In this insideHPC Guide, “10 Questions to Ask When Starting With AI,” our friends over at WEKA offer 10 important questions to ask when starting with AI, specifically planning for success beyond the initial stages of a project. Reasons given for these failures include not having a plan ahead of time, not getting executive or business leadership buy-in, or failing to find the proper team to execute the project. Chasing the hot technology trend without having a proper strategy often leads companies down the path of failure.
Artificial intelligence (AI) and machine learning (ML) technologies are disrupting virtually all industries globally—and AI technologies are not just being applied within robotics and vehicle automation. Companies from financial services to retail, from manufacturing to health and life sciences are seeing business improvements through insights generated by AI and ML.
Introduction
Planning for success beyond the initial stages of a project is key.
Artificial intelligence (AI) and machine learning (ML) technologies are disrupting virtually all industries globally—and AI technologies are not just being applied within robotics and vehicle automation. Companies from financial services to retail, from manufacturing to health and life sciences are seeing business improvements through insights generated by AI and ML.
Digital leaders are also paying attention to this emerging technology.
- According to the 2019 Digital Business study by IDG, organizations planned to spend $15.3 million on digital initiatives, with AI and ML high on that list.
- Despite the enthusiasm around the technologies, however, failure rates on AI and ML projects range anywhere from 50% all the way up to 85%.
Reasons given for these failures include not having a plan ahead of time, not getting executive or business leadership buy-in, or failing to find the proper team to execute the project. Chasing the hot technology trend without having a proper strategy often leads companies down the path of failure.
Fortunately, enough lessons have been learned through these failures to give companies a better game plan for their next AI or ML project. Listed below are 10 questions AI teams should ask themselves when they are beginning new AI projects.
#1 Have we clearly defined a goal and identified the right questions to get us there?
Amazingly, many companies don’t have a clear vision of the goal they want to achieve from an AI project. Moreover, they don’t have a good sense of the questions they need to ask and answer in order to take the necessary steps on the path toward achieving it.
“A lot of companies will start with ‘We know that AI is a game changer, so let’s see what we can do with it,’” says Shimon Ben David, the Field CTO at WekaIO, which offers a parallel file system to help companies with large storage problems—very much like those faced by companies starting their AI journeys.
Like an explorer preparing to reach a destination, a project leader needs to establish a final end point and then provide a map that includes specific directions to follow for each step of the journey. For an AI project, the specific outcome needs to be identified, and then directions are formed by asking and answering questions to help reach the goal and achieve the desired outcome.
The key here is to create a good AI team with the ability to ask and answer these initial questions. This could include software engineers, business leaders, subject matter experts, and possibly even customers within the environment.
For example, let’s imagine a financial institution that has the ultimate goal of increasing its bottom line by improving its profit margin. The first question to ask is, “How do we use AI to do this?” One answer could be to consider using AI to help decrease the high rate of defaults on loans, thereby getting a better return on their investments.
Therefore, who could ask the right questions to identify customers with the highest risk of defaulting on a loan? In this case, the institution’s individual account team members would be good candidates to ask questions and gather the data because they’re the ones closest to the sources of data—the customers, themselves. The account team knows the situational problems that customers face and routinely interacts with them, often hearing the reasons why payments come in late, which leads to loan status jeopardy and, sometimes, default.
For the good customers, the institution can then provide either rewards or incentives, such as a lower interest rate. For the high-risk customers, the institution can offer programs and monitoring to make sure they stay on track or get them out of the high-risk category.
Keep in mind that the questions a company creates to get to the final goal can change and evolve as more data is gathered. If you’ve chosen the right goal, it should stay the same, but the steps to get there might change as you encounter roadblocks and obstacles. If you haven’t identified the right goal, asking questions will make that clear so you can pivot in the right direction.
“Companies need to maintain a constant set of questions, and chances are good that they will morph with a project’s progress, but it’s imperative that you have an initial response to them at the beginning, thereby giving yourselves a place to start,” says Ben David.
#2 What data is required to achieve your goal or solve your problem?
After an AI project team has identified the goal or specific problem that AI can solve, the team continues to ask questions to determine the data or variables required to achieve the goal or resolve the specific problem.
In the case of the financial institution, after customers in the at-risk category have been identified, the team has reached only one step toward reaching the goal. Remember that the goal is not just to define those at high risk for defaulting, but to keep them from actually defaulting so that the institution can increase its profit margin.
The team asks more questions to reach the next step: Does everyone in the high-risk category face the same situation that keeps them from paying? Probably not, so how does the team identify and categorize customer subsets that need different forms of help to achieve timely payment? What are the remedies that might help people in each subset and keep them from defaulting?
This is where the actual data comes into play. A bank may have a customer’s name, personal information, banking details, social media postings, images, videos, and other records in order to answer those questions. Multiple data points exist, and companies might not need all of them. On the other hand, some information might be missing. In reality, most companies begin an AI project thinking they have enough data to answer the question, but a good portion of the data is missing, or the data that they have isn’t useful to answer the question. In his experience, Ben David says that he has never encountered a company that has collected too much data.
“Maybe I have bank records, but they don’t have a credit score,” says Ben David. “Maybe I don’t have social media with relevant tags that they posted that could help me understand their financial situation. Understanding the data and what’s in the data is important.”
Sometimes, companies have to come up with their own data to fill in for what’s missing. The tools that you would use to extract your data set would vary depending on the type of data you need to collect. For example, Google Analytics provides website visitor data and metrics, but you might also have a customer or contact database through Hubspot, Salesforce, or numerous other services.
Fair warning, however: Keep everything! Companies tend to acquire a vast amount of data, distill it when creating the AI or ML model, and then either store the raw data somewhere to never be accessed again or, even worse, delete the unused data. Ben David says that data can be critical later when reassessing a particular model where raw data is needed again.
For example, look at how criminologists utilize newer DNA technology and methods to help find or exonerate suspects in a crime that happened years or decades earlier. Because evidence is stored and saved in these cases, criminologists can go back and re-analyze the clues. The same principle applies with AI: you might not think you need all collected data now, but years down the road a better algorithm or new technological advancement may elevate a piece of seemingly useless data into a highly relevant piece of evidence (think DNA sampling of hair strands), and you could be pleased, to say the least, that you have that old data available.
Over the next few weeks we’ll explore Weka’s new insideHPC Guide:
- Introduction, #1 Have we clearly defined a goal and identified the right questions to get us there?, #2 What data is required to achieve your goal or solve your problem?
- #3 Where will I get my data if I don’t have it already?, #4 What is our organizational compute strategy: on-premises, cloud, or hybrid?, #5 What is our plan to move and store the data?
- #6 How will we remove bias and validate our model’s results?, #7 How often will we fine-tune the models?, #8 How do we deploy a new model?
- #9 How does my infrastructure look on day 3 vs. day 300?, #10 How do we future-proof the project?, Conclusion
Download the complete 10 Questions to Ask When Starting With AI courtesy of Weka.