Transform your R&D process with airSlate SignNow's customer lifecycle funnel for R&D

How to create outlook signature

we are so excited for maria to be joining us today uh so maria abrar she is a senior data scientist at trulio as we have gone through the day we've dug in deeply into customer tracking optimization of marketing tactics based off customer lifetime value even writing some code in r and now maria is going to do a presentation on data science and customer life cycles journey so if you have any questions as maria is going through please drop them in the q a i'm going to aggregate them and we'll do them at the end so maria i'm going to go ahead and stop sharing and i will hand the controls over to you thank you for joining us today thank you for having me it's i'm super excited to be here and hopefully everyone can learn something from me too today um can you see my screen yes that is perfect thank you i'll turn my camera off have fun thank you um hello everyone my name is maria thank you jenny for the introduction so i am a senior data scientist based in vancouver originally from pakistan i have been working in analytics um campaign management also a little bit of software development for the past 12 years today in this session i'm going to cover how data science can help in the customer life cycle journey i'm also going to make sure that i'm not getting too much into the technical details like the models um what are you know the best practices in in python or are i am going to focus majorly on what are the best approaches where we can the best steps we can take to make sure that the data science approach is in the best practice and also how can we achieve that in not only in a b2c market but also in the b2b market as well so even if you're coming from a different background like retail telecommunications or software as a service company um you will for sure get something out of this okay so introducing um we can basically split up a complete customer life cycle journey into three major steps the first is attract then engage and then reading basically so what happens is that attraction is where you are identifying who your next customer is going to be and that person needs to be engaged or attracted in the right way so what needs to be done here is you need to make sure that the customer needs to know exactly what the product can give how it can add value to the customer for example if it's a software as a service then how can that service help the customer in his business in his daily day-to-day activities how do we actually reach the customer and then make sure that the customer is required and the customer from the lead to opportunity to a new account cycle that successfully gets done so there is a hundred percent conversion rate if it's a normal customer for example from a telecommunication or you're coming from a retail background so how do we make sure that someone who has absolutely not spend any um anything on your product or on any of your clothing in a retail or a new shoe how do you make sure that the customer actually becomes more um you know frequent in your store or your website so it's so this step is important for all backgrounds for all domains next is the engagement part engagement is now that the customer is successfully part of your base the customer is now um aware of your product the customer has actually used your product for some quite some time so now this is the tricky part where you have to make sure the customer gets engaged and make sure that the customer keeps on coming back keeps on spending more and adds more value so this is where you can already you have a lot of information about the customer the spends the behavior patterns um likes dislikes based on that you can identify the next step make sure that the customer moves from a lower value segment to a higher value segment and actually grows instead of going back you know in the dormant phase the last step which is to retain the retention step this is the most important one at this point the customer decides that the customer is going to stay as the customer or the customer goes back you know of course torment um what are the best practices how can we make sure that the customer retains in the base make sure that the customer is at a peak level of spending um the customer is getting exactly what he signed up for the product or the retail and it also makes sure that the customer if going towards dormancy then we are having a very very strict proactive strategy in place to make sure that the customer comes back revives so all of these steps in the coming slides i'm going to go through them each one of them and then identify some of the models i've felt in past or some of the projects that we have done throughout my experience and ensuring how we can use data science in achieving these goals so attract and acquire attract and inquire the first part is to choose the customer how do we choose the customer so at this point if you're sitting in a b2b company you already have your leads and you need to make sure that the customer moves from a lead to and convert it into account or if it's um you know um if you're working coming from a retail background and you want the customer if the customer is coming on your website and identifying it's a completely new customer and he wants to know what he actually wants to buy and since in these cover times we have seen a very high increase in this pen pattern through our digital platforms so this is a very useful use case in there what happens is that when a customer comes on board you can actually cross reference that new customer to your existing customer base so based on your existing customer base you can build an audience modeling uh model where what happens is that existing customer has already a behavior pattern a spend pattern you might categorize them based on so these are the kind of customers who are more into um if it's issues um what kind of shoes what are the expense buttons what says you have also another kind of customer who is into more of sales the clearance section so you can basically map the new customers based on these existing customers and then recommend the next page or recommend a part of that website streaming to that specific customer so this not only gives the customer a very personalized um experience but it also makes sure that the customer is satisfied with the overall experience the exact same approach can be applied if you are putting a software as a service where for example you know exactly by now you you can you can hand out customer service and you can identify what exactly your customer need and then you can also map that with the existing businesses um which are operating as your customers in your company and you can say that these are the customers who are already using our product and these are the best configurations or the best way they're using the phone and you can use the exact same technique to engage your new customer because at this point in time you do not have the spend pattern or you do not have the behavior pattern of your new customer but what you do have is the existing existing customers and a lot huge information about your existing customers all you need to do is map it based on your audience modeling the next step is to target them with the right offer this is super important in the sense because maybe your customer is not looking for what is exactly being given to him in that personalized you know offer or that customer service so it's very important to listen to the customer here and then based on that give him the next best um offer based on what the customer actually needs how to do that so um one thing which has helped me a lot throughout my career and with the teams i've been working with too is the segmentation there are a lot of segmentation which you can done they're also known as clustering where basically what happens is that you divide the whole base in into homogeneous clusters so any group of customers exhibiting similar behaviors they get plugged into one cluster and then you can treat that one cluster as one micro segment this is also very contextual marketing approach where you are giving one specific offer to that specific micro segment based on the kind of customers clubbed in so segmentation can be divided into different streams based on high value low value i'll touch that later on in the slides as well at this point since the customer is a new sale you need to wait for a few weeks days or maybe one or two months based on the background you're coming from and based on that you can basically split your new customer base into specific segments so new segments every segment is automatically one cluster or two clusters and based on that cluster you can identify what exactly the customer needs and then target those specific customers in those specific clusters based on the past few months few days whatever your new sale duration would be ing you based on what they're doing in that specific customers this is also a very good technique because um i i've seen that in my experience as well you can not attack every single customer with one offer it's always the best practice to use a segmented broken down approach because in their funds you always have the best upticks um the customer is going to respond more positively to your campaigns or to your initiatives or even to your offer another thing which can be catered into this is the kind of customer for example if you're working with a b2b company if it's an smb or an enterprise you can also give them you know um customized products based on the kind of background the customer is coming from so if this at this point you need to understand as much as you want you can about the customer because at this point the customer has not generated enough in activities you know on your um using your product but you can know a lot about where the customer is coming from and why exactly the customer choose your product another very good um use case is the product affinities prod affinities is basically something which i will also be discussing throughout the slides as well at this point front definitely is will be used for new opportunities these are intelligent recommendations based on um there are many different algorithms available a priori and then you can also use neural networks so basically what you can do is that you can identify existing behaviors and then based on the existing customers and their behaviors you can map the next best opportunity or the next best product for that customer um it's very similar to segmentation you can also live them up together like identifying your clusters and then based on every cluster you can add a layer of your product affinity on top so this also helps in identifying the right product for that specific cluster um i personally am a big fan of this because even as a business owner you do not really um have the information of what the best product is you can always map productivity on top and then the affinities are going to give you the next product based on what the existing customers are doing and what your new customer is actually doing at the base right now so the first n days of activity n is basically based on completely relevant to your domain um a retail company might have a different end days of activity by end days i mean the total number of days you mark a customer as a new sale before you move it to the next segment so the best approach is to make sure that once you know what the value of n is for example in some companies it might be 90 days for some companies it might be two weeks for some companies it might be six weeks do you want to treat your customers a new sale customer for the next six weeks or three months based on the end sales you can update your affinities very frequently obviously it depends upon the kind of infrastructure you have but the best way to do it is to update them as frequently as possible because a new sale customer is the most active and showing the most dynamic behavior change versus the next two segments i'm going to talk about so it's very critical here to keep on updating your current affinities and every month a customer might change from one segment to the other be in segmentation or be in the within the affordability state so last month the customer might the next best offer might be you know offer a or campaign a but in the next month based on the last 30 days activity the change will be huge and you'll actually see the dynamic shift of a customer behavior based on what you've been doing in the past one month um based on the offers you'll also see um cluster movement so when you're converting a new sale from you know i'm i'm relatively new i've just been using the product for one week watches i've been using for three months or two months you're going to see a very distinctive change this change can be increase in usage this can be this change can also be a decrease in usage which automatically decreases revenue um it's decreasing engagement increase in engagement so these changes need to be tracked very regularly in the first three months and again it's super important because the customer is in dynamic the customer doesn't exactly know how can what is the best way to use the products so it's always always important to make sure that these models new sales and affinity are being updated very frequently um one of my favorite models to be used in new sale is the quality of sales um so in a previous presentation um uh he was talking about um random forest and you know how we can use random forest so we've used random forest a lot and one of the best use cases of the forest was to use in the quality of sales model because that was an organization where we had to identify a good quality sale versus a bad quality of sales because of which an accommodation structure was created and commission needs to be discharged so initially it was just one hard-coded value you know if the customer is doing xyz activities we mark it as a good sale versus a bad sale but um we replaced that with a very good quality of sales model which basically monitored um customer activity and within a seven days window it based on the random price outputs it used to give that these are the customers who are actually a good quality sale versus these are the customers which are bad quality of sales and this can be helped in use in your commission structure this can be used in the overall sale process and identifying what exactly needs to be done to make sure that there is 100 conversion rate or to make sure that if you are selling your product to one specific customer and it's not a business make sure that you're making that sale to a future high value customer the next is the engage and draw so engage and grow is the most critical part based that at this point in time the customer is already using your product the customer is has already generated a lot of data where you can on top apply a lot of business methodologies you can identify exactly who the customer is what the customer wants from your product um and at this point in time there are two things start needs to be done engage the customer more and then identifying growth potential so how do we base understand the customer again this is another leg of the segmentation um at this point you already know that the customer is no longer part of the new sale the customer is already part of the base for some specific time period which you've already said like the end set i was talking about so what you can do is split it into two kinds of segmentation one is the unit segmentation and the other is the life cycle segmentation by lifecycle segmentation this is specifically can be generated for those subscribers or those customers or those businesses who've been in the engage part of your customer life cycle and then you can split that into further streams for example if you're talking about telecommunication um i've had a lot of experience in that where we have customers who are mostly terror savvy they only use your connections for um a good quality of internet but they're also customers who only use your same your mno offers for texting or voice offers you know so they're different kind of customers you can split up your customer base based on the type of usage the type of products and then further split them into segmentations this not only helps you to go even more distinct like even more clear and concise micro segments but it also pinpoints exactly what the customer needs and reduces the error margin too um and then one of the thing which um is this this can be implemented across in any company is the customer profile 360. so what a customer profile 360 is that by this time you know how much the customer spends on your product you know how much the customer use it um on your product you also know how much the customer what how the customer is actually using the product um you can add in more psychographics and demographics variables on top based on the regions based on the background the customers coming from is the customer a student is the customer um a teacher is the customer you know a banker based on that you can identify how what's the best time to reach the customer how to reach the customer and then get the best customer level support or experience the customer can get based on not only what the customer is doing using your product but also pulling in all the psychographics for the demographics data as well um product upsell is also a very good way to increase the customer from a low value product to a high value product um there are multiple higher value propensity um prediction models which can be built based on one or two or multiple use cases for example um one of the business objectives which um top of my mind is that currently your customer base are generating a revenue of let's let's assume ten dollars per day and but you know that these customers also show a potential or this might be a target that you need to obviously increase your revenue um so what you can do is identifying um set of subscribers set of customers who are predicted to actually move towards a higher bundle or a higher product a higher value product where you can generate more revenue so you already have that prediction model in place all you need is a very good offer on top uh an offer which excites them an offer which makes sure that they actually move towards the higher bundle and based on that they will actually give you an incremental value you will not only move the needle but you're also going to move them from um giving them the more best experience and it might be much better than what they're already using this can be um you know for example you just might have noticed um especially in the north american market when you have a very big issue with your phone you call the contact center you tell them that i have connection issues or maybe i just want to cancel my connection they always have an extra package available to you which might be of lower value but they are going to give you some extra minutes or some extra mb to make sure that they retain you and make sure that you do not leave the network because they make sure that you stay on as a customer so that's that's where upselling or down selling the product um similarly you can also apply a product lineage part which is basically identifying monthly weekly or even daily prediction to identify forecast models there are a lot of models available forecast models based on the kind of data you have you have a lot of early based era you can apply specific model to that you have um region-wise tera you can apply a specific model to that and you can build up a lot of forecast models which can tell you that this is what the customer is doing right now and this is what the peak looks like in the next three weeks four weeks six weeks whatever your business objective is this is super important when you are identifying you know revenue potential you want to identify the markets where currently in existing scenario you do not see enough customers but the forecast model will give you a very far predicted date that these are the emerging markets or these are the markets where if you invest more you're going to generate a lot more revenue so forecast models are something which are very easy to implement um you have a lot of variety and then they not only help in the customer's overall experience but it also helps in the overall company's growth as well value realization and expansion so obviously when you're sitting in a company where you need to make sure that um your company's always making profit and you're not losing any customers you need to be very sure of what exactly the customer lifetime value is um the there are multiple different models available where you can identify the customer lifetime value or you can identify the next best action um create a pricing and profit optimization model to make sure that if your customer is currently existing you know in in this in a smaller segment but you see that the customer is showing the potential to move through a higher segment you not only make sure that the customer moves to a higher segment but this is where the account manager relationship comes in and then a proper pricing strategy is in place and this can be done by making a very perfect pricing and profit optimization model um one of the very good examples with i normally give here is so everyone uses instagram everyone uses netflix so whenever you open up netflix um it gives you a recommendation for a 96 person relevant and it pops up a new movie or a new series um that's what's happening behind that is there's it is they have this a huge huge ai model which is a pro recommendation engine which not only takes into account what how much time you're spending on um what category of movies like for example i i love watching horror movies so i always get more of you know if there's a new horror movie the trailer pops up and it shows me 98 relevant because number one i spend more time on netflix watching those movies number two anyone else who has the same profile as me is also watching those movies so they're mapping my behavior with the other person's behavior or that segment of clusters um that's the segment of customers and then making sure that i am recommended the best next product this hybrid approach is has worked in so many organizations um that it becomes even much better if you're increasing engagement if you're increasing revenue if you're reducing churn because let's take instagram so how instagram works is that every time you go on your explorer field and you're looking into pictures or movies of any any you know anything on the insta feed um it stores every single information which is being you know uh been generated by your phone for example how much time does mario actually spend on dog videos um how much time does maya spend looking at you know a specific friend or a best friend's group how much time do i spend in actually messaging on instagram how much how many times i spent liking you know all that information is being stored and then it's being utilized to make sure that every time i'm open instagram i'm only shown the relevant information so i'm why do i only see those those pictures or those videos which you know are super interesting to me and i spend more time because one they are increasing their goal of engagement and two they also pinpoint those specific ads you know uh via youtube or facebook that's that's exactly how they're monetizing all of this and that's that's how the whole model is running behind this all you need is a business objective so to increase that you need a professional recommendation engine you can also um apply a segmentation approach on top because then for example instead of putting one big huge recommendation engine my my suggestion would be that split your base into let's assume smb enterprise or high value low value customers and then a separate recommendation engine for a high value customer versus a separate recommendation engine for a low value customer this is going to be more customer centric this is going to be even more accurate and then this is going to be add more value to your overall organization as well um all of these models um on individual they do obviously give a lot of value but one thing which i've learned is when you club them together the output gets even much much better because it's it's not only a model in isolation if you pick up a base and then apply segmentation you might get five amazing clusters and for every single cluster if you have the infrastructure available you can recommend you know specific streams based on those clusters on top so you know apply recommendation engine on top of these clusters and then based on that you'll be even getting a more better and a concise micro segment the last is basically retention and revival so retention and revival is the trickiest the hardest of them all because this is where you want to make sure that the customer stays at the top the mature level which you receive by now and the customer doesn't go into the declining phase the dormant phase or the joint phase so you need to be very proactive here predict the decline there are many different models available which can be done and the model's aggressiveness can be defined based on the business objective you can add dormancy and prediction for example customers who are existing currently using your product they are currently part of your network they're currently part of your base but they are predicted to go dormant because number one they might be um in a region where they aren't too many marketing campaigns number two they might be showing a decline based on their existing trends versus what they were showing in the previous months or weeks so at that moment they are currently part of the base but it's very tricky to identify that they would actually go dormant so and and tormented prediction churn prediction this can be made multiple layers as well for example predicting subscribers who are going to stop using a product the next in the next week seven days or customers who are going to stop eating from the next 20 days 30 days versus 90 days so you can break up your base into three different categories and for every single category you can have a separate retention campaign available so for example if a customer is part of the top predicted bins in the next seven days you might have a very less aggressive campaign but you might you need to put the customer back into the old usage pattern which was a high value or more spent versus a 90-day prediction where you are sure that the person is going to go towards the 90-day you need to be even less aggressive because you still have 90 days to get the customer back versus the 7-day which the customer is going to be more urgently getting out of your base or stop using your product so based on these levels based on these you can call them churn scores um you can mark them based on different turn level turn scores um a spectrum where you have specific campaigns available or initiatives available to win back the customer um another good use case here is to identify um churnhub using fna and then apply mgm member get member initiative so for example um i know this example because um i've seen this happening with instagram and facebook as well so if i stop using facebook for a few days i get this email from facebook and based on that email it shows me some of my close friends a recent post my close friend has made on facebook and tells me or makes me excited that you know a friend of yours has posted this go check it out so how is that working um facebook has my sna so facebook knows that maria interacts with these are my closest connections facebook also know that maria has decreased her activity in the past three four days so i am in a potential churnhub for them but they also gave me initiatives that you know attract me back using my sna a member your member initiatives can also be applied similarly for example um in a telecommunication world we've done a lot of mdm initiatives where we know that um maria was talking to um my best friend all of all of these like past few months and i was constantly messaging her she was constantly messaging me so he was directly connected to each other in the social network analysis but in that sma map suddenly my friend's phone turns off um the person goes torment might be sudden normal might be slow dormant but based on that i am going to get a campaign an sms or a campaign offer that i can you know um win something or maybe get something free free db free minutes if i get my friend back on the network and this way my friend is going to get also um a retention off if you come back on the network but since her phone is closed they need me to contact trent llsd both can win extra minutes or extra 3gb just because you are dormant um and this has worked too because this work in those cases where the customer is already dominant and it's it's more of a reactive approach but this is a good way to approach your customer if the customer phone is not working or if the customer is not reachable um bring back customers of these so now we've talked about how exactly to identify that these are the customers but what are the best offers to you know and how do we choose the best offers that's where you can again use the product recommendation engine but this documentation engine is only using the data of a dominant subscribers so um the recommendation engine the logic the methodology is the same but every time i discuss it i'm using it in a different segment on a different set of customers so my output is going to be different this way so in a pro recommendation engine it's the same example as the call center one um if the customer is always dormant or predicted to be dormant you can upsell a product you can down to the product just to make sure that the customer sticks to your network the customer sticks as your customer and the customer doesn't go back um you can select the best low price of uh you can you can have windback offer um you you call your internet providers and tell them that your internet isn't working so you want to cut off the connection and go to a new internet provider they will send you some person right there and then at your place just to make sure that it gets better and give you even more customize and a special service that's how these works um another thing is to identify new opportunities here like fusing code recommendation or per dfinity you can say that a lot of customers who are existing showing propensity of something which for example um if you go back to the netflix example um i just saw a trailer of a new comedy show and i liked it but netflix doesn't know that i like it but what nice list does know is that the people in my cluster the best of the all of the customers netflix users in my cluster they are also using um going into that you know that category too so they are enjoying that one specific comedy show too so netflix will automatically recommend me a brand new product or a brand new thing um that this is something new for me and i might like it too because everyone else in my segment is like you get two so you can use again your user segmentation and prior affinities for the new opportunities and then identify those new opportunities as well this way you don't have to you know identify exactly what to offer because the recommendation engine is going to tell you that these are the products which are being you know enjoyed the most by similar group of customers or customers who are also um at the same level you know be it our pool use it or spend and then ashley paul center analytics this is again you can flag your churn customers or flag customers who are predicted to go dormant or fly customers who are very high value and have a very separate stream of high value you know calls calls from this and stream for them so basically whenever all these people call they all have different experiences when they're calling or when they are you know complaining about any offer or the product so they get a very customized experience which is going to make them feel special too and come back using your product too um another thing is the loyalty analysis so basically you can treat your customers with rewards or loyalty based on the age they've been part of your you know customer base or based on uh how much they've spent so for example you give them targets that if you spend this amount of you know um money you will get this much discount on a shoe or a code or anything retail wise and the same goes for in a b2b market too when you're talking about a customer and you say that if the customer stays with us and this is the amount of volume they generated we're going to give them this much more discount based on the amount of volume so so the more they spend the more discount they get uh and even with the loyalty you can give them loyalty discount as well which are super useful too in retaining the customer even if the customer is at the maturity level and not going for the dormant it it actually helps them to stay at the which is the maturity as well so i've talked about a lot of models and how we can use them in the customer journey all of these need to be have a very proper model evaluation and the results need to be carefully refreshed and make sure that they are being used in the best possible way i've worked with a lot of amazing data scientists who have produced amazing models and with a very high precision or accuracy and everything looks perfect after six months the model becomes void the model is not being used the way it was supposed to be used how do we make sure that the model gives the right quality of our food and it's being used in the best possible way so one of the things that we can do is to identify um you know how the customers are responding to the offers based on the model output which is known as the response modeling so it is a big feedback loop which we can implement so if i give a customer an offer like if you go back to the netflix example the cust netflix offers me a brand new comedy show so that recommendation is coming from the product affinity netflix product affinity engine whatever the ai is behind that if i actually watch the show till the full extent i actually spend the time allocated for that show to watch that means that i qualified for that campaign based on that they can then improve the model and then use my behavior to further retrain and refresh the model and provide even better outputs and this can go the other way around for example if i do not go and watch that show this might affect on the other way that they are going to use that information and then tone down tone it down in a way that they do not offer that show to other customers who have more similar behaviors than me and maybe use this information to the model might actually extract me from one cluster and put me in another cluster as well so this response modeling is super important because once you close the whole full back loop you get to know exactly how exactly the model worked and was it successful or not um another important thing is the campaign analytics so you had an output you identified the churn subscriber who was predicted to go dormant and then you had the best offer available for him you targeted the subscribers with the best customer with the best offer it was backed by marketing it had a very positive business case everything was working fine but how do we make sure that the customer actually retained back to the space um stayed back in the base because of what we did and it was not something that happened in the base in the control so it's very important to have a target control methodology based on this methodology you can identify what is the incremental uptake so control is a group of subscribers with exact similar behaviors exact same market segment it's the exact same um you know uh you can call them statistically significant too because they have the um right our pool right usage they are from the same segment but we do not offer them the whatever the campaign was we only target the target group of subscribers the target pool and then after the campaign ends you evaluate the uptake so this is the uptick how much the control actually did that specific business objective like coming back to the base not getting dormant or maybe um uh buy a new product versus how much did the target for who we actually made an effort to pitch them that product and how much did they basically um you know went for it so the target uptake versus the control update if you just convert compare these together you're going to get an incremental update so the target and control updates are almost the same that means that our efforts did not work as much as we hoped it do versus if the uptick is high like even a one to two person uptick based on your base that means that we did something right so it's very important to evaluate the campaign effectiveness because it not only shows you how effective was the offer but it also shows you how effective was your targeting how effective was the micro segment which you've identified based on the segmentation or prediction model or affinity or maybe an sme um real-time data is also super important because this is the basic of any kind of data science model you need to make sure that your information is as much recent as possible so the recency might be different for some companies it might be important for some like in a b2b it a recency might be one day or two day but if you're talking about a customer where you have huge amount of volume of data being generated like you know all these big major social platforms recently needs to be very very recent like the last one second or the last one minute but whatever information is being incorporated into your data pipeline it needs to be perfect 100 percent accurate if the data is not recent this accuracy decreases exponentially the customer might not be you know doing the showing the exact same spending patterns which the customer was showing one week ago two weeks ago but your model is going to keep on making the same predictions because your data was not refreshed in the right way more frequently or as much as it was supposed to get refreshed from um and then the last is the democratic democratization of data where it's it's super important for everyone across the organization to see all numbers and how is that done is by real-time dashboard so these real-time dashboards are um this can be made by many open source tool or you can actually pay for bigger amazing better tools too but it's super important to make sure that across all the organization everyone is looking at the same number the data is consistent um it's it's available to every decision maker be it marketing be it the product team be it um even the top management everyone used to know the exact same number and it needs to be one true picture i've seen organizations where they are multiple dashboards and every team is you know generating their own dashboard so if it's uh you know um average revenue per user one variable it's a one number but the same number can be calculated differently in a finance team um in a bi team in in marketing which it actually leads to many other different problems as well so this number needs to be consistent every number needs to be consistent it needs to be refreshed and then every person in the organization especially the decision makers um and decision making does do not need to be in cxo or cmo they need to be um they can be anyone in the department and they need to have the full visibility of what's happening in the whole company using those numbers everything needs to be data-driven if that is happening if the company's terror driven then only you will get the right resources and you will get to you know experience how to develop the best models and make sure that the models are being utilized in the best possible way um one more thing which needs to be done at this point is to make sure that um if if it's a new product or if it's a completely new um even from the technology perspective they need to know exactly how the marketing are doing what campaigns and how do those campaigns are you know improving the overall customer behavior or adding value to the customer's whole experience with the product and based on these numbers since everything is going to be data-driven the technology team can make their own improvements ingly so all teams need to work together based on um you know these numbers or based on the decisions they make so they need to make sure that everything is um consistent throughout the organization be it big b or small um so summarizing a little bit of what i've been you know talking about this is a very famous curve whenever you google customer life cycle you'll see this curve but what i've done here is i've plotted down um the different models which i've been using in the past or which you can use in your organizations or in your setup so the easiest one which i think are the easiest but they are um a little bit tricky too are the behavior and the life cycle segmentation you can build a behavioral life cycle segmentation based on even with a very less amount of data you have these these kind of models these are basically just segmentation you just need to be make sure that you're only giving in the right variables if you give in too much extra information the clusters are going to be um you know more interlinked together which is not a good quality to have in the behavior segmentation and then it's also easier to maintain so for example if you're talking about a b2b organization you you'll see exactly where your s b customers lie or where your enterprise customers are um it's it's also a very good step as a first you know i do it you know when i'm doing something um i just give it all of the customers and identify so the first level of clustering it shows you so much information like it clubs into specific clusters and then you see behaviors coming out of it and these behaviors are not something which you know already they're not you know something which are already eminent or completely hidden in your base you might see a big chunk of subscribers who are high value but they've not been tagged as high value or you might also see a big chunk of customers like retail you'll see that these are the customers who are who are showing the same you know spend part in like these are the customers who are buying shoes as well as accessories so there a lot of insights would get generated by simple business behavior segmentation then there are a lot of forecasting and proactive models that you can make turn prediction high to low low to high product usage device prediction um what kind of device the customer is using especially in the telecommunication domain you can also add like quality of sales model here so and these prediction models need to be very concise based on the business objective business objective is super important here every single model is going to be um giving you a very different output if your business objective needs to be changed so the list these are just some of the lists this can go on and on based on your company what you're looking for how you want to improve the customer experience and then my favorite deploy recommendation engine it not only gives you the upsell the best high level product it also gives you a downsell for the churn customers you know the customers work around you in a higher bundle or a higher package you want to give them a lower package just to make sure they stick to the network but it also gives you a cross sale so like the example i gave i'm into more into horror movies but maybe i might change into a comedy genre as well so that's how the recommendation engine works like if you find the most relevant products based on what the other customers are doing and what yourself are doing and then lastly the social network analysis um social network analysis are complicated and they're also um actually the prereq is to get abused infrastructure too and obviously more specialized data scientists do but they show you they give you information about especially in the large companies which information you cannot get out from any other models like your communities your queen bees you can identify which are joint hubs just based on the social media network analysis um and based on this you you'll see that the higher you go up the more business value these models create and obviously the more complex they become um it's harder to maintain them too but it's again it's it's not something which you know you can't live without social network analysis or it totally depends upon your business objective what you need to do and then pick the right model the right way to do it um that is it i think i'll hand it over to jenny hi wonderful that was such a great presentation thank you so much we have a handful of questions that people have submitted and if you have other questions please be sure to throw them in the q a and i'll pee them up so first is you walk through a lot of different ways to be able to aggregate and create different data models how do you actually recommend or what advice would you give to be able to communicate and convey that data and insight to a non-technical person um so it's it's it's something which we've learned um the hard way because once you're a data scientist you just so much invested into numbers and into algorithms that once you start talking to a business user you can't just bombard them like this is the precision this is the recall this is the confusion metrics um so the best way to do it is to go back to the business objective and then use the words that the business normally uses you know like for example in my case um i would say um a top predicted bins of turn subscribers you know for me that's basically those subscribers who are generating the highest score and coming in the top bins which are more likely to go down but to a business user i'm going to use a different line i'm going to say that these are the these are the set of subscribers out of the total customer base we have who have the most propensity or the highest probability to go dormant or to leave your customer base you know they don't understand bins or they won't understand the numbers as such as say you need to tell them exactly what they will understand based on the terminology they use um similarly if you're you know talking to a cxo or a cmo all they care about is engagement levels the number of customers engaged you know revenue um they don't really care about accuracies they don't really care about the algorithms so you told them when they attend that this is the revenue generated right now and this is the expected revenue that we can generate based on this implementation sure sure so then flip it if you are a marketer and you're uh lucky enough to either have a data scientist on your team or you're collaborating with the data scientists within your organization what are some tips that you would give that marketer to be able to get better alignment and stronger communication with their data scientists because they can't many times flip their language to be more technical like you have the ability to flip it to speak more of the business jargon what tips would you give them um so it's it's it's more like the business objective comes from the marketing team right so the business objective is something which is something common between the two teams all together um but but right now since like 10 years ago i would have had more challenges explaining what a data scientist team does but right now they're a lot more awareness and there there's a new terminology which is you know business analytics where the marketing genes are doing their own analytics too so they actually are more interested in the data science portion too i've i've received so many questions from people from up around the globe that how do we you know we have a complete marketing diagram but how do we run a python code how do we do resql so at this like right now it's even more um people are more into data science so it's it's easier for them to understand it more and then again when you even even as a data scientist um there's no harm in explaining them exactly how a model works like you need to explain them that this is how i predicted it plot a graph show them this is how it got predicted or these are how the slope goes what says um even in a random forest you can show them that this is what a tree looks like it's it's overwhelming but you'll see the percentages you know splitting up together and and they are so data driven that they get that too um so it's it's it's more like you know how i'm interested in marketing how they are using my model and versus how they interpret what the model is doing sure great tip so if uh you were to give advice to a marketer who is hiring a data scientist for their team what advice would you give them as far as finding a good candidate so i think um if if a marketing team is hiring a data scientist team they would expect the data science team that that specific data scientist to do a lot more than build models the data scientist is also going to be in charge of the etl the data is also going to be in charge of converting the external data sources into actual you know um an analytical layer the data source centers is also going to be responsible of visualization so the marketing um the hiring manager needs to look into even even if the data scientist is is not a phd you know in a machine learning specific role if the data scientist has relevant experience in managing um a complete data pipeline on his own starting from connecting the right sources to building a model be it python bhr and then visualizing it you know and look or tableau whatever tool the person gets um and then making sure that the visualization is as much descriptive you know even if the data scientist is not there in that meeting the marketing team understands exactly what's happening in the dashboard that's what they need to look for in in that data scientist and and at this point i think that it's it's a mix of all different fields it's um you know a bi developer a visualization expert and you know someone who's expert in python or good coding techniques yeah yeah i definitely agree and i did the presentation very beginning of today and it actually laid out all the different roles if each person were to have a different role or function within your bi team all the different roles you would need in order to have a really fully staffed updated team and it's quite a few people if you're right if you're only going to have one they need to be very well-rounded to be able to do that so last question maria if the marketers here want to continue learning more about the specifics of models um where should they go what kind of education or training is available online that you would recommend so um i know there are um there are a lot of different available certifications or universities are offering online you know classes for specific roles but the marketing the person needs to know exactly how much into data sciences the custom person needs to go into because it's it's more like you know even me being in the in this in this field for 12 years i still don't know all of the data science like i'm still learning something new every single day so the first question for me you know for them i i would ask like how much do you want do you want to just do you know um a simple python code you know some statistics in our or do you actually want to build you know an sna or you know a big you know a prediction model or a churn framework and based on that my recommendation would be that there are a lot of schools um and a lot of online courses um which also have a lot of course materials from ivy schools like john hopkins and even ibm and all these different companies and they have a very good material on that and they give you a little bit of you know introduction to that if this is going to be intermediate data scientist or a beginner data scientist and i i always route them towards them um another very good is dela camp um but again the problem with data camp is that you need to have a lot of time because it starts from absolutely scratch and then build it up you know to the actual work um one more thing which they can do is how i actually learn new stuff is that um in our day-to-day role in in our you know um in our work in our jobs we face these a lot of different questions you know like for example how do we do this how do we do that and if the person is very much enthusiastic into data science they can just actually note all of those questions and then start something on their own you know when they're in the spare time and that's what i do i just you know pick up the question like you know what if we have this or what if we have that kind of model you know and when you get time you just you have the business objective right there you're sitting on top of the data all you need is you know run an open source python code or r and then start testing it i think that's the best way to learn it and you'll get so many questions like all these online platforms like github stack overflow you just google one question and you'll have millions of you know search results with actual code working code and you yeah you just start off from there um i think that's the best way to do it i love it trial by fire that's been a general consensus for all of the presenters today is like just do it just jump in so well maria thank you so much the presentation today was absolutely wonderful and for all the attendees stick around because in four minutes we have chris and sophie from facebook we're going to have about a 20-minute presentation from them that we're going to watch and then we're going to do a fireside chat to round out the day their presentation is around using measurement strategies around innovative and up and coming technology so should be pretty interesting but maria thank you so much for your presentation it was a joy to have you thank you so much for having me have a good day