Here is a simple base R implementation, see my comment for details.

I will leave it to you to work out a version which works well with {dplyr} syntax. The data masking is different. "dplyr-style" is close to subset(). If you need assistance, do not hesitate to comment.

Data

data3 = data.frame(
  customer = c(1,2,3),
  frequency = c(30,32,36),
  recency = c(72,71,74),
  TX = c(74,72,77),
  monetary_value = c(35.654,47.172187,30.603611))

Implementation

of log_div_mean() (do you have a reference for the calculation?)

log_div_mean = \(.data, # data 
                 .x, .y, .z, # columns of interest
                 a = .6866195, b = 2.959643,  # default values 
                 r = .2352725, alpha = 4.289764 # which can be overwritten
                 ) {
  .u = .data[[.x]]
  r1 = r + .u  
  r2 = log( (alpha + .data[[.y]]) / (alpha + .data[[.z]]) )
  r3 = log(a / (b + max(c(.u, 1)) - 1)) # typo in your max?
  rr = r1 * r2 + r3
  1 / (1 + exp(rr))
}

where we use the variable naming routine present in the {tidyverse}.

Application

> log_div_mean(.data = data3, .x = "frequency", .y = "TX", .z = "recency")
[1] 0.9619502 0.9730688 0.9340070

Correct results?