28 - Intro to Einstein Discovery - Part 1 - The Business Scientist and Use Case Foundations

28 - Intro to Einstein Discovery - Part 1 - The Business Scientist and Use Case Foundations

Show Video

in this chapter we are going to cover einstein discovery einstein's discovery comes with tablocia platform or the product uh especially if you have the tcrm plus license or the einstein predictions licenses or any license related to the product that provides the einstein discovery feature and we are going to cover a few basic terms and basic introductions so we're going to go over the introduction to einstein discovery ed the role or the skill sets needed for a business scientist this is aka also known as citizen data scientist this is the new persona for which this tool would be uh pretty much their uh tool of choice and how to approach a use case so we've had a lot and we will be covering the technical aspect of the product how to use it but we also want to cover how to approach a business use case or a use case how to build that up and know how to apply the product for that use case the importance of the business knowledge this is going to be very important again relating back to the business scientist persona the good and the not so good fit for ed some important terms as i mentioned and the typical analytics assets related to einstein's discovery a quick exercise as an introduction and talk about limits um what you can do the product from volume from predictions etc all right what is tableau serum einstein discovery what are einstein's discovery pretty much the predictions that get deployed from that um i would like to start say that it's in the same way tableau is data for everyone tcrm is actionable predictions for everyone or you know specific einsteins covers actionable predictions for everyone that is what we're trying to do uh democratize democratize ai and predictions for the end users and spread the sme knowledge uh to the end users too and i'll i'll cover that in a bit more details so the product is a leader in augmented analytics ai built in it's part of the platform you'll see it in a bit prediction scores but also explanations and natural language and actions recommendations recommended actions what you can do to increase or decrease and maximize minimize that particular variable that you're trying to predict um through these predictions it is native to salesforce meaning the platform itself the data the uh the place where the model sits everything is in salesforce so and it is native again to the salesforce environment it's api ready you can take the predictions deployed and score any data outside as long as you can you know connect to the api or send the information to the api it is app centric we build it around a business use case and again we will cover this in details and the value is pretty much this driving the efficiency of the end users augmenting the daily business flow that i deal with as an end user with intelligent insights that allow me to do my job much more efficiently and again contribute to that particular valuable variable that we're trying to maximize and minimize um so einstein the word einstein by itself there are many products across salesforce that you can utilize or use the word einstein so i want to clarify a little bit where einstein's coverage sits if you think about the spectrum of einstein or ai products we have you can think about these as three main buckets to the left are out of the box ai features you have a license for them you pay for it for it and it automatically shows uh let's say if it's a predictive predicted field it shows automatically on the object and you use it you don't control how it's spelled um there probably are requirements to have x y z of these particular fields and these particular objects to have that feature works so pretty much these are out of the box not a lot of control we move to the middle bucket this is where it's supervised ai with the tool that helps you build the model create the stories uh and deploy deploy models as predictions so you have um you have a say in how the model is built you can see the explanations um you can bring in data from outside salesforce you can combine data you can use the predictions outside so there's a lot of features here and this is pretty much where we're targeting that persona of the supervised ai or with einstein discovered and of course there's the third bucket where it's pretty much the deep learning uh data scientist heavy heavy and this is where we just provide i mean but in a simple way we provide the platform to build that particular ml model whatever you need to do and again it's more targeted to the heavy data scientist persona so with that einstein discovery again it is integrated in tableau serum so i know we call it different names but it's pretty much it is within the stack um you know you get the data to a data set through the data platform you can prep it through data prep you combine it add the right fields etc and then you build a story or a model we use those two words simultaneously it's pretty much where you are again doing your ai work building that model and then you deploy it as a prediction and that prediction can be consumed somewhere else or outside um you know or you can rescore back into the data layer but pretty much what i'm trying to cover in this particular slide is that it is part of the platform uh it is a layer intelligence layer intelligent layer in the platform in tableau crm and all of this is within salesforce um and again as a reminder you can bring data that is natively from salesforce or from outside and this is the beauty here that you can actually build models based on both uh you know combined data and you can then deploy again back to salesforce objects or you can just deploy outside as an api service or even to tableau whatever you want to do again depending on the use case what you're trying to do with the product and this is the ultimate end user experience um specifically in this particular case is the salesforce user and what we're trying to do again with all this augmented analytics is provide that rich daily experience for an end user who's logging into their salesforce environment looking at their accounts they're dealing with their transactional data on a daily basis this is what i do on a daily basis maybe you know adjust the field create an activity or a task etc but also i have my typical analytics and this is my trends my historical analytical insights this is your typical reporting or dashboards or you know typical analytics that you know embedded uh in the account page but also on the right side i have the predictions and this is the new right i have a prediction score i have causes so it's not just a black box uh number that i don't know why this number is 77 i actually see some of the uh leading causes i keep even can see the how to improve it so this is very important this is where i can take an action immediately without even going back and you know emailing someone or waiting on a resolution or etc i am again spreading that that's the thing i mentioned i'm spreading the sme knowledge knowing what to do in a certain situation to the end user through these recommendations i i don't want to use the word recommendation through these improvements for this particular or these actions for this particular score or this particular field and again this is the augmented analytics this is how we want to drive uh predictions like i mentioned it's predictions uh actionable predictions for everyone and it's very important as we're going to mention on this you know probably multiple times that it's not just getting the model that gives me the right number and the right action but also operationalizing ai and this is very important because as you're going to see in this particular video here that we're going to cover um you know the terms and we're going to do an exercise how to build this or at least get the get the the predictive field but how do you operationalize this how do you bring everything that you build fast and uh in in a fast way to the end users that's what we're trying to do uh and we're gonna show you in the other chapters or videos when you deploy predictions so einstein's discovery is again i mentioned this is a powerful augmented analytics tools it focuses on four questions that have become a staple in analytics which are the descriptive diagnostic predictive and prescriptive analytics and essentially what they mean is what happened i'm looking at the data what happened this is historical this is typically where we would use reports and dashboards why it happened we could solve this historically with some dashboards granted that we had the right chart and the right kpis and metrics and dimensions looking at them at the same time and then we could deduce something why a trend is going down or up but now with ai this has become automated and i'm looking at all of the possibilities of my dimensions and measures interacting and i'm looking at it statistically in in a way that you know it's being statistically proven not just like a uh you know business hunch or uh an estimate of why it could it could have happened based on my experience and then more importantly the what will happen and what to do about it and these are very important meaning that now that i know why it happened and what can happen in the future what should i do about it so these are what we talk about unbiased explanations predictions and recommendations i'm biased in the senses i'm not coming here with my two years or four years or 10 years of experience which might be true which might be you know uh i might be saying the same thing as what td is going to say um but that's the validation for it because it it is validating that it's equal to the same experience that i've had but also through math and through statistics so a few things also too important that the the insights the predictions of prescriptions are very important in the sense of we providing a prediction with explanation this is very important business use cases and you're going to see that we are focusing on solving you know typical business use cases we're not going out and trying to solve for example you know uh you know climate change and and predicting big you know complex problems out there um this is important because we want to keep the ability to explain the insights that you will see this also as we go through the rest of the chapters here or videos it is rapid in the sense of i can load data run a story check the model i you know the model metrics are not good go back change something the data run it again fast or change the inputs or add some derived fields again that iterative process or the ability to just load the data of course after i've prepped it and changed it and run the story of the model multiple times is a is a very fast thing and it helps me get you know faster time to results or faster value of my product without sacrificing the quality of what i am trying to create here and uh it is supervised or augmented in the sense that we control as as the business scientists or the experts using this tool we control what's going on what's what are the input variables you're going to hear me saying input and output right or explanatory um so output is what we're trying to minimize and maximize input are the fields we are including this data set we're including the story we're saying you know go back and study the relation between these fees and their values if they if they are affecting the output variable in a certain way and it is important and i'm going to focus on this multiple times here the business understanding of uh or the business business knowledge uh of of what we're trying to do so if if it's a customer chair if it's a sales opportunities if it's a lead conversion if it's uh service csat etc etc you need to understand what the business cares about what is uh what makes sense from a business perspective and uh this will influence the data to be used the the scope of the data to be used and and even insights that you're seeing and validating for example recommendations in the engine maybe it says because you're going to see understand discovery as data improvements it's going to say like you know these fields are duplicates uh they they provide the same information so maybe that's something you want to verify with the business before just going with a biased opinion um you know if if it's not super clear if they are duplicates or not etc and uh what can we do with einstein discovery we can create again supervised models that are used to predict uh or to be deployed as predictions i'll go over how these terms or assets fit together on a different slide but mainly what we can do with the product right now is binary classification and numeric uh use case so binary classification use case or numeric use case meaning we can uh tell you uh you can use this tool to minimize maximize a number uh that can go you know from you know it's an amount of opportunity it could be anything that's a linear that's a numeric so units sold days to close revenue customer satisfaction that's a number and i sense cover automatically once you select that this is the field i'm trying to predict to maximize or minimize um it detects what type it is and it implies it applies the correct for example linear regression in this case if it's a lot if it's a binary if it's a did i win or not this opportunity true or false one or zero chain will not churn it will detect it's a binary and do a logistic regression now under the hood there are different algorithms we're going to cover that later on but this is at a high level what kind of regression is going to be used so um and i'm covering this so to clarify that for for example we don't do multiple clustering we don't do uh classification more than two um in this particular use case for example that that's uh that's another one and um to just visualize it really quick uh typically you want to predict a number in this case it goes maybe from zero to whatever value it is so that's how a linear might look like and again we're trying based on the data that we have the historical data these are all data we know now we're trying to predict so in the future if any any new value comes around here we're going to say okay well our prediction is going to be this value okay and uh if it's a binary classification again we're trying to split uh if it if data comes in if or if you know whatever that data comes in and maybe it's here well it's okay it's then it's a it's a circle if it comes here then we're going to say it's a an x etc and uh here's some more examples uh on the cases or the the the einstein discovery what what it works on numeric outcomes like we mentioned translated to business questions how much revenue can i expect how long will this deal take to close how much will this house sell for for a true false outcomes is this customer going to churn is this lead going to convert will the student graduate on time etc etc now why i'm covering this again one more time because there's going to be an important role for that business scientist and again this is a a persona think about it it could be the analyst or the data if they know the business uh you know well or a combination of all these three people two people even the data scientists if it is uh on the team sorry if if it is uh uh you know if it is a requirement or if that team exists and you know they are engaged in the beginning or at the end to validate the models etc all of these together i'm just going to call them business scientists and the business scientists job is going to be crucial to take the business problem which could be like we're losing money or we're not moving uh fast enough or customer we're losing customers or you know there's some efficiency that can be done take those statements translate them to what metric um that can be minimized or maximized to provide the the that value that that benefit uh value from the product okay and uh here's some more common use cases you can look at the uh you know under the different industries or under the different clouds whether sales or pharma or fund you know financials or wealth retail or other all of these are valid use cases that we have customers that have covered using einstein discovery so you heard you heard me mentioning about the business scientist and this is going to be important the business scientist is a persona that understands the business so 33 business expertise they have to because and i will cover that a little bit more detail but pretty much because they need to relate what the tool can do for the business 33 data and analytics expertise so they can't understand some data concepts um they might have dealt with some reports and dashboards before etc or they have came i can't come up with the kpis and metrics that the business care about or cares about and the last circle the red which typically these personas or these teams lack is the statistical terms knowledge or some predictive tool uh familiarity so what i mean by this is you're going to see in iceland discovery there are the easy parts where you look at the chart and look at the explanations etc but there's also the not so easy parts where you're looking at the model performance model accuracy model metrics r square or msc or even compu confusion matrix and and the threshold those things need a little bit of a statistical information or knowledge you don't have to be an expert but it is good and it is kind of uh you know assume that you will be able at least to glance at them and know what they mean or know what if the values are you know are high or low or something needs to be changed or not also if you typically have a data scientist team who will be validating or providing the last approval on the model accuracy to be deployed you want to be able to talk the same language and kind of relate so again you don't have to be a data scientist but you need to know these statistical terms a little bit and again as a reminder that we are focusing on the business use cases and we are providing this in the ui you don't have to write python or art from scratch and even the need for developers and integration and deployment or operationalizing the the final product which is the prediction everything is taken care of for within the product itself all right we're going to cover the technical terms and again how to use the product but how do i approach my business problem and categorize it in a way that i can use einstein's discovery to help me you know solve that problem so there's going to be six questions typically six five or six questions and you want to follow this template now again there's multiple ways there's different ways out there or you might find that you know you might add one or two steps but this is pretty much the core of it so first you want to start with what is the problem you're trying to solve what is it exactly the problem you're trying to solve because you're gonna walk into a meeting or you're gonna go you know if you're on the customer side still your own customers are the business and you're going to hear a lot of different business use cases a lot of metrics a lot of asks or demands so you're going to have to walk through that kind of prioritize them or get an agreement on what use case you're going to start solving and define if you know go on defining that then you're going to create the variable that tracks the performance or tracks the metric that affects that problem and again we're going to cover customer change so if you think it's customer channel then it's the churn flag like i need the churn flag did the customer chain or not for example so again you have to create the variable because that is what einstein's discovery does it maximizes or minimizes a variable um are there any glaring problems so for example when you create the variable is there any problem with that variable do you need to maybe create a a new variable for example if we're talking about uh um promotions i want to see if certain uh promotion is working as expected well one when do i measure do i measure the day i uh deploy or or put that promotion on the shelf or after three days or after a week customer churn do i look at all you know for a customer to churn if there's a flag for that customer one or zero well a customer can be 10 years and a customer can be one year and you know eventually all customers change so how do you find that flag that valuable you just agreed on how do you know it is uh you know it works for your business use case because again all customers churn so if you say i'm gonna just use the flag by itself that's a problem there's a problem with that variable you're not tying it to a time a window so that you know so your definition is a customer that churns within a year or two years right so again that's kind of you know could be the problem with the variable you you just thought it's very simple and just just a flag um the level of insight so when you this is very important important because we're going to create data sets and that data set has to have a grain that is the one record per customer if it's customer chain for example that would be one record per customer if it's opportunity it's one record per opportunity so you want to make sure your data set has a unique record pair up so you know when you build the story model there's no confusion it is one record one occurrence one observation of that particular opportunity or customer decide the appropriate scope of the data how many years you want to go back and look at the data do you want to look at your last five years of data or two years did the business change you know maybe your business model change or the service you just deployed only two years ago for these particular customers versus your business has been running out for you know for 10 years so you want to know what's the scope of the data you're going to bring into that data set and what the variables you you are including in the model the variables that you're including that story or model do they make sense are they historical meaning if you have a variable that is input or it's being populated after the event happens that might not be helpful for the model for example if you take an opportunity and you are trying to uh predict if this opportunity is going to close one or close lost and you're including multiple fields you know from the opportunity object well one of the fields might be opportunity lost reason so if you include this in the model on the story then einstein's cover gonna predict 100 saying i know this opportunity if it's going to be won or lost if the last reason is populated well that doesn't help me that this this last reason is always populated after the opportunity closes after the prediction should have happened or might or would have happened in the future so again these kind of fields you don't want to include them in your story or in the data set [Music] you

2021-05-08 01:50

Show Video

Other news