Sudha Jamthe’s WeeklyWed AI Speaker Series, featuring Emanuele Bianchini (click link to go to video)
As a participant in Sudha Jamthe's Capstone AI Lab on Business School of AI online, Emanuele and his team applied Akkio’s no code AI to look at a real business problem for one of the members of Emanuele’s team. Their task was to leverage the data the conference planning business collected to increase revenue by 10%. Ultimately, they fine tuned their goal to look specifically at ways to increase attendance rates by analyzing a number of factors related to their potential audience.
Business Owner @Innovatech : Roberto Amigo
BizDev specialist : Emanuele Bianchini
Data Science Advisor: Kajal Gupta
Product Manager: Ananya Sen
Going in, Emanuele and his group thought it would be easy to load the data into Akkio, and then let the program tell them how to increase sales. They learned that their task was not quite that simple. Emanuele notes, “Data is the foundation but it’s not the answer. It’s what you do, the decisions you make, and the conclusions you draw,” that make the difference. He added that it’s an iterative process where you constantly “manipulate and assess what you have and what you get.”
The beauty of Emauele’s team’s Capstone Project was that they used Akkio to analyze real, raw data (unlike many academic projects wherein students use a canned set of data), so they experienced the struggle of applying this technology in a highly authentic way.
“Data science solves math problems,
AI solves business problems” Emanuele Bianchini
Emanuele discusses the “two perspectives” they had to balance, referring to the business problem on one side and dealing with the data on the other. He emphasizes the need for thorough communication and the importance of the “explanation,” which refers to defining the precise business problem, and deciding how data analysis can contribute. They quickly realized that their problem statement was too broad, and it took them several weeks of narrowing their statement to come up with an adequate description. Just saying they wanted to increase revenue was not enough to train the AI, and to get value out of the process.
Emanuele and his partners found that they were constantly tweaking their Problem Statement, trying to get more and more precise, something that most businesses will likely experience when they begin to apply AI. The graphic belows shows how they kept coming up with new and more specific questions as they moved forward.
Like anyone applying AI for the first time, Emanuele and his group questioned whether they had enough data, whether inherent biases impacted their results, and how can they verify the veracity of their assumptions? He acknowledges that there will be biases in any set of data, not because an individual has actual biases but because there are limitations to what you can see and know. “Part of the process,” he says, “is to constantly re-evaluate and iterate your model,” and to fine tune your problem statement throughout. He also emphasizes that every business situation will be unique and different.
Emanuele defines the process as a “continuous loop,” whereby they constantly go forward and look backward, improving their model all along the way. He talks us through the process of entering the data, modifying it, training the model and then running it on real data. He explains the two levels of initial data they utilized, both “basic” and “specific”. “Basic” data typically relates to demographics, whereas “specific” data relates to the actual problem or decision under examination (in this case if they attended the conference previously, or if they subscribe to the newsletter). Emanuele explains that about 80% of the data will train the no code AI model, and that a designated predictor allows them to exercise the model.
By the end of the Capstone Project, Emanuele’s group was able to build the model such that it could compare various input factors and determine the Key Drivers that determine sales, without using Python or writing any code.
Akkio is an easy-to-use, scalable, and affordable AI platform for real-time data driven decision making.
Among the numerous questions Emanuele and his team considered, just as any new user will, “Does this data help us predict future conference attendance?” Initially, they thought “no,” so they added more columns (more data) and continued to train their model.
Throughout the process, they continued to ask, “Are we learning anything?” and “What are the key drivers?” One of the things they found is that whether the conference was online or in-person had a big impact on whether people attended. And when they dug deeper into the details, Emanuele’s group noted that a person’s title has a big impact on whether they will attend a conference. They also learned that whether a potential participant was “invited” to attend has a significant impact on registration. In addition, he declares that some of the categories (descriptors) reported so little information that they eliminated those categories from consideration.
By the end of their project, Emanuele’s group felt they had learned a lot. They continued to stress the importance of deciding how reliable their reports are and how they can continue to improve them.
His final take-aways are
Richard Meyers is a writer/editor/blogger, UX customer success and business consultant.
Go to Business School of AI, Weekly Wednesday, for more information about how you can learn to use AI.