Artificial Intelligence Machine Learning Customer Support

    The Machine Learning BIOD-ome

    By Sinan Ozdemir on September, 25 2017

    Stay up to date

    Back to main Blog
    Sinan Ozdemir

    Business leaders from a wide variety of industries and from companies both large and small are feeling the pressure to implement viable and effective AI strategies to keep up with the many innovations happening in the field.

    As a CTO and a former consultant in the data science and machine learning space, I often get asked the question, “How do I go about planning out and deploying a machine learning project at my company?” Most of the time, managers do not have intimate familiarity with machine learning or how to apply machine learning to their business problems. When I get asked this, I almost always reply with a simple answer, “BIOD”.

    If you are considering incorporating a machine learning model into a project you are working on, I suggest you ask yourself four questions:

    • What is the Business application of the machine learning model?
    • What is the Input to the model?
    • What is the Output to the model?
    • What is the Data and where is it coming from?

    A machine learning model is an artificial intelligence learning algorithm that can estimate a map between your inputs and your outputs using simple and sometimes complicated mathematics and procedures. Let’s take a look at an example of implementing artificial intelligence in a customer support application.

    Suppose you wish to surface article links to customers upon request in a chat box. The customer will ask you “how can I reset my printer” and the model with respond with a link called “How to reset my printer”. Simple enough.

    The business application here is quite clear: satisfying your customers’ questions promptly and accurately will help customer retention and prevent against lost revenue. So now the question is what is the input to our model that will surface article links? In this case, the input would be raw text coming from the customer. The output would be the article’s link and perhaps a quick summary of the article. To train this model, we need data. We can either look back at historical data of humans surfacing articles upon request (if it exists), or we can directly gather data from the company knowledge base of FAQs, help articles, etc.

    Once we are clear on the business application, input, output, and data source, a skeleton of the project becomes clear. Gather data from the company knowledge, train a machine learning algorithm to match incoming requests to outgoing links and deploy in the forward-facing chat box.

    Using BIOD is a great way to articulate the goals of machine learning models as well as get on paper how you are going to do it. It is also useful in that if your BIOD is not very clear and you struggle coming up with these four ideas, perhaps a machine learning model is not the best way to move forward. Either way, BIOD might help reveal the path forward.

    Submit a Comment

    Stay up to date