7  Audit

[To my understanding] Audit includes tools and work stipulated by Standards. Audit Data Analytics (ADA) replaces excel-related tools with R/Python to improve efficiency/effectiveness. It does not necessarily reduce audit work required by ISCA. The following example is to audit expense claim based on data from payroll, hr, and finance departments, which demonstrates ADA is a vital move for auditors from all possible perspectives.

Compared to excel-related tools, it could be easily used to test audit assertions (e.g., occurrence, existence, completeness, cut-off, valuation, classification) after reconciled in terms of P2P, O2C, Payroll, R2R, GL.
1. benefit: version control diff, lightweight size, powerful 1m rows, automation script.
2. pattern recognition: spot deviation and inconsistency.

It also addresses common mistakes throughout the audit process. For instance,
1. version control: which version of PBC data is the latest?
2. reproducible: my result is different from yours after rerun.
3. report: check if number in working papers tally to those in financial statement.
4. automation: roll out audit work next year by copy+paste.

Diff 1

Diff 2

7.1 Cleaning

exp_claim_raw <- readxl::read_excel("isca_cpe_2023/1. Anomalies in Payroll data.xlsx", 
                                    sheet = 1, 
                                    range = "A1:G33") %>% 
  janitor::clean_names()

hr_data_raw <- readxl::read_excel("isca_cpe_2023/1. Anomalies in Payroll data.xlsx", 
                                  sheet = 2) %>% 
  janitor::clean_names()

pay_data_raw <- readxl::read_excel("isca_cpe_2023/1. Anomalies in Payroll data.xlsx", 
                                   sheet = 3, 
                                   skip = 2, range = "A3:D25") %>% 
  janitor::clean_names()

df_comb <- exp_claim_raw %>% 
  full_join(hr_data_raw, by = c('staff_id' = 'staff_id')) %>% 
  left_join(pay_data_raw, by = c('staff_id' = 'staff_id'))

df_clean <- df_comb %>% 
  mutate(across(contains("date"), lubridate::dmy)) %>% 
  mutate(on_leave = lubridate::dmy(on_leave)) %>% 
  mutate(staff_name = coalesce(staff_name, name.x)) 

# check if amount is correct
sum(df_clean$amount_s.x, na.rm = TRUE)

df_clean %>% 
  distinct(staff_id, amount_s.y) %>% 
  summarise(app_c = sum(amount_s.y, na.rm = TRUE))

sheets <- list("comb" = df_comb, "clean" = df_clean) 
writexl::write_xlsx(sheets, here::here(paste0('audit_sit/audit_payroll', Sys.Date(), '.xlsx')))
openxlsx::openXL(here::here("audit_sit/audit_payroll2023-12-22.xlsx")) 

df_clean <- readxl::read_excel(here::here("audit_sit/audit_payroll2023-12-22.xlsx")) %>% 
  mutate(across(c(contains("date"), on_leave), lubridate::dmy))

7.2 Procedure

# cross check payroll amount against finance amount
df_clean %>% 
  group_by(staff_id, staff_name) %>% 
  summarise(amt_exp = sum(amount_s.x), 
            amt_paid = sum(amount_s.y) / n(), 
            amt_diff = amt_exp - amt_paid, 
            .groups = 'drop') 

# compare date to ensure no claim happens before incurred or after resigned
df_clean %>% 
  dplyr::filter(claim_date > expense_date) 

df_clean %>% 
  dplyr::filter(claim_date > last_date | claim_date == on_leave) 

# identify multiple claims for the same expense
df_clean %>% 
  group_by(staff_id, staff_name, purpose, amount_s.x) %>% 
  dplyr::filter(n() > 1) 

# ensure staff name and their bank account number updated timely
df_clean %>% 
  dplyr::filter(!is.na(edits_to_hr_data), 
                bank_account_no.x == bank_account_no.y) 

df_clean %>% 
  dplyr::filter(name.x != name.y)

# produces audit working paper
library(pointblank)

ag <- df_clean %>% 
  create_agent(label = "A very *simple* example.", tbl_name = "payroll") %>%
  col_vals_between(columns = claim_date, left = vars(expense_date), right = vars(last_date)) %>%
  interrogate()

ag

Audit Procedure 1

Audit Procedure 2

7.3 Enhanced

df_clean %>% 
  count(staff_name, sort = TRUE)

df_clean %>% 
  dplyr::filter(grepl("\\d+?", purpose)) %>% 
  mutate(purpose = gsub("\\d+?", "", purpose)) %>% 
  mutate(across(where(is.character), ~na_if(., "AB99"))) %>% 
  mutate(staff_id = replace_na(staff_id, 0)) 

df_clean %>% 
  select(contains("date"), purpose) %>% 
  mutate(if_taxi = case_when(str_detect(purpose, "Taxi") ~ "taxi", 
                             TRUE ~ "other"),   
         total_date = lubridate::floor_date(claim_date, "week"), 
         first_date = first(total_date)) %>% 
  slice_max(order_by = claim_date, n = 3)

df_clean %>% 
  dplyr::filter(!is.na(amount_s.x)) %>% 
  mutate(new = (amount_s.x %/% 100) * 100) %>% 
  group_by(new, amount_s.x > 300) %>% 
  summarise(new1 = mean(amount_s.x), .groups = 'drop')

df_clean %>% 
  dplyr::filter(!is.na(staff_name)) %>% 
  group_nest(staff_id, staff_name) %>% 
  mutate(new = map(data, ~pluck(.x, 4))) %>%
  mutate(new1 = map(new, ~paste(.x, collapse = '|'))) %>%
  select(-data, -new) %>% 
  unnest(new1)

df_clean %>% 
  dplyr::filter(!is.na(staff_name)) %>% 
  select(staff_id, staff_name, purpose) %>% 
  summarise(new1 = paste(purpose, collapse = '|'), .by = c(staff_id, staff_name)) 

df_clean %>% 
  select(staff_id, staff_name, division, purpose, amount_s.x) %>% 
  dplyr::filter(!is.na(purpose)) %>% 
  separate(purpose, into = c("type", "info"), 
           extra = 'merge', remove = FALSE, fill = 'right') %>% 
  group_by(division, type) %>% 
  summarise(n = n(), 
            amt_type = sum(amount_s.x), .groups = 'drop') %>% 
  arrange(-amt_type)

library(lubridate)

df_clean %>% 
  pivot_longer(cols = where(is.Date), 
               names_to = 'activity_date', 
               values_to = 'detail_date', 
               names_pattern = "(.*)_.*", 
               names_transform = list(activity_date = toupper))