• +31 64 056 8403
  • shane.pillay@shanepillay.com

Smart Analyzer

What if you could predict stuff? Around 2014 I developed, along with Freek Meerhof, a tool called Mobile and Housekeeping that generated predictions based on historical data. This was our initiative.

Click a brilliant colleague for Freek Meerhof's Linkedin profile.

That involved into the Smart Analyzer. The premise? Well, what if a customer starts typing in a question, and as he types each letter, our tool, based on some smart analysis, tries to predict the correct answer.

I explain our algorith with a simple example and table.

I run a business called "Mobile and Housekeeping". I offer a phone service with mobile phones and, being incredibly horizontally integrated, I also offer housekeeping.

Starting with the table below (actually you can start with a table with zero elements), I receive a complaint from a customer saying "My phone bill is too high".

How do I know if this complaint has to do with Mobile, Housekeeping, Laundry or Invoices?

I do the map by analysing each piece of text in the sentence. Eventually, I would analyse all combinations of texts, including spaces. But for this example, I analyse whole words.

Text Mobile Housekeeping Laundry Invoice Total
bill 100 200 50 80 430
phone 70 20 40 30 160

Here is the sentence "My phone bill is too high".

For each word, I calculate it's weighting per column

For Mobile: my=0, phone=70/160, bill=100/430, is=0, too=0, high=0. Final value=0.67

For Housekeeping: my=0, phone=200/400, bill=20/430, too=0, high=0. Final value=0.54

And so on and so forth.

In this case it is possible called about Mobile. This will be suggested to the customer care agent. The agent can correct the answer, if necessary, and this increments the relevants columsn by 1 - a learning process.

How powerful is this? Not sure, but we built it in a day!

I will add on some sample scripts soon.