Kirill Eremenko: 00:00:00
This is episode number 415 with Head of Data Science at the Customer Journey Division at MassMutual, Asieh Ahani.
Kirill Eremenko: 00:00:12
Welcome to the SuperDataScience Podcast. My name is Kirill Eremenko, Data Science Coach and Lifestyle Entrepreneur. Each week, we bring you inspiring people and ideas to help you build your successful career in data science. Thanks for being here today and now let’s make the complex, simple.
Kirill Eremenko: 00:00:44
Welcome back to the SuperDataScience Podcast, everybody. Super excited to have you on the show. Today, we’ve got a very interesting guest, Asieh Ahani, who combines a background in academia, a very deep background in academia and her current profession in the world of data science in industry. This podcast is interesting because we spoke about both and it’ll be useful really for all levels whether you’re a beginner, you’re an advanced data scientist or you’re a manager or a data science leader.
Kirill Eremenko: 00:01:20
The reason is because, at the beginning of the podcast, we talked a lot about her academic background and her career, her progression through there and we touched on some interesting technical topics. For instance, you’ll learn about brain computer interface and some of the research that she was doing there, what kind of tools she was using. We talked about the curse of dimensionality. We touched on kernel analysis, classification theory. You’ll hear about signal processing. We also talked about quite intensely or quite deeply about stochastic versus deterministic signals, quite a few interesting things to pick up even if you’re quite advanced already in data science, you might hear some new areas that you could add to your tool kit.
Kirill Eremenko: 00:02:11
Then we moved on to discussing academia versus industry and why she made, why Asieh, made the move to industry. Then we moved on to discussing her career path, how her career developed which is very impressive how she has progressed through her career.
Kirill Eremenko: 00:02:29
We talked about topics such as being an individual contributor in data science versus being a leader, what it is like to be a manager in data science and how to lead a team of 13 people, some of the pros and cons of management in data science. You’ll find out how data science divisions can be structures for instance in the example of MassMutual. Also, heads up. Asieh is hiring at her division so in this podcast, you’ll find out how you can apply for a position and that’s specifically for advanced data scientists. They’re looking for somebody to join the team.
Kirill Eremenko: 00:03:07
At the end of the podcast, we touched on a very important topic: women in data science and you’ll be able to find out some advice from Asieh and the things that you can do to help data science be even more inclusive.
Kirill Eremenko: 00:03:23
We’ve got a cool podcast coming up with lots of topics. Can’t wait to get started so without further ado, I bring to you Head of Data Science of the Customer Journey Division at MassMutual, Asieh Ahani.
Kirill Eremenko: 00:03:40
Welcome back to the SuperDataScience Podcast, everybody. Super excited to have you back here on the show and today we’ve got a special guest calling in from New York, Asieh Ahani. I hope I pronounced your name correctly. Welcome to the show, Asieh.
Asieh Ahani: 00:03:54
Thank you very much. Yes, you were correct.
Kirill Eremenko: 00:03:57
Awesome. Fantastic. Well, excited to have you on the show. How’s New York these days? We already talked a bit about coronavirus, is it cold? Is it warm? How’s the weather? How’s the winter going?
Asieh Ahani: 00:04:10
Yes, it is getting cold which is fab because the city was on lockdown in the summer. We couldn’t really enjoy the warm weather so it’s fab that it’s getting cold but I like fall.
Kirill Eremenko: 00:04:24
I see you’ve got a lot of plants in your house. You like to have a lot of nature inside.
Asieh Ahani: 00:04:31
Yes, there’s not much nature outside so you have to compensate for that.
Kirill Eremenko: 00:04:35
Yeah, yeah. Okay and are you far away from Central Park?
Asieh Ahani: 00:04:41
I am. I live in Financial District so at the edge of the island. Central Park is in the middle so I’m not very close by.
Kirill Eremenko: 00:04:52
Yeah, New York is huge. It takes a long time to walk the distance. You’re near Manhattan, is that right?
Asieh Ahani: 00:04:58
Yeah, I am in Manhattan but I’m at the edge of the island in Financial District where the World Trade Center and 9/11 Memorials are.
Kirill Eremenko: 00:05:08
Okay. Okay. Got you. You have a very interesting story. Before New York, you were in Boston and before that you originally are from Iran if I understand this correctly. Walk us through this. You’ve studied at several different universities. You’ve traveled the world. We’d love to hear how it all started.
Asieh Ahani: 00:05:32
Yeah, I mean, it all started in high school. I always loved math. I was very much involved in a lot of different extracurricular courses around mathematics and statistics. That’s why I chose electrical engineering communication for the college degree. I got my Bachelor of Science in Ferdowsi University of Mashhad in Iran. That’s the city that I’m from. It’s in the eastern part of Iran. I studied in electrical engineering because to me that was the field within engineering that was mainly using mathematics and the statistics compared to physics or more applied sciences.
Asieh Ahani: 00:06:19
In retrospect, I stepped on to theoretical math instead of engineering but at least I think I found my way along the way. Then I moved to Boston for my … actually I moved to Tehran. This is the capital of Iran for my master’s degree in bioengineering because I was very interested at the time in neuroscience.
Kirill Eremenko: 00:06:45
That’s a big shift from electrical engineering to bioengineering. What made you choose, make that change?
Asieh Ahani: 00:06:52
Well neuroscience is very much, so again bioengineering have different fields. There are different fields within bioengineering. There’s biomechanics, bioelectric and biochemistry. I chose bioelectric. It was very similar to electrical engineering. The only difference was that the application was in biomedical fields at that time, they had neural signals. You’re basically using the same statistical methods like signal processing or machine learning what probability theory but to analyze brain signals not to analyze electrical data.
Kirill Eremenko: 00:07:38
Got you, so you made the shift and was it a good move? How did you feel?
Asieh Ahani: 00:07:40
It was. I think because again the biomedical applications are really challenging which makes it very interesting. I really love that shift and I think I learned that you don’t have to be afraid of just changing course because you might find something that is more interesting.
Kirill Eremenko: 00:07:59
Okay, awesome and so you completed your master’s in Tehran and what happened after that?
Asieh Ahani: 00:08:04
I think at that time, I was just trying to decide whether I want to pursue a PhD program, PhD degree or not and I applied for a couple of universities. I got an admission from Northeastern University from Boston and I realized okay, that’s a good challenge. Not only I have to change the country, my country of residency, also I can pursue a higher degree. I looked to United States.
Kirill Eremenko: 00:08:34
What was the PhD in?
Asieh Ahani: 00:08:37
My PhD was also in bioengineering. My adviser was in electrical engineering department but most of his projects and funding were in bio fields specifically neuroscience and brain-computer interface. That’s why although I was in the bioengineering department, I was working with an adviser who was in the electrical engineering department.
Kirill Eremenko: 00:09:02
Okay so your past continues, right? Your bachelor degree. All right so bioengineering, neural science, and if you were to describe bioengineering especially the electrical part to somebody, what is it all about?
Asieh Ahani: 00:09:22
It’s about signal processing and machine learning, I think. I mean at least the area that I was involved with. Our brain, even our body, the muscular, our muscles, there are so many data that you can collect from a human body specifically brain but not excluding the brain. There’s a lot that you can do on that data not only to understand the biology but also to use that data in order to help people that have any disability whether it’s visual disability, brain disability or physical disability. That’s basically everything is based on signal processing, processing that data and then machine learning which means coming up with prescriptive analysis based on that signal processing portion.
Kirill Eremenko: 00:10:21
Awesome.
Kirill Eremenko: 00:10:23
This episode is brought to you by SuperDataScience. Our online membership platform for learning data science at any level. We’ve got over two and a half thousand video tutorials, over 200 hours of content and 30-plus courses with new courses being added on average once per month. All of that and more, you get as part of your membership at SuperDataScience. Don’t hold off. Sign up today at www.www.superdatascience.com. Secure your membership and take your data science skills to the next level.
Kirill Eremenko: 00:10:57
Can you share with us what was the main topic of your research for your PhD?
Asieh Ahani: 00:11:08
Yeah, I was doing both signal processing and machine learning. I did analyze a lot of EEG data to understand the changes in brain signal when somebody’s meditating and what is the changes in EEG. Also, I used machine learning to develop brain-computer interfaces. Brain-computer interfaces are basically tools that you train to collect the data from human’s brain and then translate that into a robot to perform the action that the human wants. For example, you want to move your arm. There is a signal going on into your motor cortex and then brain-computer device can detect that and moves a robot arm. They’re basically moving that arm using your brain without even moving your arm.
Asieh Ahani: 00:12:05
My project was detecting attention signal just to help people with disabilities who cannot move their hands to build a brain-computer interface keyboard. Let’s say for example, I can show you A, B, C, D, E, F and you are actually looking for the letter E. When you see E, there is an attention signal in your brain, “This is what I’m looking for.” Our device could capture that attention signal and type E for you. Basically you don’t have-
Kirill Eremenko: 00:12:41
You’re not doing anything. You just see the right one. You don’t even have to point with your eyes or anything like that. You just see the right one because you have the letter in your head already, something different happens with it. Wow!
Asieh Ahani: 00:12:53
Yup.
Kirill Eremenko: 00:12:53
That’s so cool. Wow! You and I’m guessing because technologically we’re still not there, you are doing non-invasive signal reading like there is that called the EEG device when you have all these sensors on your head but there’s no surgery. There’s no invasive connection with your brain. Is that correct?
Asieh Ahani: 00:13:18
That is correct. Yes. It was mainly non-invasive. In my master’s, I was working on some MEMS technology to build biosensors that you can actually implant in someone’s brain but I changed my course instead of going to that MEMS field I moved to more machine learning as another field.
Kirill Eremenko: 00:13:42
Okay. How accurate are these EEG devices?
Asieh Ahani: 00:13:48
It depends on what type of signal you’re trying to collect. The signal such as attention and visual or motor cortex are pretty strong. You should be able to actually collect them with a high level of accuracy using EEG. But then, you go into cognition, how can I understand someone’s behavior or emotion using their brain signal, I don’t think EEG can help you that point.
Kirill Eremenko: 00:14:18
Okay, okay. Got you. Very interesting field. Very interesting. What kind of machine learning techniques did you use for processing those signals?
Asieh Ahani: 00:14:34
First of all, I think the biggest challenge is curse of dimensionality. Then you want to train and classify. You have a limited amount of time to train and classify. It means that you have a low number of observations and the dimension of the data is really high. Using different algorithms to reduce the dimensionality is key. We use a lot of that.
Asieh Ahani: 00:15:02
Then after that we did kernel analysis and classification theory to classify the brain signal. For example, in this example of brain-computer interface keyboard, when you see A, B, C, those are the data points that is not your desired data points. Their target class is zero for them whether you see D which is the letter you’re looking for, it should be target class one.
Asieh Ahani: 00:15:36
First, separating that data and building a classifier to train on these training data set that makes standard you’re looking for the data that is your desired letter, the classifier can actually detect that and classify that accurately. It’s a lot of classification, regression analysis and also the basic signal processing work to remove noise, clean the data, and again reduce the dimension of the data.
Kirill Eremenko: 00:16:04
Okay, wow! There’s a lot there. Let’s go through those step by step. Curse of dimensionality, can you please refresh my memory on what is the curse of dimensionality?
Asieh Ahani: 00:16:16
Yeah, it’s generally when your dimensions are more than your observations. That means the covariance matrix you cannot have the inverse of the covariance matrix, obviously. What you want to do is to reduce the dimension of the data so there are many transformation that you can do on your covariance matrix. You can do regularization. You can do shrinkage, which means that you shrink your covariance matrix values toward the pooled in average. Your eigenvalues are actually not all zero. You can actually calculate the inverse of the covariance matrix. Regularization as well, you want to shift the values of your covariance matrix against your the pooled in average. With that-
Kirill Eremenko: 00:17:08
This is all so easy to you. Sorry to interrupt. This is going to sound funny. Can you refresh my memory please on what is a covariance matrix? I know it might sound really obvious to you but I need a refresher.
Asieh Ahani: 00:17:20
Sure, I mean, EEG data is you have a Gaussian assumption on an EEG data. We assume that EEG is Gaussian. It might not be, but yeah, that’s the basic operation where you have to assume that my data has this specific form so I can use this model to solve it. For EEG data, often times, you have that Gaussian assumption. When you-
Kirill Eremenko: 00:17:48
So normally distributed data.
Asieh Ahani: 00:17:50
Exactly. It has a normal distribution but it’s multivariate. When you try to solve a multivariate, Gaussian model you have to solve it using least square method which is basically it’s not a linear-
Kirill Eremenko: 00:18:13
Yes, yes.
Asieh Ahani: 00:18:14
You cannot solve in a linear form. That’s why where you actually want to solve a Gaussian model using least square method, you need to calculate the inverse of the covariance matrix. Then your dimensions are higher than your observations, it means that the eigenvalues in your covariance matrix are mostly zero and it means that your covariance is not inversible. That’s when you have to reduce the dimension of the data to make sure that you can calculate the inverse of the covariance.
Kirill Eremenko: 00:18:47
Okay, and so you could use something like principal component analysis to reduce your dimensionality.
Asieh Ahani: 00:18:52
You can do that. Again, there are multiple ways that you can do it. We did regularization and shrinkage in this specific project. It’s all about which methodology gives you the best accuracy. For us, this methodology, regularization and shrinkage, provided the best accuracy.
Kirill Eremenko: 00:19:11
Mm-hmm (affirmative), how can you describe regularization?
Asieh Ahani: 00:19:15
Yep, so regularization means that you’re looking at again your covariance matrix. This is just an example. It can be any analysis of where you have a matrix and you want to regularize matrix. It means that you want to shift the value within that matrix toward a pooled in average so that your eigenvalues are not mainly zero or most of them are zeros. That’s regularization.
Kirill Eremenko: 00:19:48
Okay. Got you. I understand. Okay, and then you also mentioned you would build and classify. What kind of classifiers did you build in this project?
Asieh Ahani: 00:19:57
I have to remember. I think we did kernel density estimation which means that one is a kernel. If you plot all your data points in the two-vector plot, you can actually draw a circle around your different classes. That’s what called radius kernel analysis which means that I’m going to design kernels and the shape of the kernel is a circle. There are different ways that you can build kernels. It can be a stochastic or it cannot be non-circular. We used circular kernel analysis and based on that, you basically found the center of the kernel and the radius of the kernel. Then we trained our database on that kernel density estimation and then when you have a new data point, we just see where that data point can reside, whether it’s kernel one or kernel two. Based on that, we did the class.
Kirill Eremenko: 00:21:04
Okay, so you would use something like a radial basis function.
Asieh Ahani: 00:21:07
Exactly.
Kirill Eremenko: 00:21:08
Okay, got you. It’s interesting. A lot of this is coming back to me because we teach this in our machine learning course and we did kernel SVMs, for example. The RBFs was there.
Asieh Ahani: 00:21:20
Yup.
Kirill Eremenko: 00:21:22
Okay, cool. Thank you very much. That was a nice dive into the technical details and it’s interesting to hear that a lot of these machine learning concepts that you could be using for marketing or for business or for some other applications, predictive model maintenance and things like that can actually be used to understand better what’s going on in the human brain. That’s very, very interesting.
Kirill Eremenko: 00:21:47
One that we don’t normally encounter like for example business or in other industry applications is signal processing. That’s something that you did use. That probably comes from your electrical engineering background more than from machine learning. Is that right?
Asieh Ahani: 00:22:05
Yeah, I mean, obviously all industries, I feel like probably finance uses a lot of signal processing maybe a stochastic signal processing not linear or alter regressive but not in marketing which is my field.
Kirill Eremenko: 00:22:22
Yeah, if you could give us just a few words on signal processing, what is the essence and what is maybe some of the basic types of signal processing? I know for example, moving average is a type of signal processing but that’s really basic. What are your comments on that for people who haven’t encountered signal processing in their careers?
Asieh Ahani: 00:22:43
My experience have been mainly on deterministic and non-deterministic signal processing. I think that’s the highest level of classification. If it’s deterministic, then moving average is one way of modeling that signal. Autoregressive models are one way of modeling that and Kalman filtering is another way that you can actually model a deterministic signal.
Asieh Ahani: 00:23:10
Then when it’s undeterministic or a stochastic, then you go into again, you can use autoregressive based on a stochastic autoregressive analysis which has a stochastic component to it. Those are the type of the signal processing that I’m familiar with but mainly on the deterministic data and non-stochastic ones.
Kirill Eremenko: 00:23:31
Okay, well what’s the difference between deterministic and stochastic signals?
Asieh Ahani: 00:23:35
Yeah, so in a deterministic one you assume that the pattern in your signal doesn’t change in time. You’re repeating the same pattern. In the stochastic, you have the stochastic assumption that your data is not deterministic the time and is changing. Examples of that is weather, stock market, gas or oil prices.
Kirill Eremenko: 00:24:01
Heartbeat would be deterministic, its repeats.
Asieh Ahani: 00:24:05
Yeah, oh, EEG.
Kirill Eremenko: 00:24:06
EEG, also deterministic, right? There’s a lot of variety in it but if you were able to classify all the variety you’d be deterministic.
Asieh Ahani: 00:24:15
Yeah.
Kirill Eremenko: 00:24:16
Brownian motion would be stochastic, right? Because, you don’t know where it will go. You can’t predict where it’s going to go.
Asieh Ahani: 00:24:22
Yeah.
Kirill Eremenko: 00:24:22
That’s my guess. I’m not sure about that but basically, I get your point. Stock market would be stochastic. Thank you. Very, very insightful.
Asieh Ahani: 00:24:32
History will be a stochastic.
Kirill Eremenko: 00:24:35
History, well some people say history repeats itself though.
Asieh Ahani: 00:24:38
I know that’s not accurate. I think that’s a problem, right? We try to over simplify very complicated concepts. History is one. Actually, history is the most stochastic signal you can imagine because it’s a combination of a stochastic human data. I am a stochastic. You are a stochastic and our collective behavior defines history. It’s really the most stochastic data you can ever imagine because then the only reason I can make it deterministic is I can predict every individual human’s behavior which you can’t.
Kirill Eremenko: 00:25:21
What do you say when some psychologists say for example that or economists say that it’s really hard to predict the behavioral patterns of one individual human but it’s much easier to predict the patterns of a group of humans or the bigger the group, the easier it is to predict the group mind and how people are going to behave and think. Don’t you think that maybe if you put a lot of stochastic individuals together, in total, the combined system might actually be less stochastic than the individuals themselves?
Asieh Ahani: 00:25:53
Yeah, I definitely agree. Again, I agree for building analytics on human behavior and find a group like mentality of a group. As a society, where are we going and how different groups are interacting with each other. It’s very difficult to come of it prescriptive analysis.
Asieh Ahani: 00:26:15
I’ll give you a very so for example why a stock market is a stochastic. If I can build a very good model that will give you the price of Apple stock tomorrow, you’re going to buy and sell based on that assumption. Your buying and selling that stock will change that prediction. It’s by nature stochastic. Even if you can predict it, the action you make based on that prediction will change that prediction.
Kirill Eremenko: 00:26:45
Just like a paradox.
Asieh Ahani: 00:26:49
Exactly. The same thing goes with history. Even if we can cluster and then try to do something, people will then behave and react based on that and then it becomes unpredictable.
Kirill Eremenko: 00:27:02
Okay. Okay. Got you. That’s a very simple proof of the theorem. Very cool. Thank you very much. Well that was a very interesting dive into your research and your academic background. Let’s talk about your career. At which point did you decide to pursue a career of following a PhD in this space? Why didn’t you stay in academia? Why did you decide to move into the industry?
Asieh Ahani: 00:27:36
Yeah, I can say it was a combination of chance and confusion. I think that’s something that most PhD students deal with at the end of their PhD. They have this question of whether I want to stay in academia or I want to go to industry. I had the same dilemma as well. I too, a doctor was looking for post-doc positions because I wanted to stay in academy specifically in Europe. I wanted to leave US at that time. Then I can’t leave without having some experience in industry and I found an internship in Visual IQ which was the last company that I used to work before moving to MassMutual in Needham, Massachusets.
Asieh Ahani: 00:28:26
That internship experience was very good because I think I was leading one of the most important projects they had at the time not only to automate that but also to building algorithms. I found that data science has a good combination of everything I’m interested in: problem solving, programming and also not putting yourself in a specific academic box. You can actually have a client facing position. You can talk to clients. You should understand the business. There’s a lot of non-technical skills that I can learn if I work in data science versus if I stay in academia. That’s when I decided to move to industry.
Kirill Eremenko: 00:29:11
Awesome. What were you doing at Visual IQ? Are you able to share this project that you said was one of the important projects that you were leading there?
Asieh Ahani: 00:29:19
Sure, I was working on marketing attribution project. Every company uses different channels to market. You can use TV and radio. You can use paid search or paid social or organic social to advertise for your company. Mixed medium marketing and marketing attributions are about attributing your conversions as sales to different marketing channels and how to attribute that is very challenging because for example if you see a TV ad and you go and buy a policy, there’s no way that it can link that together, right?
Kirill Eremenko: 00:29:58
Mm-hmm (affirmative).
Asieh Ahani: 00:29:59
What are the modeling methodology that you can pick to attribute your sale to different marketing channels offline and online is a very interesting theoretical problem.
Kirill Eremenko: 00:30:11
Yeah. That’s very interesting. How did it feel moving from working on brain-computer interface and meditation like signal processing to marketing which is completely two different worlds even though some of the perhaps, a lot of the skills are transferable in data science but the underlying essence is different. How did you feel? I’m just curious.
Asieh Ahani: 00:30:41
It was definitely a change. In academy, a lot of concepts are very abstract. It’s not practical. Again, brain-computer interface is different. You know what you’re trying to do. You want to move a robot arm but most academic concept are very abstract but in industry and in marketing, things are pretty pragmatic. I want to increase my number of customers by 10%. I want to reduce cost in this part of operation. I want to retain my customers. How can I find customers that are most likely to lapse and I reach out to them. Very pragmatic, use cases.
Asieh Ahani: 00:31:25
Then often times, there’s no solution. Somebody tells you this is my business problem. Go figure out how you want to solve it and that’s where the interesting part comes in that you just have to come up with a hypothesis. You have to test that hypothesis and the last and most important part is to really understand how does this hypothesis will work out. Then you actually implement it in the business. It’s very different as it’s not only technical. There’s so many different factors in it that just makes it way more challenging, at least for me.
Kirill Eremenko: 00:31:59
Okay. Understood. All right and after Visual IQ where you did the internship and you quite quickly went from intern to data scientist to senior data scientist in a matter of nine months and then five months then you moved to MassMutual where you are the head of data science in the Customer Journey Division. Tell us a bit about that. Maybe, let’s start by tell us about MassMutual. What is the mission of this company?
Asieh Ahani: 00:32:30
MassMutual is Massachusetts Mutual Financial Corporation. MassMutual provides a variety of financial products. Most people know MassMutual because of the life insurance product which is one of the biggest bucket of business for MassMutual but MassMutual provide variety of financial products such as disability insurance, annuity, different financial products and also retirement services.
Asieh Ahani: 00:33:05
I think MassMutual as a whole is first of all is a mutual company so an old policyholders have a share in the company. Actually, that was one of the reasons that I accepted the offer. To me, not having a private owner or even working in the public company, is actually very interesting that you’re actually more-
Kirill Eremenko: 00:33:27
More fulfilling.
Asieh Ahani: 00:33:29
Yeah, you’re working for your policyholders. I think the mission of the company is really great because these financial products help people to protect the ones they love. College can be afforded. Retirements can be planned. I can plan for the worst case scenario which is disability or death. To me, that provides a service to the society that is actually much needed and I really like the work that MassMutual is doing for the society.
Kirill Eremenko: 00:34:02
Yeah, and it’s a very big company. I’m just looking up the total of 7,000 employees in the US, a total of internationally 10,600 employees, was founded over about 170 years ago and the revenue in 2016 according to Wikipedia was or, no, according to the MassMutual website was 29.6 billion. It’s a huge company and it would be a very responsible role to be the head data scientist in one of the divisions of the company.
Kirill Eremenko: 00:34:41
Tell us about your role maybe, how it’s structured. We were talking about the podcast at the start. You have data science and data analytics and then in data science you have three domains. Can you paint this picture for us, please, so we know exactly or roughly how it all works?
Asieh Ahani: 00:35:00
Sure. We are a part of the DSA Department which is Data Science and Analytics. It includes data governance, data enterprise architecture, data science and data analytics. My department which is data science and data analytics include both analytical teams and data science teams. Data science itself is divided into different domains depending on what type of projects we accept and who we work within the company.
Asieh Ahani: 00:35:37
We have cybersecurity and the folks that are generally working in security and finance, beside security domain. We have investment and finance data science domain and those folks mainly work with investment management team and the finance department. We have risk and product domain and they usually work on mortality models, pricing, actuarial and all the different underwriting processes which is a big part of MassMutual business.
Asieh Ahani: 00:36:09
My domain is customer journey. We are involved in every level of customer journey from awareness, consideration, purchase. After we have a customer we have all these post issue operations, retention, loyalty, making sure the contentment is high and also cross sell opportunities for our existing customers and finally claim which is basically that post issue operations efficiency and also marketing can sell before we want to attract the customer.
Kirill Eremenko: 00:36:45
Wow! That’s a huge area of responsibility. You would work quite closely with the marketing team, the supports team, well, the marketing team and the support team, right?
Asieh Ahani: 00:36:56
Yes. We almost work with every part of the organization. I mean it’s great but it’s also a bit challenge because we are stretched. They have to work with everyone so the prioritization and making sure that we focus on the most important business problems is really key. Otherwise, we will be buried under a lot of projects.
Kirill Eremenko: 00:37:17
Okay. Got you. How many people do you have on your team?
Asieh Ahani: 00:37:27
I didn’t move to MassMutual as the head of data science. I was a lead data scientist and individual contributor for two years before I moved to my current position as the Head of Data Science Customer Journey. I started a team of at that time, I think six. Now we are 13 within a year.
Kirill Eremenko: 00:37:45
You grew pretty well.
Asieh Ahani: 00:37:46
Yeah.
Kirill Eremenko: 00:37:49
That’s impressive growth and it’s also impressive that with just 13 people, you can support an organization that employs 10,000 people. You must be so busy.
Asieh Ahani: 00:38:00
We are very busy.
Kirill Eremenko: 00:38:00
Okay. Okay. Got you. I have quite a few questions here. The first one would be on your progression. It’s impressive to see how even starting for Visual IQ, it took you seven months as an intern, nine months as a data scientist, five months as a senior data scientist. Then MassMutual two years, three months as a lead data scientist and then Head of Data Science Customer Journey in which you’ve been for just over one year. Very rapid progression through your career from an intern to individual contributor to senior to a new company to then head of a whole division for a giant organization.
Kirill Eremenko: 00:38:54
Can you tell us how did that feel? Sometimes people stay in the same role for a decade or maybe five years but you were constantly going up and up and up. What was driving you and what are the positives and the negatives or the pros and the cons of such rapid career growth as a data scientist?
Asieh Ahani: 00:39:19
That’s an interesting question. Actually, those are the things that I think about constantly myself especially when you move from a technical role to a management role. That’s the big change and that means that you have to say good bye to some of your technical skills. It’s inevitable, right? We are in a very highly technical field. I think 88% of people in data science have a post graduate degree or some sort of master’s or PhD. It’s a very highly educated and highly technical field. It’s very important to have your technical skills. When you move to a management role, you know that it’s inevitable that you might lose some of those.
Asieh Ahani: 00:40:01
I would say I don’t think that’s definitely a concern or a bad thing but it’s the reality that you have to accept and you have to be prepared for. But there are much more to learn. Gaining all the softer skills, understanding how to delegate effectively, how to lead effectively, those are really things that you won’t learn until you become a technical lead or people manager. Communication, learning how to listen, being more nurturing, I think a lot of skills that you can actually use in your regular life. Those are the good things in moving up into your career because you’re learning that. I think that you also have to be aware of as your responsibilities change, you can not gain a lot of expertise in a specific area which you can use that to your advantage in your career.
Kirill Eremenko: 00:41:08
Okay. Interesting. Do you ever regret the move into management and do you miss the ability to just poke around in an algorithm and spend three days building a regression or something like that?
Asieh Ahani: 00:41:23
I don’t regret but I miss. That’s the different answer to the different adjective I think. Again, I don’t regret it because I’m learning and growing definitely but I do miss technical work a lot. I mean I’m coming from years of technical work and just moving to a management position has been challenging.
Kirill Eremenko: 00:41:48
If somebody is in a similar dilemma as you were, is there anything that can be done for a manager to still maintain this technical expertise and not give it up completely?
Asieh Ahani: 00:42:08
Yeah, I, for example, every day after work, I solve an algorithm problem. I use different softwares. For example Leetcode, or HackerRank, different sites that you can actually go and solve an algorithm problem. To me, it’s a good exercise because first I use Python so I practice my programming. Also, it makes you engage in actually solving an algorithm problem.
Asieh Ahani: 00:42:37
I’m thinking about maybe also including building a model or solving your modeling problem as there. I know Leetcode and HackerRank don’t provide that opportunities but there are other websites that can actually go and solve a simple data science problem using Python or R. That’s something that I personally do after my work.
Asieh Ahani: 00:43:00
At true work, I think there are managers who do both technical and management work. I think you in the size of my team, I haven’t had that option yet but maybe in the future I can manage the team in a way that I can take some of the technical responsibilities.
Kirill Eremenko: 00:43:19
Okay. Amazing. Speaking of managing teams, can you share some things that you’ve learned managing a data science team of six and then 13 people? What are some of the main challenges and what are some of the tips you can share with people who are considering becoming managers or who have recently become managers?
Asieh Ahani: 00:43:43
Yeah, I think the most important thing is really first as a manager, learning the increasing your business acumen and understanding the business as best as you can. Then the second layer is to develop your team to also understand the business. I think that has been a challenge especially when you have a company that is a bit huge. You have 7,000 different people working in the company, different types of business, different financial products, really understanding the business is really key because you can actually frame the work of your team to be aligned with the strategical goal of the company.
Asieh Ahani: 00:44:31
Then the second layer is to make sure that your team also understand the business and have that vision and direction to I don’t want to work on hundred different things, let’s understand what are the big business questions. Is it a strategic question for the company and let’s focus our attention on that. That to me has been the biggest challenge and I’m learning even now that I’m trying to develop my team and push them toward that direction as well.
Kirill Eremenko: 00:44:59
Okay, and tips?
Asieh Ahani: 00:45:06
Tips, so generally I think on that business acumen side, building relationship and communication is key. Also, building your influence. Basically, building relationship with the right people and asking the right questions on coming up with effective solution is key. But also, when you’re talking about people management, I think communication is really important but also being empathetic to people’s need especially this year, it has been very challenging to be motivated and keep people motivated. You have to know how to manage and not only you have to work on yourself but you have to make sure your team is motivated. It’s difficult when you have a team in New York, they’re all just in one-bedroom apartments trying to work from home.
Asieh Ahani: 00:45:58
Knowing how to listen and how to relate to people’s problem has been key. I think that’s very important to manage people and also at the same time, making sure that you push people to be the best version that they want and keep them motivated and motivate and work on important project.
Kirill Eremenko: 00:46:21
Would you say that management in data science is for everyone or some people should really consider it and decide if that’s what they want?
Asieh Ahani: 00:46:31
It’s hard for me to tell. Honestly, I think humans are capable of learning anything. It’s difficult for me to say that it’s not for everyone. I think if you are interested, you can learn the skills to be an effective manager. It all goes through whether you’re interested or not. If you’re interested, then you can learn and you just be motivated to learn.
Kirill Eremenko: 00:47:01
Awesome. Got you. Thank you. And you also mentioned before the podcast that well we’ve got an exciting surprise for our listeners. You’re hiring at MassMutual specifically into your team. Tell us a bit about the roles and the candidates that you’re looking for. If someone listening to this podcast is interested, would be able to go and apply.
Asieh Ahani: 00:47:20
Yeah, thank you so much. This is actually a lead data scientist position. We’re looking for somebody with experience. I think the job requirement is seven years of experience but we include some of the graduate degree as a part of the work experience.
Asieh Ahani: 00:47:38
I definitely want to have a diverse set of candidates coming in and I want to make sure that I want to emphasize on people who are familiar with the industry but also I don’t want to exclude anyone who is not familiar because I think people can learn the industry very well if they are really technically as strong.
Asieh Ahani: 00:48:03
This position is basically lead which means that the candidate has to lead multiple projects at the same time. It’s very important to be agile and also be accountable. Also, there’s an opportunity to lead junior data scientist because as the portfolio of data scientist grow and they have multiple projects and competing priorities, it’s very important to assign junior data scientists to those projects. There’s an opportunity to also exercise, develop their leading … learning how to lead people.
Asieh Ahani: 00:48:45
It’s an exciting opportunity and I would appreciate if people can apply and please reach out to me on LinkedIn if you are interested.
Kirill Eremenko: 00:48:54
Okay so that’s the best place to reach out to you directly on LinkedIn.
Asieh Ahani: 00:48:57
Yes, I don’t have any other social media.
Kirill Eremenko: 00:49:00
Yeah, no, no. That’s okay. I guess people should just put a note that they heard about this on the podcast so you know where they heard it from. Does that person have to be in New York?
Asieh Ahani: 00:49:14
Not really. I think New England is perfect. Generally, within a state of Massachusetts or in New York. Just because most of the team are either in Boston, Springfield or in New York but given the current climate, I don’t want to exclude anyone.
Kirill Eremenko: 00:49:34
Awesome and speaking of the current situation, how have you found, oh, before I continue. Anybody listening, if you are interested, make sure to find Asieh on LinkedIn and apply directly with her.
Kirill Eremenko: 00:49:52
I was asking, speaking of the current situation, how have you found it leading a data science team specifically a data science team, I know a lot of managers are learning a lot of new things in this coronavirus time, leading remote teams, but what are some of the specifics around leading a data science team remote?
Asieh Ahani: 00:50:15
Oh, it is challenging especially where you brainstorm and you want to actually formulate or write an equation. It has been pretty challenging. Personally, there are tools that I can use. For example, we use these tools that you can actually write on a notebook and then you can share that notebook. It looks like a regular notebook but they can connect that to a computer and it’s like you’re writing on the screen. That’s good for brainstorming but again, I think the biggest challenge is that communication and keeping people motivated. It’s very difficult to do that when you’re not sitting in the same office or you cannot have lunch with your team or go for a coffee after work to make sure that you have that relationship.
Asieh Ahani: 00:51:06
I had to onboard two people. We hired two people in the middle of pandemic. I have never seen them. Two people on my team, they were onboarded digitally. For them, I feel like they often might feel isolated because they never saw me or anybody else on the team versus somebody who used to work before the pandemic. Those are challenges that I face every day. We just have to adapt.
Kirill Eremenko: 00:51:41
When your data science team works on a model, do two of the same people ever work on the same code? How do they share that especially in a remote environment?
Asieh Ahani: 00:51:58
Yeah, so we do use GitHub. It’s easier to actually work on the same code when you’re using things like GitHub or different platforms that can share codes. I think the challenge is mainly when you are actually brainstorming. You’re trying to ideate. You think, “Oh, how about this?” You want to easily walk to the board and start writing that down and that’s not an option when you’re working remotely.
Kirill Eremenko: 00:52:29
Got you. All right, I can imagine you’re busy managing this massive team and in the evenings, after work you also solve these data science challenges. How do you maintain work-life balance?
Asieh Ahani: 00:52:44
Good question. At the beginning of pandemic, I can say that I wasn’t very much aware of all the changes and all the pressure. I think I focused a lot on work and I tried to basically use work as a distraction to everything else that was going on. I can tell that that’s not an effective way of doing things. Actually, it’s important to understand your challenges, think about them and come up with solutions outside of work to fix them.
Asieh Ahani: 00:53:20
Once I got to that point, I tried to really set up time that I’m going to close my laptop. There’s no way that I’m going to open the laptop after this specific time. I try to avoid working on weekends no matter what except when there’s some sort of a deadline on that. I try to schedule a time to run by the rural outside or different things outside of the work so I have a full schedule of things that I should do beside work to make sure that I have that distance. It’s key.
Asieh Ahani: 00:53:57
I think people are in the same situation as me that they’re using work as a distraction especially now that we’re working from home. There’s not really a line between your life and work anymore. You wake up and you start working. It’s very important not to get into that mindset. Work is not a distraction from a stress, anxiety because of pandemic or economical situations. It’s very important to find different hobbies for that.
Kirill Eremenko: 00:54:28
Awesome. How is that going for you to close your laptop at a certain hour? I’m trying the same thing but it’s so hard. You close it and then you’re like, “Oh, I still have to finish this.” What’s your success rate so far?
Asieh Ahani: 00:54:49
I have this progress. It’s getting better. I think it’s important to have a full schedule for other things. I signed up for yoga classes, make sure that you have an alarm for running or exercising outside. I signed up for some galleries in New York that I want to see even in the week day which I didn’t used to. I scheduled the rest of my day to make sure I definitely have something to do that I have to close my laptop.
Kirill Eremenko: 00:55:16
Awesome. That’s good. That’s good advice. One very important topic I think we should touch on is women in data science. You’ve had tremendous success. Were there any obstacles along the way and what can you share with our female listeners who are excited about data science but they can see that this is indeed like a lot of STEM is a male dominated space. What kind of encouragement can you share with them?
Asieh Ahani: 00:56:01
Yeah, I think that’s something that you realize very early on when you go into tech that there are more men than women. I just think that so you can’t deny that but I don’t think you should get it as a roadblock or a problem either.
Asieh Ahani: 00:56:23
The most important thing is to have confidence and not only confidence to speak up but also confidence to ask for help. I think as a woman, you don’t have to have insecurities about asking for help, asking for guidance because you think that well what if people judge because I’m a woman and I cannot perform as well as a man. These are all in our heads.
Asieh Ahani: 00:56:47
Really working on those insecurities, making sure that you believe in yourself, you’re speaking up, not only in meetings but also with your team, not being afraid of saying, “I want to meet with this influential person in my company,” or questioning the status quo that I think this project is not important. We should divert our focus to this other project. Having that confidence to speak up is very important and also I think you just have to have a passion about what you’re doing and not being afraid of making mistake. People do mistakes. You just have to learn from them and move on.
Asieh Ahani: 00:57:32
I think again women have maybe more difficulty dealing with mistakes because they think they are being judged mainly more harshly which is a reality. I’m not trying to say that it doesn’t happen but we should basically say, “I’m going to do my best. There’s going to be mistakes along the way. I’m just going to learn from that.” Be accountable for those mistakes and just move on. Most importantly, to make sure that you play fair and square. Again, at the end of the day, it’s all about confidence.
Kirill Eremenko: 00:58:07
I love your comment about asking for help because everybody asks for help. I ask for help all the time. Everybody regardless of your gender or background, you should be comfortable asking for help and if you’re being judged for asking for help then you should ask yourself is that really a company you want to be working for.
Asieh Ahani: 00:58:31
Exactly and actually, you can lead by example. You can basically say, “I’m a woman leader and I’m going to speak up and ask for help.” You actually lead by example to all the other women in the company that it’s not a sign of weakness. If you need help, you can ask for it and then even if the company doesn’t have that culture, you help building that culture for the company.
Kirill Eremenko: 00:58:54
Yeah, that’s great advice. You spoke about confidence, having the confidence to speak up, to challenge what the company is working on, through the confidence to ask for help. How do you build that confidence?
Asieh Ahani: 00:59:14
That’s a tough one. I was trying to think about how, again, I don’t think I have gained that level of confidence that I want to myself. It’s a day to day struggle, I would say. I think being self-aware is very important, understanding your weaknesses and strength and just double downing on your strength and also work to develop your weaknesses.
Asieh Ahani: 00:59:43
Once you see yourself that you’re actually growing and developing, you are going to build that confidence. I am still trying to build that for myself because I am self-aware about my weaknesses and I’m trying to improve on those. When I see progress, I’m like, “Okay. Good. Now I can just be more confident in myself.” Just self-awareness I think is the key to build that confidence over time.
Kirill Eremenko: 01:00:12
Yeah, that’s a good example or good advice. Self-awareness to build confidence. Asieh, I have one more question for you before we wrap up and that is where do you think the future of data science is going for people who are looking to build a career in this field, what should they be preparing for in the next two to three years?
Asieh Ahani: 01:00:48
I think the work is getting more and more data-driven and the key is that now people are generating more data. Everybody and all the population and most of the population are generating data every day. The data is only growing. Given that the data is growing, the need for data scientist and data analytics and data engine only going to grow. I don’t think that there is any limit really in the demand for data science in any time in the near future. Some of the fields might get a different name. A lot of data scientists are now machine learning engineers or machine learning researchers. The name changes but really what you do doesn’t change and the demand for this field is only going to increase.
Asieh Ahani: 01:01:40
That’s good that there is demand but given that now all the industries from health care to business management to finance, to government, everyone is going to need data scientist and data analytics resources, it’s very difficult to know okay, what skills do I lack because based on the industry, you need different sets of skills. If you work on finance, you have to give very good technical, you have to be very fluent in statistic processing. If you want to work let’s say for AWS in Amazon, you have to have a very foundation on infrastructure knowledge although you’re doing machine learning on top of that data.
Asieh Ahani: 01:02:26
My advice is that choose a field and in that field, make sure that you are always on the edge of that industry and that field. You’re pushing the envelope and you’re questioning the status quo and you are always on the border of that specific industry given that it’s very difficult to be a knowledge expert now in data science in every possible field.
Kirill Eremenko: 01:02:52
That’s the researcher speaking in you, pushing the borders of what humanity knows. Great advice. How do you choose a field?
Asieh Ahani: 01:03:06
I think a lot of that comes from graduate so college. For example, if you’re very much interested in pure statistics and pure probability theory, you know that you most likely want to be a researcher I don’t know either in Microsoft or go to finance and doing quantitative finance or very theoretical mathematics.
Kirill Eremenko: 01:03:34
Or actuarial maybe.
Asieh Ahani: 01:03:35
Exactly, exactly. Really focusing on those field is more important and maybe finding an internship in those fields so you can examine that actually that’s what you’re looking for. Generally, taking advantage of internship through college is really important because that’s the time that you can actually work in multiple industries and find the industry that you are interested in.
Asieh Ahani: 01:03:56
To me personally, if I never had that internship in Visual IQ, I wouldn’t be where I am. Before, I had no idea about the application of data science in marketing and I just learned from that by that internship. That’s very important. I think that’s a way that you can actually have different experiences on pick a field and just double down on that field.
Kirill Eremenko: 01:04:19
Fantastic. Asieh, it has been a pleasure. It’s been super exciting to speak with you. Before you go, please tell us what are the best places, you already mentioned LinkedIn. Is that the best place to get in touch with you for our listeners?
Asieh Ahani: 01:04:32
Yes, I mean, I don’t have social media unfortunately, so I cannot provide a Facebook or Twitter account. LinkedIn, it would be the best for reaching me.
Kirill Eremenko: 01:04:43
That’s perfect. That’s perfect. And one final question, just a quick one. What’s a book that you can recommend to our listeners?
Asieh Ahani: 01:04:51
Yeah, for data science, again, I’m really interested in the theoretical aspect of data science. For me, the go-to book has always been the Pattern Recognition and Machine Learning by Christopher Bishop. Honestly, I’ve never had a question that I couldn’t find answer in that book. It’s really comprehensive from graft theory to neural network to basic classification and regression analysis.
Kirill Eremenko: 01:05:19
Got you. Pattern Recognition in Machine Learning by Christopher Bishop.
Asieh Ahani: 01:05:25
Yes.
Kirill Eremenko: 01:05:27
Awesome. Well Asieh, thank you very much for coming on the show and sharing your time and insights with us.
Asieh Ahani: 01:05:34
Oh, no problem. Thanks for having me.
Kirill Eremenko: 01:05:41
Thank you everybody for sharing this hour with us and I hope you got a lot of value out of this podcast, got some useful insights and we’re excited to hear about Asieh’s story, how she went through academia and have a bachelor’s, master’s and PhD and all the exciting research that she’s done and also that move to industry and how she is progressing through her career and leading people in the space of data science.
Kirill Eremenko: 01:06:11
My favorite part personally was our discussion about stochastic versus deterministic signals. It was a good refresher of that world and we don’t often talk about it in data science because for instance like in marketing or in many other areas of data science it’s just you just don’t work with signals but it was exciting to see that you can use signal processing both for in computer interface and at the same time you can use it in financial modeling. There are areas where data science will overlap with signal processing and it’s important to be aware that these are two separate or slightly separate domains that you might actually be interested in if you’ve never tried signal processing or you want to look into it, that might be something you’re passionate about or if like Asieh, you’re coming from an electrical engineering background where there’s a lot of signal processing, those skills can be applied in certain areas where data science is also used.
Kirill Eremenko: 01:07:11
As always, you can find the show notes for this episode at www.superdatascience.com/415 that’s www.superdatascience.com/415. We’ll link to Asieh’s research papers there. You can read them. Also, Asieh said she’ll send through some more information on curse of dimensionality and some other things that we spoke about. That might be useful to you too. Of course, there you’ll find the URL to Asieh’s LinkedIn where you can connect with her and specifically if you’re looking to apply for that job that they’re currently hiring for, I would encourage you to connect with Asieh and let her know. As she mentioned we’re, they’re, not we’re, there at MassMutual they are looking for an advanced data scientist or lead data scientist.
Kirill Eremenko: 01:08:04
If you enjoyed this podcast and you know somebody who could be inspired by it, somebody maybe who’s in the world of academia or who’s excited about signal processing and is considering data science, send the link to them. It’s very easy to share the podcast, just send the link www.superdatascience.com/415. On that note, it was very exciting to have you here and I look forward to seeing you back here next time. Until then, happy analyzing.
