By Christie Huber
Data science is a hot topic across many industries with marketing being no exception. While many marketers are still getting up to speed on the topic, here at DialogTech, we are fortunate to have an in-house team of artificial intelligence (AI) and data science experts that can answer all of our questions.
We had a chance to sit down with Tim Hoolihan, Sr. Director of Data Science at DialogTech, to talk about his experience in working with data science for marketers. Particularly, we wanted to get his thoughts on how using AI and data science can help uncover a better understanding of the customer journey, and provide a clearer picture of the return on your marketing dollars.
I started at DialogTech nearly three years ago, but I have always had an interest in analysis and understanding mathematical aspects of systems. By vocation, I worked in software development and focused on process, evolving technologies, etc. When I began working with the R language, I found the blending of these two interests. My degree is in Information Systems through the business school, instead of the more traditional computer science route. Therefore, I enjoy that data science is often applied to core business problems, such as marketing strategy and sales.
I enjoy almost all aspects of data science, but there are a couple of areas that stand out. Natural Language Processing (NLP) is a very interesting field that opened my eyes to the way I think about language and the meaning of words. By facing the challenge of making decisions based on language and the ambiguities involved, you gain better understanding of other related fields.
For example, the field of NLP has given me new respect for translators and the difficulties they face. Additionally, as I watch my own children learn to read and write, I have a new appreciation for the inconsistent and arbitrary rules of English they are trying to learn. For native English speakers, we learned these rules a long ago and by rote memorization, so we forget how quickly they crumble when trying to apply logic. That is one reason that probability-based methods in language systems have improved the understanding of conversations so much recently, as compared to older rule-based language systems.
The second aspect I really enjoy is simulation of systems and processes. Often analysts and data scientists can get wrapped up in correlation or modeling. Those techniques can be used to predict future outcomes, but are really an application of patterns on past data. It is easy to believe those patterns, models or insights tell the whole story. When you go the extra step of simulating something, you attempt to create the system yourself. You can then compare the simulated results to the actual results. It’s often a humbling experience that gives a new perspective on the complexity of systems.
For example, a manager of a car dealership may think they know the percentage of drop-off at each stage of the funnel as customers go through the sales pipeline. But if they sit down with the data about the customers that start in the pipeline each month and write a simulation of the stages, they may find that they should have double the sales they actually have. It’s a stark realization that the mental model of the sales pipeline isn’t always accurate. You can reach that same realization via descriptive statistics of the drop-off rates, but I often find simulation a better mental exercise for getting the size and scope of a problem.
The Data Science team at DialogTech works on a variety of problems across product lines. It is our challenge to research our data and the latest data science tools to ensure we are constantly improving our customers understanding of their phone conversations. The sales and marketing professionals that rely on our tools are the best people to decide how to communicate with their customers, but it’s our job to give them as much insight as possible into those conversations and customer journeys.
In day to day terms, that means working with our analytics team to refine and tune our Dialog Analytics product, which provides conversation classification driven by machine learning. In addition, we have exciting work underway to tell our customers even more about their calls using artificial intelligence, which greatly reduces the burden of listening to large call volumes.
I am the organizer of the Cleveland R User group, an after-work meet-up group for those interested in the R language, which is well-suited for statistics, modeling and other data science uses. I also attend other related groups in the area and have published a three-part video series this year with Packt Publishing about machine learning in R. In early 2018, I will be speaking about optimizing R performance at Codemash, an annual conference for developers.
We work to maximize the use of transcriptions in modeling in order to provide insights across our customers’ calls. There is more we can do in this area and exciting stuff to come on this front.
Another area ripe for innovation is combining conversation analysis with website usage data and customer information in order to provide a broader picture of the customer journey. Depending on how a customer integrates with our system, they may already have insight into this. I see us bringing this customer journey overview to even more of our customers.
Finally, great data visualization is challenging, but worthwhile for marketers. Insights are valuable, but even more so when you can effectively communicate those insights to your peers and sales staff. You maximize the return of your data science work when you have clear, concise communication of the actionable insights you have derived.
Always question your assumptions. Can they be verified with data? If you simulate your process using your assumptions in code, do the results look anything like the reality you see in the market?
I would also emphasize focusing on sound, basic principles when analyzing problems. It’s easy to get wrapped up in the modeling of a problem. However, if you’ve gathered a poor data set for your analysis because of sampling problems, all the algorithms in the world won’t save you. It’s easy to get distracted by new tools and techniques, but you can’t forget that other problems haven’t gone away, like confusing correlation for causation, confirmation bias, survivorship bias, etc.
The tools are catching up rapidly, democratizing the ability to work with data, meaning that more and more people in an organization will be able to do analysis work. This will emphasize the role of understanding the fundamentals of how to prepare data for analysis and how to interpret the model. There will be a need for sharing and growing data science skills within the organization, as well as leveraging the platforms that are built on sound data science principles.
I see reinforcement learning as a growing field. It focuses on problems that occur and change over time and attempts to model the impact of a single action in a larger context. These same techniques that are useful to build AI systems for games could be applied to support decision making in contextual time-driven business processes, like the sales pipeline.
With the recent launch of data science and artificial intelligence (AI) capabilities, DialogTech is pushing the boundaries of innovative technology to offer clients cutting-edge marketing strategies.In the meantime, be sure to check out the on-demand recording of our popular webinar, Beyond the Hype: 6 Ways AI Works With Calls to Improve Sales and Marketing.
In a recent survey, 34% of marketing execs said Artificial Intelligence is the technology they are least prepared to leverage in 2018. Our AI experts explain specific ways AI can help you acquire more customers.View Teh On-Demand Webinar →
Get the latest straight to your inbox.