Multitasking? Great! You can listen to this blog post by clicking above or find our podcast, Intelligent Data Exploration, on major podcast platforms.

Gartner reports that just over half of all AI projects make it into production. And of the few who do, many will go unused, either because the business doesn’t trust the models, or because they’re just not solving the right problem.

So how can you find the AI use case that will go the distance and kick off a project that will have an impact? It all starts with a good foundation.

Don’t skimp on data exploration

You can’t fix a problem you don’t understand. And you definitely can’t fix a problem if you don’t even know it exists. Data exploration is the only way to make sense of all the moving parts in your business and it’s a critical first step in any potential AI project. You need to discover where the real challenges lie if you’re to have an impact on the business. And you need to understand what is driving those challenges in order to target them effectively.

Exploration is often guided by a hypothesis–we think the challenge could be X so let’s explore to see if that’s true or not. The challenge with exploring this way is that it really limits the scope of what’s explored and has a high risk of introducing confirmation bias. Hypotheses can lead data scientists and analysts down the wrong path and away from the most meaningful discoveries. And if you’re not solving the right problem then your project is based on a false premise and you won’t be able to find and implement a solution that works.

Exploration should focus on the business challenge. Sales are down? Explore as many attributes relating to sales as possible. Want to try generating personnel schedules that anticipate demand? Explore all the attributes that impact staffing. The key is to look for all the relationships and drivers that exist, with an open mind, so that nothing gets overlooked. 

If this feels daunting, don’t worry. This is where the use of AI to explore data—we call it Intelligent Exploration—can help. Traditional BI tools aren’t designed to support the breadth and depth of exploration that good AI demands as a foundation. But when you leverage Intelligent Exploration, you can start to surface the most impactful opportunities for AI.

Identify your opportunities

Very few business challenges are straightforward so your exploration will probably identify a few challenges and, for each of those challenges, a few contributing factors. The next step is to refine your results to identify which challenges are significant and worth pursuing.

Start by consulting your business stakeholders and reviewing the results of your exploration. Their knowledge of the business can help inform your understanding of your findings and together you can translate your findings into potential use cases. Involving the business stakeholders early also helps to ensure they understand how AI projects develop and what data will ultimately be used to drive the model.

Making sure that your business partners really understand the insight you’ve uncovered, particularly if it involves multiple, interconnected relationships, can be challenging but it is possible:

  1. Structure and present your discovered insight with a storytelling narrative.
    • Set the scene (the high-level challenge you were exploring)
    • Point out the early areas of insight that were surfaced by the AI and caught your attention, then what you chose to look at next as a result
    • Describe the big ‘ah-ha’ insights that came into focus as potential use cases
  2. Use plain language wherever possible. It’s so easy to slip into the jargon that we use every day but unfamiliar terms can be distracting and make them feel like what you’re showing is beyond their comprehension. 
  3. Be very thoughtful about which visualizations you use to illustrate your findings, particularly if you need to highlight the interplay between attributes. 
    For example, if you want to show the interplay between 3 attributes, showing three 2×2 plots will not get that information across in a consumable fashion but a true 3D visualization of a 3×3 will.
  4. If your toolset allows it, leave the analysis with the business leaders to consume and explore on their own. Ideally, you’ll be able to annotate it, calling attention to the areas of interest. Doing this provides the team more time to digest what you’ve told them, and to “kick the tires” of the results. 

Refine use cases and create a ranked list

Together with your business stakeholders, plot the use cases by feasibility and business value. Create and prioritize the list of possible problems to solve so that the most important challenges are addressed first.

When determining feasibility consider the following:

  1. Should the use case be tackled with an AI model or could it be solved some other way? Just because you have a hammer, doesn’t mean that everything is a nail. Other solutions could be a process change, a system modification, or better or more timely analytics.
  2. Determine the cost or impact of not addressing the problem. This will help determine impact, mobilize the business to embrace the solution, and provide you with key metrics to measure success down the road.
  3. Can you put this model into production, support it, and foster the organizational change required to leverage it? A complex model that can be seamlessly integrated into a current workflow may be more feasible than a simple model that requires a big change to established processes.
  4. Fairly assess your data sources. If the data you have is weak, it won’t matter how good your report or AI model is; it won’t be usable. In fact, it could do harm. 

Select The Right Use Case  

Armed with your carefully plotted use cases, the result of thorough exploration and consultation with the business, you, your team, and your business leadership will be able to select the right AI use case to move forward with.

To learn more about how to successfully create the next-generation AI strategy that leadership is asking for, download our eBook “Building a sustainable AI strategy from the ground up”.

Lung cancer is the leading cause of cancer-related death, making up almost 25 percent of deaths from all types of cancer in the U.S. However, thanks to positive lifestyle changes such as quitting smoking and advances in research, detection, and treatments, that number is dropping.

Individuals with cancer are at greater risk of developing complications and dying from common illnesses and viruses, including influenza. So, when a research group from Columbia Medical School launched a study to explore the link between lung cancer mortality and flu epidemics, Virtualitics joined the team to provide state-of-the-art data analytics and multidimensional visualization capabilities.

The purpose of this study was first to determine whether a link existed between death from lung cancer and influenza and then to apply any significant findings to increase patient and provider awareness of the danger of influenza infection for patients with lung cancer.


How the Study Was Conducted

The study was conducted using 195,038 patients with non-small cell lung cancer (NSCLC) located across 13 states using data obtained from the Surveillance, Epidemiology, and End Results (SEER) Program and the Centers for Disease Control and Prevention (CDC).  

The research team compared monthly mortality rates during the high and low flu months between 2009 and 2015 for all at-risk patients in the study, as well as newly diagnosed patients. Influenza severity level was defined based on the percentage of outpatient visits to healthcare providers for influenza-like illness and matching CDC flu activity levels with SEER data by state and month. 

The data was then analyzed using high-dimensional visualization coupled with AI routines so researchers could fully understand and explore the complex relationships within the data.

What the Study Found

The research team observed a significant difference between the monthly mortality rate in patients during high flu months compared with low flu months.    

The positive relationship between flu severity and mortality was observed across all study participants, as well as at the individual state level and among new patients, specifically.

The results of the study, which were published in the Journal of Clinical Oncology, proved that increased influenza severity was positively associated with higher mortality rates for NSCLC patients. At the conclusion of the study, researchers indicated that the study’s findings support the need for future research to determine the impact of influenza vaccines on reduced mortality for patients with lung cancer. 


3D Visualization and AI-Driven Analytics Enable Research Teams to Get the Most Valuable Information from Data

The Columbia Medical School research team was able to determine that a correlation exists between influenza severity and patient mortality by using three-dimensional visualization and artificial intelligence to analyze and model data. This type of visualization has several key benefits that improve data analysis capabilities across not just research but all industries.


Increase Visibility

Unlike 2D models, multidimensional visualization allows researchers to add and remove variables to measure the impact on the subject and get deeper and more complete insight into the relationships between datasets. For example, during the lung cancer study, researchers examined flu severity level by month and year simultaneously. 

Identify Interrelationships

Immersive data analytic technology is essential for identifying patterns, trends, and outliers that researchers can easily miss when the data is only presented in two dimensions. Without the additional plane, critical data points may be occluded that could make a significant difference in a study’s results.

Three-dimensional visualizations can also be applied to unstructured data to enable research teams to gain valuable insight and identify relationships that exist between datasets from otherwise hard-to-analyze, disparate data sources.

Improve Communication

Applying multidimensional visualizations to complex AI and machine learning models allows researchers to display data relationships in a variety of geospatial and graphical visualizations that are easily understood by researchers, data scientists, and the non-scientific community alike.

AI-driven data analytics and 3D visualization tools also enable distributed research teams to collaborate over geographic locations. With collaboration and presentation tools available in both desktop and VR, researchers can work together securely to find, visualize, and share data insights, anomalies, and patterns, no matter where they are located.

Multidimensional visualization paired with the power of AI and machine learning is driving advances in healthcare and medicine. When the Columbia Medical School research team and Virtualitics partnered to tackle the world’s deadliest cancer, they were able to look at the data in a whole new way.

Contact us to learn how Virtualitics AI Platform can help your organization get the most information and value from your data. 

Detailed paper: 

Influenza and mortality for non-small cell lung cancer. Authors: Connor J Kinslow, Yuankun Wang, Yi Liu, Konstantin M. Zuev, Tony J. C. Wang, Ciro Donalek, Michael Amori, and Simon Cheng. Published on Journal of Clinical Oncology, Volume 37, Issue 15.