JavaScript (Node.js), 61 55 54 bytes

Saved 1 byte thanks to @nwellnhof

Takes input as an array of characters.

s=>s.map(p=c=>r+=p>(p=(+c?~c:1-Buffer(c)[0])%6),r=0)|r

Try it online!

How?

For all characters but digits greater than $0$, the 0-indexed column $x$ is given by:

$$x=(c-1)bmod 6$$

where $c$ is the ASCII code of the character.

For positive digits $n$, we need to do instead:

$$x=(n+1)bmod 6$$

Examples:

"a" --> (97 - 1) mod 6 = 96 mod 6 = 0

"b" --> (98 - 1) mod 6 = 97 mod 6 = 1

"0" --> (48 - 1) mod 6 = 47 mod 6 = 5

"3" --> ( 3 + 1) mod 6 =  4 mod 6 = 4

Commented

s =>                       // s = input string (as array)

  s.map(p =                // initialize p to a non-numeric value

  c =>                     // for each character c in s:

    r +=                   //   update the result r:

      p > (                //   compare p with

        p = (              //   the new value of p defined as:

          +c ?             //     if c is a positive digit:

            ~c             //       -(int(c) + 1)

          :                //     else:

            1-Buffer(c)[0] //       -(ord(c) - 1)

        ) % 6              //     apply modulo 6

      ),                   //   yields 1 if the previous value is greater than the new one

    r = 0                  //   start with r = 0

  ) | r                    // end of map(); return r

Jelly, 11 bytes

⁾04yO‘%6<ƝS

A monadic Link accepting a list of (uppercase) characters.

Try it online!

How?

First replaces any '0's with '4's (so the rest of the code treats them as being in the rightmost column). Then casts to ordinals, adds one and modulo's by 6 to get 0-based column indices. Then compares neighbours with is-less-than and sums the result.

⁾04yO‘%6<ƝS - Link: list of characters         e.g. "BLAST20"

⁾04         - list of characters = ['0', '4']

   y        - translate                             "BLAST24"

    O       - ordinals                              [66,76,65,83,84,50,52]

     ‘      - increment                             [67,77,66,84,85,51,53]

       6    - literal six

      %     - modulo                                [ 1, 5, 0, 0, 1, 3, 5]

         Ɲ  - neighbourly:

        <   -   less than?                          [  1, 0, 0, 1, 1, 1  ]

          S - sum                                   4

Perl 6, 45 39 bytes

{sum .[1..*]Z<$_}o{(2 X-.ords)X%46 X%6}

Try it online!

Works with uppercase letters. (2-ord(c))%46%6 computes the reversed x coordinate.

K (ngn/k), 32 31 bytes

+/>':(+6 6#,/"a10"+!'26 9 1)?0+

Try it online!

Clean, 85 bytes

import StdEnv

(s=sum[1\a<-s&b<-tl s|b>a])o map(e=(toInt e+if(e-'1'>'/')1 -1)rem 6)

Try it online!

Ruby, 56 bytes

->s{w=9;s.count{|c|w<w=[*?a..?z,*?1..?9,?0].index(c)%6}}

Try it online!

Preliminary naive version, will be golfed.

Japt -x, 14 bytes

®rT4 c Ä u6Ãä<

Try it online!

Port of this Jelly answer. Takes input as an array of characters, with letters uppercase.

Explanation:

®rT4 c Ä u6Ãä<    :

®          Ã      :Map each character through:

 rT4              : Replace 0 with 4

     c            : Get the char-code

       Ä          : Increment it

         u6       : Modulo 6

            ä<    :Replace with 1 if you had to move right, 0 otherwise

                  :Implicitly sum and output

Java (OpenJDK 8), 73 bytes

Not a bad solution for Java! That zero being on the right-hand side cost me several bytes.

t->{int a=9,c=0;for(int d:t)c+=a<(a=(--d+(d/48==1?2:0))%6)?1:0;return c;}

Try it online!

Explained

t -> {                          // Lambda taking a char array as input

    int a=9,                    // Initialise last column value

        c=0;                    // Initialise frustration count

    for(int d:t)                // Loop through all chars in title

        c+=                     // increment the frustration count if...

          a<                    // The last column is smaller than the current column

            (a=                 // Set last column to current column

              (--d+             // Decrement ascii value of char

                  (d/48==1      // If ascii decremented ascii value is between 48 and 95

                    ?2:0)       // increment by 2 (1 total) or 0 (-1 total)

                )%6)            // Mod 6 to retrieve column index

            ?1:0;               // Increment if to right hand side

    return c;                   // return calculated frustration count

}

05AB1E, 12 11 bytes

-1 byte thanks to @Kevin Cruijssen

¾4:Ç>6%¥1@O

Another port of Jonathan Allan's Jelly answer. Takes input in uppercase.

Explanation:

¾4:Ç>6%¥1@O   //full program

¾4:           //replace all '0's with '4's

   Ç          //get ASCII code points

    >         //increment

     6%       //modulo 6

       ¥      //get deltas

        1@    //is >= 1

          O   //sum

Try it online!

Retina 0.8.2, 46 bytes

T`l1-90`1-61-61-61-61-61-6

.

;$&$*

&`;(1+);11

Try it online! Link includes test cases. Explanation:

T`l1-90`1-61-61-61-61-61-6

List the alphabet and digits in the order on the OSK and map each one to a (1-indexed) column number.

.

;$&$*

Convert each column number to unary.

&`;(1+);11

Count the number of columns that are followed by a larger (i.e. rightwards) column. The &` allows the matches to overlap.

Python 2, 84 bytes

def f(s,k=6):j=(ord(s[0])+('0'<s[0]<':')*2-1)%6;return(j>k)+(~-len(s)and f(s[1:],j))

Try it online!

Mathematica, 102 bytes

Differences[Last@@Join[Alphabet,ToString/@Range@9,{"0"}]~Partition~6~Position~#&/@#]~Count~_?(#>0&)&

Pure function. Takes a list of characters as input and returns a number as output. This is a pretty naive solution, golfing suggestions welcome.

PHP, 74 81 77 bytes

for(;$o=ord($argn[$i]);$i++&&$f+=$p<$x,$p=$x)$x=(--$o/48^1?$o:$o+2)%6;echo$f;

Run as pipe with -nR or try it online.

C (gcc),  82 79  77 bytes

o;c(i){i+=i<60&i>48?1:5;i%=6;}f(char*s){for(o=0;*++s;o+=c(*s)>c(s[-1]));o=o;}

Try it online!

This function will only support lowercase inputs

Ungolfed and commented:

o; //Used for output

c(i){             //Calculates the column of given character

     i+=          //Correct i to get the correct column

        i<60      //If i is a digit...

        & i>48   //... but not '0'

        ?1           //Then move it one column on the right

        :5;          //Else move it five columns on the right

     i%=6;        //Get the column number

}

f(char*s){                        // The actual "frustrating" function

          for(                    //Loop for each character

              o=0;                //reinitialize output

              *++s;               //move to next character / while this is not ''

              o+=c(*s)>c(s[-1])  //Increment if current character is on the right of the previous one

             );

           o=o;                   // Outputs result

}

If my function is allowed to accept wide character strings, it can be reduced to 76 bytes with:

o;c(i){i+=i<60&i>48?1:5;i%=6;}f(int*s){for(o=0;*++s;o+=c(*s)>c(s[-1]));o=o;}

Try it online!

This version just accept input as int* instead of char*

Edits:

• Golfed 3 bytes in the calculation of the column (function c)


• Golfed 2 bytes thanks to ceilingcat