Cost Benefit Analysis Real Life Examples - Sensitivity vs Selectivity

1,024 views Wait, is this logic right? • Apr 05, 2023

Slog Reference: The ideal amount of fraud is not zero

Description

Why credit card fraud hasn't stopped in the US? How would you react if I said that the ideal amount of fraud in a credit card or similar company should not be zero? What would you say if I can prove to you that the ideal amount of unemployment in a country is non-zero? What if I say that the ideal number of heartbreaks is non-zero? Sounds extreme and unreal, right? Well, let me prove to you all these things in this FutureIQ episode where we talk about selectivity vs sensitivity in any system. Let's understand how security and convenience are correlated to each other and how can you benefit from this knowledge.

Listen it on the podcast provider of your choice: https://tapthe.link/FutureIQRSS

Hope you enjoyed FutureIQ by Navin Kabra and Shrikant Joshi. Do hit us up on Twitter:
@ngkabra http://twitter.com/ngkabra
@shrikant https://twitter.com/shrikant

References from the video and other episodes:
ROC Curve: https://tapthe.link/ROCurve
Prisoner's dilemma: https://tapthe.link/Prisoner'sDilemma

How to detect fraud using Benford's law: https://youtu.be/JHA2QpJKGN4
Understand Goodhart's law: https://youtu.be/OU5W-b_e8a4

Chapters:
00:00 Ideal way to prevent frauds
01:08 Sensitivity vs selectivity
03:19 Credit card frauds
06:00 Security vs convenience
07:35 Real life applications
09:36 Thinking like a VC
10:12 The balance

#futureIQ #costbenefit

Related Slog Matches

The ideal amount of fraud is not zero

Fuzzy Text

85.29

Transcript

- so in an Ideal World how much crime should be there what do you think zero Ideal World zero no crime at all please let me rephrase the question okay suppose I gave you a method a hundred percent guaranteed method of preventing all credit card fraud right do you think a credit card company would implement it I have a guaranteed method of course they had implement it yes the method is don't allow any credit card transactions of course they wouldn't implement it yeah you see the product I wanted to make right is that it is very easy to prevent fraud by preventing transactions completely okay I mean it's like you if you don't want to give birth to kids

0m 52s I think you you find it funny but this is a serious issue right because in reality the reason why fraud exists is because of this a more subtler version of this okay okay the problem is that to reduce fraud okay you end up reducing some actual transactions right in any attempt to prevent fraud yeah and that's why yeah today's episode is that the ideal amount of fraud is never zero it's not zero okay it's it's a very counterintuitive statement when you think about it but it is a very important and highly applicable concept okay in uh the parlance it is called sensitivity versus selectivity okay sensitivity as an if I have a method of preventing fraud Okay then if I if it

1m 52s is able to prevent hundred percent of fraud then I say it is a hundred percent sensitive method it can detect every possible possible fraud but the other problem is that it should detect only fraud and it shouldn't have false positives it shouldn't flag real transactions as fraudulent right right that is called the selectivity it should select only the fraud and nothing else okay and almost a law of nature is that you can't make selectivity and sensitivity both one right you can't make both hundred percent right uh in a different episode we discuss something called The ROC curve okay okay where for anything that you are doing in life I mean for fraud detection but even for simple things like using machine

2m 43s learning to detect if you have cancer uh this concept the the problem is that selectivity and sensitivity are a trade-off to increase one you have to decrease the other and when you plot that on a graph you get something called The ROC curve we'll put links to that in the video here on your screen or in the description go check it out definitely uh coming back to our episode for today right what we are going to talk about is credit card fraud let us take that as an example right if you try to reduce the fraud to zero that means you are trying to increase the selectivity to this sorry the sensitivity to hundred percent

3m 33s correct what is going to happen is that the selectivity goes down which means that valid transactions get flagged and people get frustrated people are not able to buy a lot of online uh you know not worth my time and a lot of merchants lose money because uh they are trying to ascertain the nature of the transaction being fraudulent or not correct makes sense if I have to pass a fraudulence check every time I transact my transactions are definitely going to decrease correct uh no no in fact it is worse than that right a whole bunch of times what will happen is that you want to do the transaction it is not fraudulent but that thing just refuses to let your

4m 24s transaction go through oh that's even worse right yeah and guess what huh all companies pretty much have decided that that's too much of a price to pay right it is okay to let a few percent of fraud go through but you would rather not inconvenience customers so it makes sense kind of like they are chalking it down as a business expense key absolutely it is this is going to happen this is bound to happen we can't really Escape it so might as well put it down in our books as business expense fraud happened yes uh about you know all over the world around 10 to 20 billion dollars worth of credit card fraud happens right

5m 05s that that is fraud which is not caught by the existing fraud systems right and this is for a 250 billion dollar industry so we are talking about you know three four percent of their uh revenues are going to fraud they just treated us you know this is you just live with that but when you take a look at the larger share of the pie you realize that the amount of money that they are letting go to fraud suddenly doesn't seem like much and also it explains why people like Frank abig Nelly can exist yes of course everywhere in life this is there this trade-off that uh I mean think of it as trying to prevent fraud is trying to increase

5m 53s security anytime you try to increase security you are going to reduce convenience and everyone in the world every business has to take this trade off do you increase your security or do you focus on convenience of customers right okay uh so uh this is why uh for example Indian government RBI has decided that they want to increase security of credit cards okay so because of that we have all kinds of otps and something and no recurring transactions and all over India people are complaining because now they cannot sign up for Apple TV is monthly subscriptions Netflix [Music] it is now working but initial initially there was a problem setting up a subscription for Netflix also but and YouTube much worse right uh companies

6m 47s software companies are not able to play their bills for cloud computing and they have to have somebody sitting there on the first of every month plane and if that person forgets or you know that person was sick that day the entire company is going to go down yeah so um but in return RBI has increased the security for you know like million people who don't use cloud computing like me yeah so that trade-off yeah is always there yeah so uh it it goes uh very well to the explanation of selectivity sensitivity and security that you just gave but my question to you here is how does it apply to me in real life is there is there

7m 39s an application for this to me personally so this basic concept that increase in security trades off with increase in convenience can be generalized even further saying that increase in trust has to be traded off against being corned by somebody right if you remember we have an episode on trust you trust the other person and in many cases that works out well and both of you uh do well because you are trust trust corporate cooperator and do well yeah but once in a while you will get uh conned by the other person they will take advantage of you and you will lose out so that trade-off is there you have to ask yourself which one do you want and you have to

8m 32s know that there is a trade-off and that being all the way here is not the right answer right the ideal amount of fraud is not zero the ideal amount of trust is not 100 yeah and you should definitely check that episode out because in that episode uh we uh kind of work the prisoner's dilemma which the concept of a relationship between two people and uh the the Crux of it in this particular context that we are speaking about fraud being non-zero is that you must get your heart broken you will rather get your heart broken um and yeah the ideal amount of fraud is non-zero exactly the ideal amount of heartbreak is also not zero right as a you

9m 20s are not going to find the person of your dreams unless you are willing to have a heartbroken eight times in a row Agony Uncle Naveen is available for consultation please send us reach send us comments this is called thinking like a venture capitalist right dating like a VC so in life is not zero right if you are not failing that means you are not trying if you don't try new things every once in now if you don't fill that means that you're not doing anything right so it's like the credit card company which doesn't allow any transactions yeah tweet to us comment send us an email whichever way you can contact us your questions will be

10m 06s answered in very interesting ways right so um the ideal amount or the ideal quantity of Rod is non-zero so is there like a specific number uh that that people can people should look out for that this is the threshold Beyond this if fraud starts happening you you start to worry you start to panic so no because like I said it's a trade-off and this is every individual business uh takes its own decision about where on the trade-off they want to be okay in fact there are some numbers which have I mean they're not hard and fast rules right they're just click rules of thumb that somebody came up with and many others agree so for example the ideal amount of

10m 55s unemployment in a country is not zero now this I would have immediately raised my hackers and said how can you say that but now having heard all of the explanation that Naveen has given yes yeah I agree there's a trade off again right if everybody is employed your country is going to have inflation and there are going to be problems so this is uh you know last hundred years of 70 years of Economics there are a number of economists and governments who believe that you know two three percent yeah employment in the country is desirable uh so that two three percent actually it used to be four percent uh is a number that somebody came up with and a lot of

11m 39s countries agree so no I think there is actually a mathematical basis for that number and there is a Planet Money episode on this NPR has a podcast called Planet Money and there is an episode on uh this uh or there is a flickonomics episode on this I'll link that in the comments in the show notes of this particular episode right yeah so uh every business will have its own individual level of uh non-zero fraud and you should try and figure out which that is and what that number is everybody doesn't know I was hoping that Naveen would give a number because that would also give us a number for so the way to think about it is that you

12m 14s have to try to estimate that by setting the uh level here huh how much of a good thing are you giving up huh to get the security right right you go all the way here and you're giving up too much of the good thing you go all the way here and you're getting too much of the bad thing so when the two numbers become roughly equal that's where you want to be and that is a number that I agree every individual every business will have to choose for themselves and that clearly also means that you don't trust everyone every time or you don't not trust everyone absolutely yeah this is where the prisoners dilemma thing uh is something

13m 02s that you might want to check out right so um huh I think I have the perfect quote to end this episode on and it comes from a brilliant author uh many many centuries ago called William Shakespeare it is better to have loved and lost than to have never loved at all the ideal amount of heartbreak is not zero Naveen future IQ thank you thank you