Loading & Preparing Data
options(scipen=10)
pacman::p_load(latex2exp,Matrix,dplyr,tidyr,ggplot2,caTools,plotly)
## package 'latex2exp' successfully unpacked and MD5 sums checked
## 
## The downloaded binary packages are in
##  C:\Users\adm\AppData\Local\Temp\RtmpUNCjYl\downloaded_packages
rm(list=ls(all=TRUE))
load("data/tf4.rdata")

購買機率與預期營收的分布

par(mfrow=c(1,2), cex=0.8)
hist(B$Buy)
hist(log(B$Rev,10))

B %>% ggplot(aes(x=age,y=Rev)) + 
  geom_boxplot() + scale_y_log10()

group_by(B,age) %>% 
  summarise(n=n(), Buy=mean(Buy), Rev=mean(Rev)) %>% 
  ggplot(aes(Buy,Rev,size=n,label=age)) + 
  geom_point(alpha=0.5,color='gold') + 
  geom_text(size=4) +
  scale_size(range=c(4,20)) + theme_bw()  -> p
ggplotly(p)


帶有「參數」的成本效益函數

定義、畫出效用函數 由於c()是一個常用的R內建功能,以下我們用x代表成本 \[\Delta P = f(x|m,b,a) = m \cdot Logis(\frac{10(x - b)}{a})\]

DP = function(x,m0,b0,a0) {m0*plogis((10/a0)*(x-b0))}
par(mar=c(4,4,2,1),cex=0.7)
curve(DP(x,m=0.20,b=30,a=40), 0, 60, lwd=2, ylim=c(0, 0.25),
      main="F( x | m=0.2, b=30, a=40 )", ylab="delta P")
abline(h=seq(0,0.2,0.05),v=seq(0,60,5),col='lightgrey',lty=2)

期望報償的算法: \[\hat{R}(x) = \left\{\begin{matrix} \Delta P \cdot M \cdot margin - x & , & P + \Delta P \leq 1\\ (1-P) \cdot M \cdot margin - x & , & else \end{matrix}\right.\]

估計毛利率(margin)

# load(data/tf0.rdata)
# group_by(Z0, age) %>% summarise(sum(price)/sum(cost) - 1)
margin = 0.17  # assume margin = 0.17

估計預期報償

m=0.2; b=25; a=40; x=30
dp = pmin(1-B$Buy, DP(x,m,b,a))
eR = dp*B$Rev*margin - x
hist(eR,main="預期報償分佈",xlab="預期報償",ylab="顧客數")

🌻 有多少顧客的預期報償大於零?

🌻 如果我們針對所有顧客做促銷,預期報償將是?

🌻 如果我們針對預期報償大於零顧客做促銷,預期報償將是?

sum(eR[eR>0])
## [1] 74578.71


市場模擬

單一參數組合
m=0.2; b=25; a=40; X = seq(10,45,1)

df = sapply(X, function(x) {
  dp = pmin(DP(x,m,b,a),1-B$Buy)
  eR = dp*B$Rev*margin - x
  c(x=x, eReturn=sum(eR), N=sum(eR > 0), eReturn2=sum(eR[eR > 0]))
  }) %>% t %>% data.frame %>% 
  gather('key','value',-x)

df %>% ggplot(aes(x=x, y=value, col=key)) + 
  geom_hline(yintercept=0,linetype='dashed') +
  geom_line(size=1.5,alpha=0.5) + 
  facet_wrap(~key,ncol=1,scales='free_y') + theme_bw()

不同的參數組合
mm=c(0.20, 0.25, 0.15, 0.25)
bb=c(  25,   30,   15,   30)
aa=c(  40,   40,   30,   60) 
X = seq(0,60,2) 
do.call(rbind, lapply(1:length(mm), function(i) data.frame(
  Inst=paste0('Inst',i), Cost=X, 
  Gain=DP(X,mm[i],bb[i],aa[i])
  ))) %>% data.frame %>% 
  ggplot(aes(x=Cost, y=Gain, col=Inst)) +
  geom_line(size=1.5,alpha=0.5) + theme_bw() +
  ggtitle("Prob. Function: f(x|m,b,a)")

市場模擬:不同的參數組合的比較
X = seq(10, 60, 1) 
df = do.call(rbind, lapply(1:length(mm), function(i) {
  sapply(X, function(x) {
    dp = pmin(1-B$Buy, DP(x,mm[i],bb[i],aa[i]))
    eR = dp*B$Rev*margin - x
    c(i=i, x=x, eR.ALL=sum(eR), N=sum(eR>0), eR.SEL=sum(eR[eR > 0]) )
    }) %>% t %>% data.frame
  })) 

df %>% 
  mutate_at(vars(eR.ALL, eR.SEL), function(y) round(y/1000)) %>% 
  gather('key','value',-i,-x) %>% 
  mutate(Instrument = paste0('I',i)) %>%
  ggplot(aes(x=x, y=value, col=Instrument)) + 
  geom_hline(yintercept=0, linetype='dashed', col='blue') +
  geom_line(size=1.5,alpha=0.5) + 
  xlab('工具選項(成本)') + ylab('預期報償(K)') + 
  ggtitle('行銷工具優化','假設行銷工具的效果是其成本的函數') +
    facet_wrap(~key,ncol=1,scales='free_y') + theme_bw() -> p

plotly::ggplotly(p)

每一個工具的最佳參數

group_by(df, i) %>% top_n(1,eR.ALL)
## # A tibble: 4 x 5
## # Groups:   i [4]
##       i     x   eR.ALL     N  eR.SEL
##   <dbl> <dbl>    <dbl> <dbl>   <dbl>
## 1     1    31 -200257.  7471  82082.
## 2     2    37 -178947.  8833 130921.
## 3     3    20  -36138. 11231 101633.
## 4     4    10 -246960.     0      0


討論問題

par(cex=0.7, mar=c(2,2,1,2))
table(B$age) %>% barplot


🗿 討論問題:
  如果上述4組工具參數分別是某折價券對4個不同年齡族群的效果:
    ■ I1 : a24, a29
    ■ I2 : a34, a39
    ■ I3 : a44, a49
    ■ I4 : a54, a59, a64, a69
  如果你可以在這4個年齡族群之中選擇行銷對象,你應該如何:
    ■ 選擇行銷對象(N)?
    ■ 設定折價券的面額(x)?
    ■ 估計預期報償(eR.SEL)?