T-SQL count distinct and group by distinct IDs - sql-server

Question

I have the following type of table:
BranchName CustID
========== ======
Branch1 1111
Branch1 1111
Branch1 2222
Branch2 2222
Branch2 4444
Branch3 1111
Branch4 3333
What I'm trying to achieve is to count distinct CustID and group them by branches, without repeating any CustID.
Basically I'm trying to get to this:
BranchName DistCountofCust
========== ======
Branch1 2
Branch2 1
Branch3 0
Branch4 1
I tried this code:
SELECT X.BranchName, COUNT(DISTINCT X.CustID) as DistCountofCust
FROM
(SELECT T.BranchName, T.CustID
FROM MyTable T) as X
GROUP BY X.BranchName
It doesn't give correct result (doesn't count the number of CustID per branch correctly, because CustIDs overlap for certain branches). Is it possible to eliminate distinct CustIDs and group them per given branches? (In the final result I need only unique customers to be listed for branches).

Answer 1

You may use NOT EXISTS for this. I'm assuming that branches have certain priority according to their name.
SELECT T1.BranchName, COUNT(DISTINCT X.CustID) as DistCountofCust
FROM MyTable T1
LEFT JOIN (
SELECT BranchName, CustID
FROM MyTable T1
WHERE NOT EXISTS (
SELECT 1
FROM MyTable T2
WHERE T1.BranchName > T2.BranchName and
T1.CustID = T2.CustID
)
) as X ON T1.BranchName = X.BranchName and T1.CustID = X.CustID
GROUP BY T1.BranchName
dbfiddle demo

Answer 2

One answer for you. As I say in my comment, ideally you want to have a separate table for your branches, as otherwise, when filtering your results, you can't do a count of 0 9as that would return no rows, and be omitted). This is why there is a second CTE to get the DISTINCT values of BranchName.
CREATE TABLE #Branch (BranchName varchar(10), CustID int);
INSERT INTO #Branch
VALUES
('Branch1',1111),
('Branch1',1111),
('Branch1',2222),
('Branch2',2222),
('Branch2',4444),
('Branch3',1111),
('Branch4',3333);
WITH FirstCustomer AS(
SELECT *,
ROW_NUMBER() OVER (PARTITION BY CustID ORDER BY BranchName) AS RN
FROM #Branch),
Branches AS ( --really you should have a seperate table to store your branches in
SELECT DISTINCT BranchName
FROM #Branch)
SELECT B.BranchName,
COUNT(FC.CustID) AS Customers
FROM Branches B
LEFT JOIN FirstCustomer FC ON B.BranchName = FC.BranchName
AND FC.RN = 1
GROUP BY B.BranchName;
DROP TABLE #Branch;