I have to say before the program started, I was still wondering whether my choice of MSBA as my career transit was correct or not. The name– Business Analytics — seemed to imply the “stuck in the middle” trouble: without a strong coding background, we could not compete against the Computer Science students for a technical position; lack of deeper development into business as MBA students, we might not be capable to bring business value to the companies from high level.
However, after 2 quarters’ study, I found the value of being a business analyst: as the linkage between business and technology, we knew more about what the company was looking for and how machine learning and other advanced technique could help with solving the problems, and what metric could be useful to measure the outcome. In addition, the practicum project was the best bridge that links the academic and professional career: by going through the whole analytical project process from kick-off to model implementation, the practicum project provided us with the unique chance to act as a real “business analyst” and correct my previous misunderstandings of what a business analyst should do.
Misunderstand 1: You need to think carefully from all aspects before we build the model
Before we started the model building process, we planned to spend at least 31 month to capture as many features that might be useful as possible so that we could evaluate all of them at once and try not to leave anything behind. However, we later found that this was not a good idea, or even, impractical to do so in the real life since you could never capture all information you need to build a “perfect” model at once.
Excellence in analytics is all about speed.
— Cassie Kozyrkov
As Cassie Kozyrkov stated in his article, whereas Statistics focused on rigor and Machine learning focused on performance, the analytics were responsible to surf vast datasets fast and pick up those that may be potentially useful to dig out insights. What the best analysts are looking for are the stimulators, those can inspire the decision-makers and may be worth for further investigation. There is no limitation of how many attributes we need to add into the model, or penalty if we choose the wrong features at the first time.
In addition, the wide spread of agile development also indicates that the world evolves very fast that things may change before we find all information we need to build the model. Sometimes the underlying products may be abandoned by the market before we build up the perfect model. As a result, the model built through iterative approach which means we keep on adding variables into the model may be more valuable even through it cannot cover all features at the beginning.
Misunderstand 2: Model building is more important than features selection
I used to believe the most difficult but important part of an analytical project was to apply the most up-to-date machine learning techniques and advanced algorithms to achieve a better model performance. As a result, when we pushed the “run the code” button the first time and saw our model performance was lower than expectation, all we did was to search for the codes of other models such as random forest and XGboost, and consult our Teaching Assistant (“TA”) whether there were any new machine learning techniques that could be applied to our case.
There is no doubt that the development of machine learning and deep learning can improve the algorithm and achieve a better model performance. However, the model improvement is built on the assumption that the features we select into the models are valuable. If we only employ the noise instead of identifying the true signals to build a predictive model, then everything is in vain even if we apply the most powerful algorithm.
Your data won’t speak unless you ask it the right data analysis questions.
Similarly, if you choose the wrong data, you will not get the answer to the business questions. The process of finding the data we need is not just one-way: when we see the model performance, we need to revisit the attributes we select again and again, to check whether the outcomes match with the hypothesis we made before, and if not, figure out the reasons behind and whether this abnormality indicates potential risk or opportunity; in addition, we also need to keep on adding new variables that may be useful, as mentioned above. The CRISP-DM (Cross Industry Standard Process for Data Mining) Model also emphasized the importance of “moving back and forth between different phases is always required”, especially between data preparation and modeling, until a “good enough” model is found.
Misunderstand 3: As a data analyst, all you need to do is to improve the model performance
I used to think as a professional analyst, the clients should trust our professionalism and follow our recommendations: just like the refrigerator, you do not need to dive into the rationale behind to figure out how it is functioning before we use it. Our sole goal is to improve the model performance and s the beautiful statistical results themselves can explain everything.
However, storytelling is the most important part for a business analyst, especially when our clients have no ideas of what they want or how to use correct metrics to measure the outcome. In this case, even you figure out how to build a sound model that can achieve high performance, without persuading our clients it is what they are looking for and can bring business impact to them, the model is just “nonsense” to them and can never be implemented or deployed.
Therefore, we should put ourselves into the business partners’ shoes and start thinking from their perspective. Try to practice the presentation with our grandma instead of our teammates or peers — that means we need to explain in plain English instead of applying jargon all the time. We cannot assume all our target audiences have the same background with us and can understand every technical term without further explanation.
Look Forward
I’m glad to involve in the practicum project, to apply what I have learnt during the class immediately into real life case. Even the barriers and the mistakes we made seem to be lovely since they train us into a more powerful analyst that we know how to handle with similar challenges next time. The experience of working in a diverse team with different background, gathering different ideas from all sources we can think about, and collaborating with each other to reach final solutions are all valuable.
I believe the practicum project is indispensable during my path to become a good business analyst and I hope this blog may bring value to you too.
