A slight variation to Luis's method and implemented as an inline table valued function

CREATE FUNCTION compressDuplicates(@code nvarchar(4000))

RETURNS TABLE WITH SCHEMABINDING AS

RETURN

WITH cteTally as (

SELECT TOP(LEN(@code) - 1) ROW_NUMBER() OVER (ORDER BY (SELECT NULL)) N

FROM (VALUES (0),(0),(0),(0),(0),(0),(0),(0),(0),(0),(0),(0),(0),(0),(0),(0)) E1(N)

,(VALUES (0),(0),(0),(0),(0),(0),(0),(0),(0),(0),(0),(0),(0),(0),(0),(0)) E2(N)

,(VALUES (0),(0),(0),(0),(0),(0),(0),(0),(0),(0),(0),(0),(0),(0),(0),(0)) E3(N)

),

startpattern as (

SELECT SUBSTRING(@code,1,1) P

UNION ALL

SELECT SUBSTRING(@code,N+1,1) P

FROM cteTally

WHERE SUBSTRING(@code,N,1) <> SUBSTRING(@code,N+1,1) COLLATE Latin1_General_BIN -- Change to suit comparison required

)

SELECT CAST(r AS NVARCHAR(4000)) result

FROM (

SELECT P

FROM startpattern

FOR XML PATH('')

) AS A(R);

If there is non numerics in your string you will need to decide how to handle case, etc and set the collation accordingly