoption.
  10. Respect but always challenge the HiPPO. AI cannot succeed with old-school thinking.

About the Author

Ross Ackerman is currently the head of analytics and enablement of Active IQ at NetApp, where he leads a team of over 100 data scientists and software engineers building cutting-edge data analytics tools. Additionally, Ross is Director of Transformation for NetApp where he leads business teams across the company to leverage artificial intelligence / machine learning to solve business problems with data and analytics. As well, Ross has 4 granted patents for his research in machine learning, internet of things (IOT), software engineering, and big data technology.  Ross and his team was recognized by the Association of Support Professionals for the 2018 best Customer Support Site for self-service with AI enabled virtual support assistance.

Add a Comment