Subscribe to DSC Newsletter

Why bother building a model while you have a good human doing the work?

Stunned by the Senior Management's basic and fundamental question, I spent a lot of hard time thinking through it. Finally I get part of the ultimate answer: Predictive Analytics move faster and more frequent than human being. What I mean is:

Predictive model is run to react to every change of independent variables: Building a good model of relating consumer's purchase behavior is only part of the story. What is more important is the trigger of the model can be defined and run at a specific time, more often than human being able to recall and think of. For example, building a predictive model for chocolate is good, i.e. festivals or broadcasting time of certain popular TV shows. What is more important is to also to run the model in due course. In my works as the investment banking professional, we need to monitor the macro-data change to make sure that  every change of the data will generate different results.

And additionally, it also enable management to view the macro-view of their business. While models can be built at the consumers/customers level, the same macro-variables, e.g. festivals or broadcasting time of popular tv shows, can also be used at a factors in the financial models, enabling management to react/counter-act to the portfolio performance. If those positive triggers can be predicted/ are known well-ahead, e.g. 5 shows and 10 festivals, each events lift the propensity to buy by 5 times, the management can calculate the yearly performance of the supermarket. With the data, managment can try to alter the prediction by improving their offer, thus boosting the performance further i.e. lift the 5 times rise of propensity to 6 times.

Guys, help me out in this, needa think really really hard to explain why a robot is better than high quality human being.

Views: 646


You need to be a member of AnalyticBridge to add comments!

Join AnalyticBridge

Comment by Shankar S on April 30, 2015 at 3:23am

Greetings, Nice article and discussion.

Just want to add a point to the side favoring models over human expertise: its laying the infrastructure for fine-tuning the model(s) to suit other scenarios/related problems etc. Once the basic process is setup, it's easy/faster to iterate many times over to solve similar problems. Maybe after 6 months, you "feel" you have a better model now than what you used "then". You can use the down time (or an alternate machine) to re-do an older analytical decision making suggestion (re-do predictions using the newer model) and check if it would indeed have been better. Most mining/analytic solutions support keeping multiple models in play instead of replacing/over-writing older models with newer ones (for same business problem)... Helpful for versioning as well as facilitating competition b/w model versions.

Comment by Jeffrey Ng on April 8, 2015 at 9:18am

can you look for a smaller problem to build up your credential? Your problem seems to be two-folded - you lack the credential internally; and secondly, the management lacks the quality to respect the model. this is two-sided. think in their shoes too: if you were them, how would you risk your career in the hands of someone who told you unfamiliar things?

1. argue that the competitors are doing it and keep fighting - you will get to nowhere. i truly understand your problem and it seems so hopeless when you are facing those challenges from above.

2. find a smaller problem first - find some neglected and boring problems that no one wants to deal with. that's what data model is very good at. boring, repetitive works. no one will fight you solving this boring problem. no one will blame you either. there is nothing for you to loss.

Comment by Thomas Lincoln on April 8, 2015 at 7:05am


I have done all the operational data collection for my division.  I did it better than the IT department in their data warehouse.  I understand the operational issues having an operational undergraduate degree and having worked in operations for 8 years.  I have built operational and financial models before, and all were correct, but weren't what sr mgmt wanted to hear, nor what they believed.  5 years later, my company is still arguing the same issues, about how saved time can't be stored up and used all at once at a later date.  And when mgmt doesn't value models because they are accountants, or have confirmation bias towards what their boss says and believes to succeed, or are applying the wrong business model, its a behavioral issue, not a math nor modelling nor data issue.

Comment by Jeffrey Ng on April 8, 2015 at 5:31am
I can share what you can do, if you truly believe in the values of models: 1. Build it yourself and evwn if u can get it off the shelf. This will help you to collect what data is important to the models. 2. Collect, store and report data on operational issues. This build up your knowledge of the data avaliable to the organization. Do build a mental model to organize the data across the organization. All these will be the building blocks for u and the organization for models building in the forseeable future.
Comment by Thomas Lincoln on April 6, 2015 at 7:42am

Actually, its more of a human behavioral issue that models can't be manipulated like people, therefore a robotic model is more risky, and less trustworthy to give the executive what he wants to hear, versus what he wants to know. . .   I see it all the time, and hear about the same in many different companies from friends of mine. . .   good luck!

Comment by Jeffrey Ng on April 6, 2015 at 6:58am
My experience told me that u need to find the less senior one but denior enough to believe in data and model. I waited 2 years before I got my first internal client, then it will spread like wild fire.
Comment by Robert Rouse on March 11, 2013 at 8:17am

I have a similar situation going on right now where we're trying to gain buy-in on a high-powered predictive modeling system for our coal plant operations.  For a host of reasons, we have to maintain a forward look on the fuel that's delivered and marry that up with dozens of other variables to predict generation and emissions levels, to name the two main ones.  

Currently it's done by humans - more specifically, humans running a spreadsheet-based modeling tool that can only realistically be run once a month. So far that works fine but new emissions requirements will lead to a need for up-to-the-minute predictive analytics.  
The sheer volume of variables and complexity of the models involved (7-10 billion calcs every time) mean that you either hire a few hundred more humans or you get a better modeling platform. 

And, for execs - that's how you sell it.  It simply costs less to "hire" an expensive computer system to do exponentially more complex and highly repetitive modeling versus hiring more people to run their own laptops.

On Data Science Central

© 2021   TechTarget, Inc.   Powered by

Badges  |  Report an Issue  |  Privacy Policy  |  Terms of Service