It's a quiet day at the Hippalus inc. like it was three months ago when the amazing drop in revenues was found. You are sitting at your desk looking at the intranet news when suddenly an email pops-up in your inbox:
You ask some of your colleagues if they know what is going to happen during the meeting. Nobody actually knows, and since the time is getting closer to 9, you decide to get to the meeting room. At the time you get there, a small chat is going on, but as soon as you come into the room the chat almost stops. And it definitely stops once Andy also gets in the room.
Dear colleagues, I have suddenly organized this meeting to share with you some good news. A sound of relief fills the room as soon as Mr. Sheen completes this pleasant sentence.
I was recently notified that the investigations on the revenues drop came to an end, and that end was a bad one. There is actually no great meaning in this talking, nevertheless, the audience is confident that more news will come in the forthcoming words from our boss.
Mr. Clough called me yesterday to provide an update on the status of the investigations and its outcomes. He repeated once more his gratefulness for the effective support we provided during their activities and wished this could be the first episode of a long history of collaboration between our two departments. Well, let’s hope this not going to become a soap opera!
The sincere and rigorous laugh welcoming this joke explains, more than a thousand words, how respected and feared at the same time Mr. Clough is.
Nevertheless, we have to admit that on this occasion our collaboration was actually constructive and helpful for the whole company. To those of who are not aware of what the investigation was about, I will briefly summarize it.
Nearly three months ago, we received an alarmed communication directly from the CEO showing an incredible drop in our revenues. It was really a surprise for the CEO and for all the company, and we were immediately engaged to discover the reasons of this epic fail.
Suddenly, it became clear that we could have not leveraged the reconciled data from the company data warehouse, which at the moment was still in the development phase.
We had, therefore, to perform the analyses the hard way: gathering, cleaning, and merging data sources from different legacy systems, trying to get ahead of that mess.
Having cleaned the data, we gained our first win: from the clean series of profit data divided by region, we discovered the drop coming from the Middle East area. This produced a kind of relief and pain at the same time, and also the involvement of the internal audit department.
We commonly agreed that the best possible way to move onward the analyses without raising any suspicion in people possibly involved in the story, was to leverage the existing data on the history of our customers to define a list of Middle East customers more prone to going into default.
And this was what we did: we applied data mining models to our data, to better understand the phenomenon and discover the attributes most predictive of a possible default. If I am right, we applied linear regression, logistic regression, and some other evil stuff… we applied random forest as well, didn’t we?
After receiving some kind of confirmation from one of his colleagues, Sheen starts talking again to the small crowd which is now sitting in the meeting room.
Those models, which they say we also combined together by means of ensemble learning techniques, were shown to be quite powerful in explaining the reality we were facing. We, therefore, moved to the next step: predicting which customers from the ones pertaining to the customer list of the Middle East, were more probably in default within the last year.
We shared this list with Mr. Clough’s department, who employed it to perform another relevant analysis. To be honest, they have done some cool analysis as well.
The internal audit basically collected customer cards for all companies we predicted being defaulted, and applied, to the text extracted from these documents, some powerful techniques of text mining.
Two big results came from the analyses. The first was related to the soundness of our prediction. The comments from our commercials showed that those companies we predicted as being defaulted were actually experiencing a really bad relationship with us. But the most surprising discovery coming from their analysis was the one obtained by exploring the network of companies and shareholders. After representing this network, the internal audit discovered that the most part of the probably defaulted company was related to a shareholder named Omid Tahvili.
I think some of you already knew this part of the story, but probably none of you knows what comes next.
The mixture of increasing silence and the sound of astonishment produced by this last sentence confirmed to Sheene that no one in the room was actually aware of the final developments in the story he was going to disclose.
After receiving a detailed report on the results of the analyses, Clough decided to break the tie and directly reach our subsidiary in the Middle East, to conduct by person the required file work, gathering evidences and interviewing involved people. What? Was he armed? No, I think he wasn’t.
A nervous laugh spread through the audience, but everyone was actually just waiting to discover the final result of this adventurous travel to the Middle East.
The final clash was directly with Mr. Leveque, who in the first days appeared to be open and collaborative, but in the end started to show clear signs of impatience and fear.
In the end, Mr. Leveque decided to confess and fully disclose what was going on: he personally got involved in a bad history of horse betting and blackmail. This was one of his achilles heels and a criminal named Omid Tahvili found an effective way to leverage it. Omid met Mr Leveque when he was in a bad day when he nearly lost a fortune.
Tahvili proposed to help him to reduce by half his gambling debt, asking in return for a particularly favorable treatment for a bunch of companies he was affiliated with. Those companies and this will not surprise anyone, were not flourishing, and actually, we at the Hippalus would have never started doing business with them if it wouldn't be for this sad history.
No surprise then if those bad customers, in the end, weren' t able to honour their debts. Since they were all substantially run by Omid himself, and this was testimoniated by our commercials lamenting a difficulty to find useful contact persons to relate with, as soon as this crimal started losing his grip on the shadow market, the business started to ruin, and we started to see our cash flows dramatically drop. Not to be mentioned, Leveque got fired, but you already know this from company news, and Mr. Sahid took his place.
The silence was not absolute; people within the room were surprised by the amazing story and how incredibly close to the truth the data had led them.
That said, even if for the company it was a hard blow, we are already recovering in that area and in the overall figures. That's why we can now look at this story as a relevant case history. Particularly, this was remembered as a great example of the power of data mining models to discover the story data that can tell. Let me, therefore, call two of the main characters of this story to join me here: Andy and his valorous assistant! We already knew about the great skills of Andy. No one would have bet on such great work from a newbie working on the topic: congratulations!
Yeah, Mr. Sheen is talking to you, and you definitely deserve his congratulations. You made a good job moving from the first uncertain steps in R to conducting a full data mining project.
You now have a powerful tool in your hands, and you know how to you use it. Good look, and do not forget to come and say hello at the Hippalus sometimes!