{
  "video_id": "wIGOnM6Cf_E",
  "channel_slug": "statquest",
  "channel_handle": "statquest",
  "title": "Human Stories in AI: Abbas Merchant@Matics Analytics",
  "duration_seconds": 3289.0,
  "url": "https://www.youtube.com/watch?v=wIGOnM6Cf_E",
  "upload_date": "",
  "transcript": "[Music]\nhello I'm Josh starmer and welcome to\nhuman stories in AI brought to you by\nStack quest in this series we'll hear\nabout the career journeys of passionate\nAI experts from their humble beginnings\nto conquered challenges we'll be\ninspired by the real world experiences\nof professionals thriving in the ever\nevolving AI\nlandscape today we have special guest\nobis Merchant founder and CEO of madx\nanalytics which uses a combination of AI\nand analytics to transform Enterprise\ndata into intelligent actions so without\nfurther Ado AIS can you tell us about\nyour journey to where you are right now\nas the founder and CEO of madx analytics\nhow did this all start so uh starting\nfrom I was in my 11th grade where it all\nstarted so everything was good life was\ngood uh and suddenly from nowhere a\nthought random thought came to my mind\nthat after doing all of this hard work\nlike schooling exams boards college\ndegrees phds of the world eventually I\nwill be joining my family business so as\nI mentioned\nuh I come from a business background\nactually so that thought came to my mind\nand I was wondering what to do with it\nlike if it is right or wrong or first ly\nit was a very dumb thought like dropping\nout of the school and it was something\nlike right now if I think it's the worst\nthing that someone can think of like\ndropping out suddenly from school and\njoining the business but looking back I\nfeel that everything happens for a\nreason so yeah I propose this great\nmulti-million dollar idea to my family\nand they straight rejected\nlike they also come from a business\nfamily but they know the importance of\neducation and you need to be qualified a\nlittle bit qualified enough to handle\nthose business right so to clar hold\nhold on to\nclarify um you're in high school yes and\nyou're like oh I've got a business\nidea I want to drop out of high school\nto do this business and you tell your\nparents yes what do you guys think of my\nplan and they say that's the worst plan\nwe've ever\nheard yeah okay okay just just to make\nsure we're all on the same page yeah so\nit was not a business idea it was\njoining my family business we are into\nretail and distribution of electronic\nconsumer goods so joining that only\nthere was not such any multi-million\ndollar idea that I have the idea was to\ndrop out okay so the idea you knew you\nwere going to be working for your the\nfamily business yes and and you're s\nyou're in high school and you're like do\nI really need to do all this when I know\nthat my destiny is to be working in the\nfamily business no matter what no matter\nwhat I study no matter how what grades I\nget I'm going to have this job and\nthat's and I just can't fight it so I\nmight as well just start now and that so\nthat's what was going on yeah like I\nthought like all these years I should\nspend in real world doing some business\ninstead of bookish knowledge that's\nright yeah that's I don't think that was\nright at that point because yeah when\nyou think back uh it all makes sense but\nyou need like education is the basic\nneed like the fundamental of a human\nbeing like what we do in our life how we\ntake decisions how we take actions it\nall comes from that like what what is\nyour mindset where do you come from okay\nso yeah and just moving moving ahead so\nthey just straightly rejected this idea\nand asked me to continue what studies\nand I was uh a little bit stubborn at\nthat point of time like I was like no uh\nthis was my mindset like it's clear I\nwant to do this I'm not wasting my time\nmore on studying and I I listen to No\nOne to be honest and just uh dropped out\nokay yeah so after few months yeah it's\nuh like going back again it it's like a\nvery bad decision but but it worked out\nI I don't know by the grace of God or\nsomething uh so after 3 to four months\nof convincing my parents they finally\nallowed me that uh okay now you are\ndoing nothing and they Tred to convince\nme very much like my family uh like you\nshould study you should get that basic\nknowledge business is year only it's\ngoing nowhere right but you should get\nthat fundamental but I heard of no one\nand uh convince them somehow that please\nallow me to join and fast forward Josh\nthree\nyears I was in the business three years\nyeah yeah uh Working Day and Night the\nrevenues were good the profits and I was\nlearning a lot of new things but there\nwas a sense of stability like I was not\ngrowing I want to grow but it was due to\nI don't know lack of qualifications or\nknowledge it was a stable like\neverything was already set up I was\nmanaging it and also I was uh using some\nof somewhat my brain to uh increase and\nsome ideas and all of that so I was not\ngrowing enough and at that point of time\nI am talking about back in\n2016 e-commerce was booming a lot of\ncourse e-commerce was quite before that\nbut in in India it was booming a lot\nAmazon flip cart\nand uh and we are into Electronics\nretail distributor consumer goods\nbusiness and I don't know how will we\ncompete I saw that it's a major\ncompetition that we are going to face\nand I am helpless but I can't do\nanything so I started sitting with\npeople business people uh from different\nIndustries what are they doing what are\ntheir thoughts but somewhere I sense\nthat feeling of missing education I\ndon't know why that I'm either I'm not\ngrasping things or I'm not able to\nunderstand or I afraid of the\ncompetition I don't know but there was\nsome a little piece\nmissing after thinking a lot or\noverthinking or anxiety or whatever you\ncan say after 6 months of doing all this\nthinking uh I decided to switch gas and\nfirst complete my schooling so the base\nI I thought like I need to make the\nfundamental the base clear then only a\nbuilding is going to be built otherwise\nyeah without base there is no building\nokay so I switch gas uh and take took\nthat decision to pursue my at least\ncomplete my schooling again yeah and I\nproposed that idea and everyone was\nhappy B but finally we were saying\nbefore three years now you are now you\nare understanding but yeah I I was bit I\ndon't know so uh yeah when I I remember\nit was around 4 months left in final\nboard exams that was that's what they\ncalled your 12th board exam after that\nafter that we go to colleges uh\nengineering or anything so uh there was\njust four months left and I have no idea\nhow in the world I will be completing\nafter this great mind shift after three\nand a half years of doing business not\ntouching the academic books how in the\nworld I'm going to do this but uh yeah\nwith family support with friends support\nand everyone came together and I like\nsupported me so I was studying for 13 14\nhours a day continuously for four months\nand fortunately I cleared the exams with\nsomewhat good grades yeah\ncongratulations that's awesome\nthank you but everyone was happy again\nthat uh please feel free to stop me in\nbetween if you have any question no this\nyeah yeah everyone was happy again but I\nwas not because I don't know how it will\nhelp me in growing my business because\nin back of my mind it was the only thing\nthat ended up me here right I I ended up\nhere so I thought of pursuing further\nlike I need to get higher college degree\na sort of educ education but this time\nit's a critical uh like it's a crucial\ndecision for me because I don't want to\nwaste any further single day so I\nstarted exploring what all business\ncourses are available and what all\nthings I can do all the bbas and mbas in\nthe world I I explored and found that\nwhatever I have learned in real world\nlike three and a half years of doing\nbusiness it was nothing different or I\nsense that it will not help me in what I\nwant to achieve in uh standing in front\nof this High competition like high\ne-commerce thing yeah so uh again uh I\nasked uh one more question to\nmyself that that changed the paths like\nthat's why we are sitting here I guess\nso what I asked was I spent last 4\nmonths studying day and night 13 14 15\nhours a day doing this completing this\nschooling now how in the world these\nseven subjects or out of the seven\nsubjects one subject will be helping me\npursuing my pathway\nforward so I expl I started exploring so\none of my favorite subjects you can say\nwas or is my favorite subject is stats\nso it's all summing it summing it up all\nagain I'm I'm fond of stats too yeah\nright so uh the question the question\nthat game is how in the world is stats\nbeing useful how can I use stats in this\nreal\nworld uh I started exploring again and\none thing that caught stats being used\nin multiple Industries but one thing\nthat caught my attention was tech\nindustry it because uh the tech\ne-commerce was the only reason I ended\nup in this situation like I am in this\nstage right now so why not why not\nunderstand Tech why not understand e-com\nso through stats so I pursu uh computer\nscience as my career I decided to pursue\ncomputer science as my career and uh one\nyear passed by I was in uh academics\ncollege degree there was just one\nsubject in semester 2 which taught me\nthe school level stats that I already\nknow and somewhat of probability\ndistribution some advanced level stuff\nbut no one answered\nhow is stats going to be useful in real\nworld how we are going to use this\nknowledge that we are getting that I\nhave got that everyone got in schooling\nin colleges how no one was ready to\nanswer that I I don't I didn't get that\nanswer to be honest I started exploring\nagain Googling right now I'm with a\ndifferent mindset when I'm researching\nbecause I have shifted myself from\nbusiness from learning to now computer\nscience so different Minds I started\nGoogling YouTube and at that point of\ntime uh you came into my\nlife I'm not saying this because we are\npodcasting and we are sitting but it's\nit's a true it's true story\nfact so I came across your\nvideos and how stats is being used the\nPractical\nimplementation AI ml data science and I\nwas like bam like\nsorry for stealing\nyour love I was like maam i w I wanted\nto do this this is the thing I wanted to\ndo uh like yes so I started learning ml\nSTS maths gradient descent algorithms\ndata science Python and whatever you can\nthink of you won't believe in at the end\nof third semester which is six month of\nme learning this technology I have an\noffer letter in the hand with an AI ml\nR D intern\nposition awesome yeah yeah that's\nawesome yeah that's a very I don't know\nemotional or situation for me that I was\nnothing and suddenly in one and a half\nyears I went into computer science and\nwithin 6 months of studying this I have\nthis opportunity yeah now I have nothing\nin my mind just to grab it just to grab\nthat opportunity no matter what I\nconvinced my uh college that it's it's\nin different city actually so I\nconvinced my principal being H ODS\ncollege that this is an opportunity that\nno one even gets after completing the\ncollege like people are looking for jobs\nright and I'm getting it so I am very\ngrateful and blessed enough and I wanted\nto use this opportunity\nso somehow I convince them\nalso yeah it sounds like a a consistent\ntheme in your life when it comes to\neducation yeah right right and I'm also\nI also mentioned this because today also\nI see people dropping out like with\nwhatever excuses like Bill Gates was a\nDropout Mark Zuckerberg was a Dropout\neducation system doesn't teach\nentrepreneurship education systems does\nnot have that curriculum no matter what\nexcuses or what reasons you bring in at\nthe end of the day it is\nimportant basic so in India there is a\nterm like uh three things you need in\nlife Roy Kaa makan which means uh food\nclothes\nshelter so I have added that like fourth\nthing you need is\neducation interesting basic education\nthat's all I don't know what excuse you\nyou bring in so ah coming back so uh to\nour story yeah sorry if you have\nanything to add oh no no no no this is\ngreat so coming back to the story uh I I\nshifted back to that City and with that\ninternship I also got a job there after\n6 months and there was no looking back\nlike I I right now currently I'm in this\nfield for more than five and a half\nyears working with multiple Industries\ndifferent domains and I can't start to\ntell you how much today also how much\nlearnings like I'm I'm learning uh new\nthings how much new learnings I have how\nmuch new things I'm getting to know and\nI keep on like I wish I keep on doing\nthis I am never uh what you say like you\nget frustrated by your job or something\nof course there are some points in time\nwhere I am also like that but I don't\nwant to stop doing it I I'm I want to\nkeep learning I'm enjoying it and this\nwas all happening five years and one\nfine day one more question came to my\nbeautiful\nmind sounds like time for change time\nfor a change time for a change yeah\nthat's where the entrepren path come\ninto picture but let me let me elaborate\nmore so one fine\nday so what uh the question was simple\nthat I was everything was was going good\nbut it was never the goal of climbing\nthe corporate lead I will tell you a\nfact like until unless we know as a\nfamily like uh over from five\ngenerations never ever someone has done\na job I was the one who in the SP\ngeneration doing a job but it there was\nno problem with anyone I was like it was\nnot a goal to climb the corporate lead\nbut I am enjoying it and I want to keep\ndoing it it's 5 years it's 50 years the\nquestion let let me come to the point\nthe question that came to my mind was in\nthis five years I have worked with\nFortune companies Fortune clients\nFortune 100 Fortune 500 anyone and\nsuddenly from nowhere uh the question\nwas where are Fortune\n5,000 where are Fortune 10,000 that\ndoesn't even exist where are smmes small\nand medium\nEnterprises and that's where already\nstarted I started exploring\nagain where are all this what are all\nthese companies doing whatever knowledge\nand skill set I have how can I utilize\nit to make a larger impact no matter I\nam I'm making impact to that particular\nsegment so along with my job I started\ndoing market research validation now\nsitting with a different set of business\npeople business owners uh industry\nnetworks all experienced people out\nthere and I found a gap or I would I\nwould like to say as an\nopportunity so what it was Fortune 500\nFortune 100 companies are all data aware\nAI aware they have all the\ninfrastructures they have all the budget\nof the world there when it comes to the\ncompanies who are not Fortune 500\nfirstly they are not aware so I did this\nresearch for one and a half years with\nmy job so I'm bit slow uh as you can and\nsee see like three and a half years one\nand a half years of\nresearch yeah so uh yeah so three things\nI found out doing my this one and a half\nyears research is firstly they are not\naware how the data that is sitting\nideally in the systems can bring value\nfor their business can grow their\nbusiness they don't have any\nidea second was affordability that we\ncan understand third was all thanks to\nat GPT open AI because of that people\nare started to getting to know that AI\nis something before that AI was like\nrobotics\nTerminators and all yeah yeah yeah now\npeople have that understanding that AI\nis something that we are using in our\nday-to-day life so third thing was how\nthey can utilize this technology in\ntheir business how in the world they\ndon't know so I grab that this this is a\ngap this is an opport opportunity and I\nwant to do this yeah so I took this step\nso before 6 month I Incorporated and\nfound and Incorporated matx analytics\nthat's what our company is called and\nfrom this year like from uh this new\nyear I qu quit my job and Go full in uh\nlike went full in you can say wow so\nyou've started your own business yes\nstarted because I found in this one and\na half years that it it will not be\nsomething that I can do as a side\nit's a problem that is going need to be\nsolved and it will take some amount of\ntime energy and\neverything and I it's just keeping in\nmind one thing that I want to utilize\nwhatever skills and knowledge I have to\nmake an impact to a larger audience yeah\nthat was the mindset yes was that a\nscary step was that\nlike you know when you leave that job\nand all of a sudden you're kind of\nfloating right when you have a job with\na company that's doing well there is a\nsense of stability there's a sense of\njust knowing exactly what's going to\nhappen today but also six months from\nnow you'll you'll be kind of doing the\nsame thing making the same amount of\nmoney and then you said I'm gonna I'm\ngonna I'm gonna jump off a cliff I'm\ngonna say goodbye to all that stability\nand all that knowing of what to expect\nand I'm I'm just going to jump off and\nI'm going to see what happens was that\nscary very scary to be honest yeah\nbecause you right now you know how I\nhave reached till here how I have\nreached it I was at senior lead data\nscientist position at Moody analytics is\na listed company in us so uh I was like\nokay now like a family also like\neveryone that I like in one and a half\nyears I have done this only like\nresearch and how will be how will I'm\ngoing to do this so I tried for six\nmonth like in my one and a half year\nresearch for six month I tried so I got\none client also but okay uh I got to\nunderstand very early that it will not\nwork out because I was the one working\nfor it for that client yeah how will I B\nbuild a business I was doing freelancing\nthat's what fre that's called\nfreelancing that's not called\nestablished corporate company and this\nis not how we build it I was the one\nworking in for that client now how will\nI going to build a business how will I\ngoing to build a team\nand five different departments you know\nsales marketing and all of that because\nI come from that I I I I got to know in\nthis six month that it will not work out\nand also I have a client that also gave\nme some sort of confidence that yeah\nokay and due to covid so I I'm just\nsaying it was scary but due to all these\nReasons I'm able to make that step first\nis I got a client thumbs up that this\nthing is working out real world\nsecond that I I realized that I'm alone\nor if I will be the work one working for\nthe projects or clients it's freelancing\nit's not\nbusiness and I'm just trading time\nwhatever I'm doing in my job I'm doing\nhere and yeah third was like how in the\ncovid yes sorry uh covid so because of\ncovid I was working from home so I have\nsome sort of savings from working so I\ncan take that risk and also I'm\nprivileged enough to come from that\nfamily or background that I like they\nare not dependent only on one income\nstream so and I also have a Runway of 2\nyears I would say right now it's just 4\nmonths but yes so all this now I am very\nlike doing calculated risk before you\nsay as you said jumping off a cliff\nleing education dropping out uh taking\ncomputer science I I have no idea what\nhow it is going to help me in my\nwhatever I'm planning to do and taking\nthis internship dropping out of college\nnot dropping out but partially dropping\nout you can say I was I was not\nattending I was just going for college\nto give exams to be honest yeah because\nthere was no such subject in my\ncurriculum which I wanted to learn then\nthat's no point so all these are jumping\noff a cliff as you rightly mentioned but\nthis one I was like now I'm experienced\nbro now I I'll be taking calculated\nrisks yeah okay so after small Cliffs\nsmall Cliff yeah small Cliffs not huge\nCliffs correct and again I proposed this\nidea and they were like now you are\ntalking my family were wow okay\ncongratulations\ncongratulations okay let's do it like\nthey were like always supportive always\nlike uh I'm uh grateful for that that uh\nif you think it's something you believe\nin if you think you are going to enjoy\nit uh like every day you need to wake up\ndoing it then go ahead please yes\nbecause it's like they are also like\nthat mindset that it's better to fail\nthan regret mhm so I jumped in all in uh\nfrom this 3 to 4 months and you won't\nbelieve whatever I have not learned in\nthis no no it's not a surprise but\nthat's\nit from your expression I guess you are\nexpecting but you won't believe from\nthis 3 to four months I have learned so\nmany things that I have not learned in\nmy 9 years of Journey like 3.5 years in\nbusiness 5.5 years in the corporate\nworld and it's been amazing until now so\nfar like how to build a team how to\nbuild a business uh marketing sales\noperations customer success and you\ndon't ask like I'm really enjoying it\nand I want to keep doing it just to end\njust to end my like this is this that's\nall like like there's no more Clips or\nsurprises for you yeah I love it I love\nit when uh so this also came from Once\nsomeone a reporter asked to Bill Gates\nthat I don't know if you are aware but I\nsaw that reel or something that what in\nthis world you are afraid of losing it's\nBill Gates afraid of nothing right and\nhis answer was I don't want my brain to\nstop learning yeah and I was like yeah\nthis is it I want to do the same I don't\nwant to stop learning until unless I'm\nlearning I want to be in this\nforever yeah I feel like that's what it\nmeans to be living is to be learning yes\num yeah um that's an incredible story\ncan you tell us about what your company\nis doing I it sounds like you're you're\nin the AI field but can you give us some\nmore details about what you're doing\nsure I would love to Josh so right now\nwe are working with couple of clients\nit's very early stage and with a small\nteam so we are working with different\nIndustries but uh mostly what excites me\nalso is ml powered marketing uh that's\nwhat we are currently working in and\nalso uh there is customer retention use\ncases so in real world how you prevent\ncustomer for from being churn before the\nchurn right yeah so that's one and then\nthere is some fraud detection in\nfinancial domain we are also there are a\nlot of AI and ml use cases but uh I can\ngo deep on what we are working whatever\nfield you are interested like ml power\nmarketing or yeah let's let's talk about\nml powered marketing uh can you tell us\na little bit a bit about that and sort\nof what you're doing and maybe how\nyou're doing it or can you give us those\ndetails yeah yeah sure I would love to\nso uh basically it's targeted marketing\non based on ML so it's a financial\ncompany based out of us that we are\ncurrently working with so the use case I\nwill be brief overview of the use case\nwas how to select right target audience\nfor our marketing campaigns this okay is\nthe question that the company was facing\nlike so marketing is cost to\ncompany that's\nit in simple words either you are\nputting out an offer 20% off 30% off 50%\noff either you are reaching out to\nanyone for opening a new credit card\nit's a financial company so if if if a\nperson is not likely to interested in\nyour credit card offer they will likely\nopt out right and when you need that\nperson when you need that person in your\nmarketing campaign he will not be there\nbecause he has already opt out due to\nthat irrelevant offers so that's again a\ncost to company which is not in terms of\nmoney but in terms of lost opportunity\nokay yeah first is the offer which is\nlike you can't\nrandomly give offer to all 50% off right\nthat's a cost to company you want to\nsend offers to relevant people so let's\nsay if Jos is buying from Nike Nike is a\ncompany and you are a regular customer\nyou already buying so there will be a\ndifferent offer for you I am a customer\nwho have bought shoes from Nike before 2\nyears and now they want to retain me so\nthere will be a different offer for me\nwe both are not same here yeah and third\none is customer acquisition if you want\nto opening a card or irrelevant\nirrelevant office that's also a cost to\ncompany so we are building machine\nlearning models a multiple ml models to\nresolve this multiple issues so it's not\njust one because marketing campaigns as\nI mentioned are for different segments\ndifferent\nscenarios so first is uh propensity\nmodeling that's what we are building the\nmodel to be in simple terms it's uh\npredicting the probability of a customer\nopening a credit card with this\nfinancial plan looking at their past\nhistorical campaign and transactions and\nall of the data second was expected\nspend like what this customers are\nlikely to spend in next 12 months so\nbased on that uh they will be uh giving\nthe\noffers propensity only and third one was\nChannel preference uh like what channel\nthey prefer now uh to elaborating more\non it before I dive deep do you have any\nquestions just just to make sure I'm on\nthe same page um so one is identifying\ncustomers two sort of categorizing them\nmaybe since this is Maybe maybe if we're\ntalking about credit cards you want to\nknow who's going to be the Big Spenders\nuh and who's going to you know who who\nwho might get the card but not use it\nand and that might create an expense for\nthe company as well and then the third\nthing that you're interested in is\numh what was the third thing uh the\nchannel uh what channel oh what's the\nbest way to reach that person if if\nyou're going to because because you you\nknow who they are and you know what\nthey're going to spend but how can you\ncontact do you call them on the phone do\nyou put an ad on their email or how do\nyou reach those people yeah that's the\nthird thing correct okay I'm ready yeah\nsure so uh after understanding the\nproblem statement of the client and so\nthe major overview headline uh problem\nstatement was how to select the right\ntarget audience for our marketing\ncampaign you want to reach to the Right\naudience right yeah from that diving\ninto de this three scenarios uh came in\nthis three problem statements came into\npicture so we started looking into the\ndata that's the first step like like\nafter we understand the business problem\nand all so they have customer\ntransactions data it's a financial firm\nof course it it's required right then\nproducts data items data and all of that\ndata that we actually need for building\nthat Predictive Analytics predictive\nmodeling okay check mark That's it uh we\nhave that data now second thing it\nsecond thing is how do we utilize so\nthey have around four years of data with\nthem like that's what we need past\nhistorical data to predict the future so\nwe come up with a performance window and\nan observation window so for to\nelaborating more on that performance\nwindow is something for example there is\na one year window that we will take that\nis a performance window in which we are\na customer of of that brand in that\nwindow and what is our behavior in the\nobservation window sorry observation\nwindow is the current 12 month window\nwhich we'll be using for model training\nand performance window will be the next\n12 Monon window so 1 January 2022 till\n31st December 2022 one year window will\nbe the observation window so training\ndata what is the uh user Behavior\ncustomer Behavior okay and performance\nwindow will be the next 12 months 1\nJanuary 2023 till 31 DEC what is this\ncustomer segment doing in next 12 months\nyeah okay now coming with the first\nproblem statement so we started with all\nthe data pre-processing getting all the\ndata data\nwarehouses and all then\npre-processing feature engineering\nmodeling all the step St first problem\nstatement we for propensity modeling\nbecause why I'm saying propensity is\nit's different I I hope you might but\nfor the audience it it's different\nsomewhat different from probability\nmodeling what we do eventually it's it\ncomes with some sort of prising like why\nwhy you are taking this decision so\nright now we also need explainability so\nwe also need to take care of that so\nthis all come after our multiple\ndiscussions with the company\nso what we identify was the data was\ngood like after doing all the feuture\nranging we have all those features we\nalso used after experimenting with of\ncourse when it comes to tabular data I'm\na big fan of machine learning models\nalgorithms boosting bagging models and I\nI like would not like to touch deep\nlearning when it comes to uh tabular\ndata to be honest like it's I don't know\nI'm biased or not but yeah well it's I I\nfeel like the the the the methods you\nwere just talking about are are lend\nthemselves to easy\nexplanations and and the the Deep\nlearning and the neural network stuff is\nit's it's somewhat more opaque about how\nthe decisions are being made and that\nmakes it more challenging to explain at\nleast to your client what's going on\nright yeah and yeah here's reasoning is\nalso into picture so what we did was uh\nwe choose like after experiment with\nmultiple models and multiple feature\nengineering iterations we chose\nR boosting algorithm so final model was\nI think one one was exib and two were\nlight GBM that where that that worked\nquite well now here so again making it\nmore clear it's a prediction problem so\nbinary classification problem what is\nthe probability of Jos opening a credit\ncard within the next 12 months and what\nis it if it is not so if it is towards\none then you are more likely to open\ntowards zero it's not likely to simple\nbut because that's a challenge of Reason\nreasoning and all so what we did was uh\nI don't know if you are aware of this\nbut we used first of all of course you\nare aware Au R Au to train our model the\nevaluation side because it matters a lot\nuh like the main part from my was Data\npre-processing feature engineering data\npart modeling was just 10% and then\nevaluation how we are going to evaluate\nhow it will work out uh so isn't that\nfunny the way it is it's like most of\nthe work is with data yeah creating the\nthe machine learning model is a small\npart and then you evaluate the model to\nsee how well it's performed and that\nstuff is almost like it's not an\nafterthought it's it's the you know\nsomething you really want to do but it's\nI found personally as well it's all\nabout the data yep I I totally agree\nlike it's all about the data we have\nspent I would say 70% of the time doing\nfeature engineering data pre-processing\nand all how do we create more and more\nrelevant features that for example uh\ncustomers spend in last 12 months\ncustomer spend in last 6 month customer\nspend in last three months what are the\namount of uh where it where he is he or\nshe is spending all those 11 features we\ncome up with and finalize on those\nfeatures and doing after the doing the\nmodeling the valuation part that's what\nI that's what I was talking about so\ntraining matrices R binary\nclassification works well so I'm talking\nGeneral like whatever three models we\nhave built the purpose was different but\nthe methodology of course features were\nalso different but I will not get into\nit because it's a lot uh but the purpose\nwas different like it was at the end of\nthe day classifying the likelihood of a\ncustomer opening a card spending so it's\na regression problem\nhow much reward is spending and again\nthe binary classification Channel\npreference what channel prefer but\nbecause there are three different use\ncases I will not go in each and everyone\notherwise it will be a long 3 hours\nalready have taken so much of time in my\njourney but yeah yeah\nyeah so uh H after evaluating the model\nand R and all we different we come up\nwith this different I don't know if you\naware but lift and gain charts that's\nwhat I have learned from my previous\nindustry experience where I also handson\nworking with marketing analytics with\ncan you tell us about these yeah sure\nyeah so lift and gain charts it's mostly\nuse in propensity modeling to identify\nthe output of your model let's say 90%\nau\nbut we we want to select right target\naudience bro uh how we'll do with what\nis 90% a how you select the right Target\non so that's where it come into picture\nwhat it what it did was uh this is a\nmetrix of course uh you can like the\naudience can also learn about it\nGoogling it but uh the major idea was\nsplitting or segmenting this customer\nbase into desiles and say Diles in\nbuckets so there will be 20 Diles in\nwhich for for example we are dealing\nwith in training data 1.5 million around\n1.5 million of customers and testing\naround 700k I don't exactly remember but\nyeah so splitting this 700k or 1.5\nmillion to test the train lift and gain\nand validation test lift and gain split\nthis customers into segments 20\ndifferent desiles and check the\nprobability likelihood prediction of\nthis particular segment not at an\nindividual level so whenever I also\nsomeone says\npersonalized uh recommendations right or\nsomething like that so there are\nsegments it's not like one particular\nindividual so uh there are like uh\nsegments in which we calculate lift and\ngain now coming to the lift and gain\nwhat is it lift is basically\nuh uh\nexplaining if we are targeting that\nparticular segment\nrandomly as compared to we are targeting\nthat segment with the help of machine\nlearning prediction okay so let me more\nbe more\nclear targeting that particular segment\nusing the prediction of an ml model and\ntargeting randomly anyone how\nmuch uh lift lift is let's say we got\n2.5x lift so it's 2.5 times more chances\nof of these customers opening a card\nwith you as compared to you targeting\nrandomly yeah so that makes that makes\nsense more business sense right yeah so\nthat's lift now gain so gain is\nbasically let's say we have 20 Diles\nthat is 100% of the customers now we\ndon't want to Target 100% of them that\nit doesn't make sense to do all this if\nyou're targeting everyone gain is how\nmuch desiles for example we also call it\nranks so if we are targeting top five\nranks how much lift it is going to be\nfor us let's say 2.5x 2x 1.5x anything\nand gain is how\nmuch percentage of audience we are uh\ncapturing okay out of this 100% if you\nare targeting top five how much\npercentage of customers we are going to\ncapture in this if we are targeting this\nsegment so it's basically to decide lift\nand gain charts basically decide what\ncustomer segments to Target how much\nrank to Target okay of course it's\nmathematical formulas and yeah uh like\nlike evaluation metrix it's it's the\nsame but it's mainly used for marketing\nand also not marketing but propensity\nmodeling when when reasoning comes into\npicture it's not just a okay this\ncustomer is going to\nCH 95% now go targetting so it's it's\nnot like that it's much more after that\nwe also come with yeah sorry if you if\nyou anything no this is good okay so\nafter that we are into of course\nexplainability we have covered the\nreasoning part explainability boosting\nalgorithm we get feature importance and\nwe also turn sharly\nvalues uh for explainability so sharly I\nwas also not aware in my uh corporate uh\nwhenever I was doing the job that like\nsharply is feure important\nwe got that beautiful graph and it shows\nwhat feature is what important but while\nworking on this project I also get to\nknow from the team actually that sharply\nvalue defines what all your features\nwhich are of course important but what\nall your features are pushing your model\nto predict towards\none what all your features are pushing\nyour model to predict towards zero so\nwhat is positively impacting your model\nwhat all features and what all features\nare negatively impacting what negatively\nimpacting means what all features are\nmaking a model to predict towards zero\nlike this this segment will not open a\ncard so that's I am also I was also not\naware that I also get to know I I I'm\npretty much sure you will be aware of it\nplease yeah yeah I mean that's fine U\nbut the audience may not know and so the\nidea is when you have a large data set\nyou know measure all kinds of things you\nmight measure uh where somebody or you\nyou're not measuring that but you might\nknow where they live uh you might know\num you know their age you might or or\ngeneral age their um what kind of Opti\nthey have what kind of job they have you\nknow lots of things like that right and\nwhat you're saying is that you're using\nthese things called shapley values and\nand they can help identify of all the\nthings that you've measured what are the\nfeatures those are the things that we\nmeasured what are the what are the\nfeatures that are helping to drive\nsomeone to make a prediction that\nsomeone's going to get this credit card\nor someone who's not going to get that\ncredit card yeah and then that that\nalone helps you maybe like well maybe we\ncan streamline our data collection\nprocess just to the these important\nfeatures yes that's that's totally\ncorrect like 100% you got it awesome\nhoay I not be saying it you got it\nbecause yeah I know I have learned from\nI have learned from you actually no it's\nit's it's it's it's important for me to\nmake sure I I understand what's going on\nI get confused I'm easily\nconfused yeah so yeah mostly we I uh got\nto understand that it's not a modeling\nactual like it's a modeling problem but\nmostly it's evaluation part and after\ndoing the uh reasoning part we go back\nback to feature engineering let's say if\nsome features are not making sense why\nis it important it is not making sense\nlet's say Josh has purchased an item\nbefore two days it is negatively uh like\nit's saying the model that it the CH is\nnot likely to open a card uh okay uh we\nhave to go back you also know it's an\niterative process yeah and it so it\nstarts out as a data problem and then\nyou do a little bit of modeling and a\nlittle bit of evalu\nand then it becomes a data problem again\nyeah yeah exactly yeah so after doing\nall this I'm I'm coming to an end so\nafter doing all of this the the client\nand the company and business are all\nconvinced we are also first of all we\nneed to convince either it is making\nsense making an impact or it's just\nrandom ml MLA right yeah so we are\nconvinced they were convinced now how do\nyou test it in real time so it's a long\nprocess first of all you select sorry\nthe companies select the target audience\nfrom the probabilities from the\npredictions made by the ml model for\nthat ranks that we have\ntheiles from that they will Target in\nreal time so test and control AB testing\nuh they will do so that they also they\nhave a marketing G that's not we are\ndoing but we are uh like working\ncollaboratively with them that this is\nthe target audience you need to Target\nand test and control all of of that\ntechnicalities in marketing after doing\nthat we will check campaign performance\nso again there is a data so now we are\ndoing campaign performance for we we\ncall it back testing so whatever\nwhatever ml model predicted what and\nwhat what happened in real world so it's\na completely real world problem so how\nit is performing in real world so we do\nback testing how this campaign perform\nas compared to test and control and\nabest and all of that if you're not\nselecting ml based predictions and if\nyou're selecting randomly does that lift\nis making sense or not so lot of testing\niterations and all we also do back\ntesting if it is working out or not and\nwe can't just blindly go with okay just\nuse ml but it's a lot of iteration so\nnow at this point of time we are also\nincluding campaign features into our uh\nmodel so again a data problem now we\nhave a new set of data campaign\nperformance results that's what we are\nincluding in our\npipeline so yeah I think that's wow I\nhope it's clear yeah I love it so it\nsounds very complicated but but maybe\nmaybe I can summarize you you've got all\nthis data you\num and you know you've got you've you've\nyou've cleaned it up you spent a lot of\ntime making the data useful you created\na model that then gave you information\nabout what features and the data set you\nneeded to uh focus on but also maybe\nsome that were confusing and maybe you\nneeded to understand more about like\nwhat why is this impacting whether or\nnot uh someone will get a credit card so\nyou were trying to understand the data\nwhich is I think Super interesting and\nfascinating but then beyond that you're\nalso evaluating whether the approach is\nworking uh at you you you've got a\nprediction of what the the lift is you\nknow how much it's going to be based\nrelative to say just taking a random\npopulation but you're actually uh doing\nthe work to actually validate it too not\nnot only you saying well this is what we\npredicted the lift to be you're actually\ngoing out and and and validating that\nthe lift is is you know does exist and\nthat your your your methodology your\nml is\nsuccessful and more successful than if\nyou just grabbed random people and said\nhey you want to sign up for this credit\ncard um so that's fantastic this is I I\nI have learned a ton from this I I want\nto I want to thank you obas this is\nfantastic um I just want to uh hear can\nyou tell us you know you've this company\nhas been around for a a year and a half\num no it's been around six months six\njust six months okay in six months how\nmany people do you have working for you\nnow uh around 15 but but not all\nfulltime not all fulltime somewhere on\nyou got 15 15 employees so you've got\nlots of management experience that\nyou've learned in the past six months\nand uh how many clients do you have\nright now uh two and uh just two yeah\naround eight are in because you know\nsales cycle is very long awareness\nunderstanding so eight or nine I guess\nare in that phas discussion calls pH\nwell I'm going to just say after six\nmonths yeah this sounds amazing right\nyou've got 15 employees you've got two\nclients now in the like and you've got\nall these clients kind of lined up in\nthe pipeline prospects for growth and it\nsounds you've got a I mean it sounds\nlike a lot of work it's not a secret\nrecipe it's not magic uh but it just\nsounds like you've got a a a successful\nsort of like way of doing these analysis\nthese marketing analysis and that sounds\nfantastic um so before we go I was\nwondering if you had any advice for\npeople that want to follow in your\nfootsteps it it sounds like it sounds\nlike I'm going to take a guess that some\nof your advice is stay in school but\nsome of your advice sound it also might\nbe like ask yourself some very difficult\nquestions don't be afraid uh to ask\nyourself difficult questions about what\nyou want to be doing and the impact you\nwant to have on the world world around\nyou um but anyways I want to hear what\nyou have to say yeah I think you have\nsaid it but I would just add you\ncompletely correct what I was going to\nsay but adding few more things on top of\nit before that I just wanted to thank\nyou like as you mentioned like you have\nlearned a lot from me right now but I am\ncompletely clueless like how much I have\nlearned from you I can't even express in\nmy words uh to be honest very honest uh\nyeah now coming coming back to the\nadvice I just have three three things uh\nsimple things when it comes to advice\nthat uh also come after our\ndiscussion first is uh start uh like\ntake action that's what right now I feel\nno matter what you think the idea or\npursuing that career or job or education\nor business or\nanything start now taking action is the\nfirst step towards success that's what I\nhave learned all my career second is\nconsistency so if you are starting now\nthe taking the first step to your\nsuccess wherever you want to reach the\ngoal the path is consistency now if you\ntaken the first step but Second Step\nthird step fifth 8 100 you need to keep\ncontinue doing it now as I like of\ncourse from my journey like I I have so\nI have keep continue doing the business\nright but somewhere or the other you\nneed to realize that as well in what\npath of course changing paths and all\nbut with consistency right now mind I'm\ndoing business but it's it's with\nconsistency so uh like two months I was\njust sitting there was no one and I was\ncompletely ready that two years I will\nbe\nsitting it will take time I was\ncompletely ready so consistency we're\ndoing a job you are learning you are\nstudying in school yeah yeah anyway uh\nand third is everything happens for a\nreason and eventually everything will\nwork out\nthat's I don't know\nwhat I can I summarize your your your\nadvice\nyeah to get anywhere you have to take at\nleast one step right you got to start\nyou must start take the first step but\nmost Journeys if not all Journeys\nrequire more than one step you you can't\njust take that first step and expect to\nget to where you want to go you've got\nto be prepared to go the distance and\nkeep walking and keep\ngoing\nand it sounds like you're an\noptimist no I an optimist you have to be\nan optimist well I don't know you don't\nhave to be an optimist I'm an optimist\nuh and so I obviously think it's a good\nthing to do but if if if you have have\nto believe in yourself and you have to\nbelieve that ultimately you're going to\nbe\nsuccessful uh if you're going to keep\ntaking those steps you know you have to\nbelieve those things if if if if you\nthink if you think you're going to fail\nif you think the paths aren't going to\nmake any sense and it's not going to\nwork out then it's awfully difficult you\ncan take that first step but it's\nawfully difficult to keep taking steps\nyou have to believe that that all the\nhard work is going to pay off at some\npoint um and that's what keeps you going\nso on that OB I just want to thank you\nfor being part of uh of the podcast I\nlike I said I've learned a lot uh and\nit's been a uh it's been great talking\nto and hearing about the process that\nyou guys use in your company yeah Jo\nthank you very much Josh for having me\nhere and uh looking forward for more bam\nvideos from you",
  "transcript_chars": 45040,
  "ingested_at": "2026-05-15T10:53:34.334050+00:00",
  "source": "channel",
  "yt_meta": {
    "view_count": 7546,
    "like_count": 144,
    "channel_id": "UCtYLUTtgS3k1Fg4y5tAhLbw",
    "categories": [
      "Education"
    ],
    "tags": [
      "Josh Starmer",
      "StatQuest",
      "Machine Learning",
      "Statistics",
      "Data Science"
    ]
  }
}