SQL进行渠道分析

Question

I have a scenario where people can do action1 and bail out; or action1 & action2 & bail out, or they can do action1, action2 & then say "I want to play more". I want to find number of people at each level:
- Did action1
- Did action1 and action2
- Did action1 and action2 and also said "I want to play more"

Each time user does an action, we log a sessionId. So if the use run action 1 and action 2, we will have two rows with same session id, same customer id and but different actions.

╔══════════════════════════════════════╦════════════════╦════════╗
║              SessionId               ║ FirstRunAction ║ UserId ║
╠══════════════════════════════════════╬════════════════╬════════╣
║ 039af321-457e-41a6-b303-41ca935b0877 ║ action_1       ║ eb6    ║
║ 039af321-457e-41a6-b303-41ca935b0877 ║ action_2       ║ eb6    ║
╚══════════════════════════════════════╩════════════════╩════════╝

The above data is in one table. Action1 and action2 are in UserAction table but "I want to play more" is in a separate table called "Play" table.

I did the following but its not correct.

I want to have an if type of logic. Search for action 2 only when action 1 is there & search for "I want to play more" when that session also has action1 and action2. i wrote the one below, but don't know how to store results at any level & if what I am doing is correct. I have 6 million data to start with. Any help?

Action1Results = SELECT [SessionId]  
                        ,[Action]  
                        ,[UserId]  
                 FROM [Test].[dbo].[UserAction]  
                 WHERE [Action] = 'action_1';  

Action2Results = SELECT [SessionId]  
                        ,[Action]  
                        ,[UserId]  
                 FROM [Test].[dbo].[UserAction]   
                 WHERE [Action] = 'action_2';      

PlayMoreResults = SELECT [SessionId]  
                        ,[Play]  
                        ,[UserId]  
                 FROM [Test].[dbo].[UserPlay]  
                 WHERE [Play] = 'I want to play more';  

FinalResults = SELECT [SessionId]  
                     ,[UserId]  
                 FROM [Test].[dbo].[Action] with (nolock)   
                 INNER JOIN [Test].[dbo].[UserPlay] with (nolock)   
                 ON [Test].[dbo].[UserPlay].SessionId = [Test].[dbo].[Action].SessionId;  

Actually there could be many actions but they will always end with _1 and _2. The list of actions could change over time.

Eg I can have actions like:

'Write_1" , "Write_2", "Birds_1", Birds_2", "Pen_1", "Pen_2".

So for these I have to see how many did

'Write_1" -> "Write_2" -> 'I want to play more'',

then how many did

'Birds_1" -> "Birds_2" -> 'I want to play more'',

then how many did

'Pen_1" -> "Pen_2" -> 'I want to play more''

and so on. So funnel for each type.

Answer 1

I assume action can only be ('action_ 1','action_2') other wise you have to compare for those value on the ON-WHERE and the CASE`

SqlFiddle Demo

WITH CTE as (
     SELECT UA1.SessionID, 
            UA1.UserId, 
            UA1.FirstRunAction Action1,
            UA2.FirstRunAction Action2,
            UP.Play Action3
     FROM UserAction UA1
     LEFT JOIN UserAction UA2
            ON UA1.SessionID = UA2.SessionID
           AND UA2.UserId = UA2.UserId
           AND UA1.FirstRunAction <> UA2.FirstRunAction           
     LEFT JOIN UserPlay UP
            ON UA1.SessionID = UP.SessionID
           AND UA2.UserId = UP.UserId
     WHERE UA1.FirstRunAction = 'action_1'
), classify as (
SELECT CASE 
           WHEN Action3 IS NOT NULL AND Action3 = 'I want to play more' THEN 'TYPE 3'
           WHEN Action2 IS NOT NULL THEN 'TYPE 2'
           WHEN Action1 IS NOT NULL THEN 'TYPE 1'
           ELSE 'TYPE 0'
       END as actionType
FROM cte 
)
SELECT actionType, count(*)
FROM classify
GROUP BY actionType

CTE portion OUTPUT

| SessionID | UserId |  Action1 |  Action2 |             Action3 |
|-----------|--------|----------|----------|---------------------|
|         1 |      1 | action_1 | action_2 | I want to play more |
|         2 |      2 | action_1 |   (null) |              (null) |
|         3 |      3 | action_1 | action_2 |          Don’t Know |
|         4 |      4 | action_1 |   (null) |              (null) |
|         5 |      5 | action_1 | action_2 |              (null) |

• So UserID = 1 has all three action will be type3
• UserID = (2, 4) only have one action will be type1
• UserID = 3 has 3 action but last one isnt I want to play more so is type 2 as UserID = 5

Final Output

| actionType |   |
|------------|---|
|     TYPE 1 | 2 |
|     TYPE 2 | 2 |
|     TYPE 3 | 1 |

If You have a table USERS you could start with that so also get count of how many user has TYPE 0

Answer 2

This should give you a count of distinct users by each session_category -

  3 = I want to play more
  2 = action_1, and action_2
  1 = action_1

Note this will count a user multiple times across session_categories if the user in fact does only action_1 on one occasion, and action_1+action_2 on another. You don't explain how you plan to deal with this situation (unless I missed something).

https://gist.github.com/leonpanokarren/56f313130118dad47113

I tried pasting this query inline here a gazillion times in vain.