The process of building an analytics model usually takes a team of data scientists experimenting on a variety of complex techniques to transform data, extract features, train the models and interpret them. They are confident with their algorithms and results, but what about the business users that will consume the outcome of such effort? Do they have this same level of confidence?
Let's suppose I am a financial advisor responsible for high net worth clients. Every month, the data science department gives me updated churn propensity for each client and I have to intervene to ensure that they remain clients. Is it a solo ride with a bunch of numbers? Or is it a draining conversation between business and data science with one side talking about clients and contracts and the other one talking about neural networks and stochastic calculus?
This is where the chatbot comes into play
Chatbots use Conversational AI to enable humans to interact with machines using natural language and instantly get a human-like, intelligent response that is tailored to the user. In the SAS world, chatbots offer another user-friendly conversational interface to the entire Viya ecosystem, bringing together reporting capabilities, analytics and artificial intelligence.
Now, back to that client churn scenario. Let's see how I can interact with the chatbot to make interpretable analytics-driven decisions out of churn probabilities numbers:
The interaction with the chatbot allowed me to review the at-risk clients:
- I was able to get an explanation of the model results to dig deeper into the most important churn drivers.
- The chatbot provided visualizations to better understand key drivers and see the whole picture.
- Now that I understand why my high-value client may churn (too many marketing emails), I can take action.
- The chatbot also provided a list of similar, high-value clients who would be good candidates for a new marketing campaign we are looking to set up.
In summary, the chatbot proved to be a useful tool to access analytical insights and results, providing me the exact bits of information I was looking for, instead of having to navigate through multiple reports and visualizations. Moreover, having a combination of graphs, numbers and explanations that the data science team designed for me, I get insights instead of numbers so I can make faster, data-driven decisions.
Now I have a better understanding of what impacts the clients and their likelihood to churn. Thus, I can take this feedback to my marketing team to adjust the campaigns we run and have a better understanding of my clients to provide better-personalized service.
I am not a financial advisor (even if my parents did push for that when I was younger), but I am a data scientist and would like to write a few comments for my fellows.
What you saw in the video was the output of a gradient boosting model for evaluating clients' churn propensity. On top of this, I ran interpretability techniques such as LIME, to get local interpretable explanations, and ICE, to evaluate the dependency of the churn propensity on one variable. Finally, I applied clustering techniques to group clients by similarity.
Talking about chatbot development, SAS Conversation Designer allowed me to build one in a visual interface that requires little to no code and can trigger code execution or APIs calls to gather the results.
Thank you for reading.
This could be good in making the strategy for a consumer to improve their customer service.
Pingback: Accessing text analytics from a chatbot: Sentiment - The SAS Data Science Blog