Einstein Analytics and Discovery Insights Specialist - Step 2

1. Create the Einstein Discovery Story from the model
   • Select Maximize Tenure as story goal, Insight & Predictions as story type
   • Include the following variables in manual mode : Subscription Date, Modem Age, Postal Code, Name, Account Number, Region, Payment Method, Contract, Online Security, Tech Support, Online Backup, Internet Service
2. Deploy the Prediction
   • Go to the model reated from the story, and select Deploy Model
   • Name the prediction as 'Predicted Tenure'
   • Connect the predictionto the Subscriber Object
   • Leave the mapping as is (All fields should be mapped)
   • Create a new prediction field and name it as 'Predicted Tenure'
   • Ignore Monitor Model Accuracy
   • Select Contract and Payment Method as Action Variables (if you can't see all the variables, zoom out with your browser)
3. Add Einstein Prediction Lightning Card to Subscriber Record Lightning Page
   • Using the Lightning App Builder, Create a Record Page named 'Default' for the Subscriber Object
   • Choose Clone Salesforce Default Page and select Grouped View Default
   • Drag and drop the Einstein Predictions Lightning Component into the Right Side Panel and choose 'Predicted Tenure'
   • Activate the page and assign it as Org Default for the Desktop
4. Add Prediction field to Subscriber List View
   • Using Object Manager, set the Field Level Security to 'Visible' for your profile (should be System Admin)
   • Create a new List View and name it 'All Subscribers'
   • Include: Subscriber Name, Predicted Tenure and any other fields you wish to show up and save the list view

And you should be all set,

 Step 1 in the superbadge is a challenge that seems to be all about understanding the windowing SAQL functions used to create the chart that creates a few of the subscriber activity. I was barely able to understand how to create this dataset from the responses on here, but eventually I got it. E.g. Some people are calling "Current quarter cancellations" as "New Quarter Cancellations" whihc is not how it is described in teh activity and in general the problem for me comes from the way people use developer lingo to describe their solutions and no one seems to comment their code which is really confusing to me. 

q = load "Beattie_Subs";
-- Current Quarter Cancellations
cancelled = group q by ('Churn_Date_Year', 'Churn_Date_Quarter');
cancelled = foreach cancelled generate cancelled.'Churn_Date_Year' + "~~~" + cancelled.'Churn_Date_Quarter' as 'ActivityDate_Year~~~ActivityDate_Quarter', count() as 'Current Quarter Cancellations';
-- New Quarter Subscribers
subs = group q by ('Subscription_Date_Year', 'Subscription_Date_Quarter');
subs = foreach subs generate subs.'Subscription_Date_Year' + "~~~" + subs.'Subscription_Date_Quarter' as 'ActivityDate_Year~~~ActivityDate_Quarter', count() as 'New Quarter Subcribers';
-- Previous Quarter Subscribers
subs = group subs by ('ActivityDate_Year~~~ActivityDate_Quarter');
subs = foreach subs generate subs.'ActivityDate_Year~~~ActivityDate_Quarter' as 'ActivityDate_Year~~~ActivityDate_Quarter', coalesce(sum(subs.'New Quarter Subcribers'),0) as 'New Quarter Subcribers', coalesce(sum(sum('New Quarter Subcribers')) over ([-1..-1] partition by all order by ('ActivityDate_Year~~~ActivityDate_Quarter')),0) as 'Previous Quarter Subcribers';
-- Previous Quarter Cancellations
cancelled = group cancelled by ('ActivityDate_Year~~~ActivityDate_Quarter');
cancelled = foreach cancelled generate cancelled.'ActivityDate_Year~~~ActivityDate_Quarter' as 'ActivityDate_Year~~~ActivityDate_Quarter', coalesce(sum(cancelled.'Current Quarter Cancellations'),0) as 'Current Quarter Cancellations', coalesce(sum(sum('Current Quarter Cancellations')) over ([-1..-1] partition by all order by ('ActivityDate_Year~~~ActivityDate_Quarter')),0) as 'Previous Quarter Cancellations';
-- Churn Rate is Current Quarter Cancellations / (New Quarter Subscribers + Previous Quarter Subscribers - Previous Quarter Cancellations)
resultchurn = cogroup subs by 'ActivityDate_Year~~~ActivityDate_Quarter' full, cancelled by 'ActivityDate_Year~~~ActivityDate_Quarter';
resultchurn = foreach resultchurn generate coalesce(subs.'ActivityDate_Year~~~ActivityDate_Quarter', cancelled.'ActivityDate_Year~~~ActivityDate_Quarter') as 'ActivityDate_Year~~~ActivityDate_Quarter', sum(cancelled.'Current Quarter Cancellations')/(sum(subs.'New Quarter Subcribers') + sum(subs.'Previous Quarter Subcribers') - sum(cancelled.'Previous Quarter Cancellations')) as 'Churn Rate';

[Attrition Cost]

q = load "Beattie_Subs";
q = filter q by 'Tenure' in all;
result = filter q by 'Churn' == "Yes";
result = group result by all;
result = foreach result generate count() as 'count', count()*950 as attrCost;
result = limit result 2000;

[Reference Line]

"referenceLines": [
    {
        "color": "⌗E84340",
        "label": "{{cell(Attrition_Cost_1.result, 0, \"attrCost\").asString()}}",
        "value": "{{cell(Attrition_Cost_1.result, 0, \"attrCost\").asString()}}"
    }
],

Happy holidays from NYC!  In quasi-code, I did step one like this:

1. Load the dataset into q
2. filter q into a new stream (cancellations) by Churn (you want the number of cancellations, so set the churn bit appropriately)
3. Group cancellations by ChurnDate year/quarter but project the date as ActivityDate year/quarter; count() as current quarter cancellations, 
4. you then need to lag cancellations by one period, so generate again but this time collect 'cancellations' by sum and in a separate column, include a window function with sum to pick up the prior period.  I won't tell you have to do it, but read up on windowing
5. group q by Subscription_Date  year/quarter
6. generate a count() for "subscribers" and like above, project subscription_date year/quarter as ActiviityDate year/quarter.
7. re-generate the stream from step 6 as a sum of "subscribers" and add lagged subscribers by the same windowing sum that you did for cancelllations.
8. then you can either cogroup with a full join on ActivityDate year/quarter (and then generate the formula you need) or union and regroup by ActivityDate year/quarter.

Remember that the formula for churn in this step is current quarter cancellations/(current subscribers + subscribers from a quarter ago - cancellations from a quarter below)

filter q by {{row(Tenure_Length_1.selection, [0], [\"min\",\"max\"]).asRange(\"Tenure\")}};