---
title: "Stage 1 · Descriptive KPI Analysis"
subtitle: "Network overview · Process funnel · KPI heatmap · Regional distribution"
---
```{r setup}
#| include: false
library(tidyverse)
library(plotly)
library(gt)
library(patchwork)
library(sf)
library(igraph)
library(tidygraph)
library(ggraph)
library(lubridate)
source("../R/utils_viz.R")
source("../R/utils_kpi.R")
claims <- readRDS("../data/claims.rds")
partners <- readRDS("../data/partners.rds")
proc_events <- readRDS("../data/process_events.rds")
```
## Network Overview
### Partner Distribution by Type
```{r partner-type-bar}
#| fig-height: 4
p_type <- partners |>
count(type) |>
ggplot(aes(x = reorder(type, n), y = n, fill = type)) +
geom_col(width = 0.65, show.legend = FALSE) +
geom_text(aes(label = n), hjust = -0.3, size = 3.5, colour = "#003781", fontface = "bold") +
scale_fill_manual(values = c(
Werkstatt = "#003781",
Glaspartner = "#0066CC",
Gutachter = "#00A9CE",
Abschlepp = "#66B5E8",
Mietwagen = "#FF6600"
)) +
scale_y_continuous(expand = expansion(mult = c(0, 0.15))) +
coord_flip() +
labs(title = "Partner Count by Type",
subtitle = "60 partners across 8 Swiss regions",
x = NULL, y = "Number of Partners") +
theme_allianz(grid = "y")
ggplotly(p_type, tooltip = c("x", "y")) |>
layout(hoverlabel = list(bgcolor = "#003781", font = list(color = "white")))
```
### Claims Profile Summary
```{r claims-profile}
#| echo: false
library(gtsummary)
library(gtExtras)
claims |>
select(claim_type, vehicle_class, coverage_type, fault_indicator,
repair_cost, duration_days, csat_score, fraud_flag, steering_flag) |>
mutate(
steering_flag = factor(steering_flag, labels = c("Not steered", "Steered")),
fraud_flag = factor(fraud_flag, labels = c("No fraud", "Fraud suspected"))
) |>
tbl_summary(
by = claim_type,
statistic = list(
all_continuous() ~ "{median} ({p25}–{p75})",
all_categorical() ~ "{n} ({p}%)"
),
label = list(
vehicle_class ~ "Vehicle Class",
coverage_type ~ "Coverage Type",
fault_indicator ~ "Fault",
repair_cost ~ "Repair Cost (CHF)",
duration_days ~ "Duration (days)",
csat_score ~ "CSAT Score",
fraud_flag ~ "Fraud Flag",
steering_flag ~ "Steering"
),
digits = all_continuous() ~ 0
) |>
bold_labels() |>
add_overall() |>
modify_caption("**Claims Dataset Overview** · Median (IQR) or Count (%)") |>
as_gt() |>
tab_style(
style = cell_fill(color = "#003781"),
locations = cells_column_labels()
) |>
tab_style(
style = list(cell_text(color = "white", weight = "bold")),
locations = cells_column_labels()
) |>
tab_options(table.font.size = px(12))
```
### Partner Network Graph
```{r network-graph}
#| fig-height: 6
library(visNetwork)
# Nodes: regions (large circles) + partners (smaller dots by type)
region_nodes <- distinct(partners, region) |>
transmute(
id = region,
label = region,
group = "Region",
size = 28,
font.size = 14,
title = paste0("<b>", region, "</b>")
)
type_colors <- c(Werkstatt="#0066CC", Glaspartner="#00A9CE",
Gutachter="#66B5E8", Abschlepp="#6B7280", Mietwagen="#FF6600")
partner_nodes <- partners |>
transmute(
id = partner_id,
label = partner_id,
group = as.character(type),
size = 12,
font.size = 9,
title = paste0(
"<b>", name, "</b><br>",
"Type: ", type, "<br>",
"Region: ", region, "<br>",
"Capacity: ", capacity_class
)
)
nodes <- bind_rows(region_nodes, partner_nodes)
edges <- partners |>
select(from = region, to = partner_id) |>
mutate(color = "#66B5E8", width = 1)
visNetwork(nodes, edges, width = "100%", height = "520px") |>
visGroups(groupname = "Region", color = list(background = "#003781", border = "#002060"),
font = list(color = "white", size = 14, bold = TRUE)) |>
visGroups(groupname = "Werkstatt", color = list(background = "#0066CC", border = "#004D99")) |>
visGroups(groupname = "Glaspartner", color = list(background = "#00A9CE", border = "#007A99")) |>
visGroups(groupname = "Gutachter", color = list(background = "#66B5E8", border = "#3399CC")) |>
visGroups(groupname = "Abschlepp", color = list(background = "#6B7280", border = "#4A4F5A")) |>
visGroups(groupname = "Mietwagen", color = list(background = "#FF6600", border = "#CC5200")) |>
visOptions(
highlightNearest = list(enabled = TRUE, degree = 1, hover = TRUE),
nodesIdSelection = list(enabled = TRUE, main = "Select a partner…",
style = "font-size:13px; padding:4px")
) |>
visPhysics(
solver = "forceAtlas2Based",
forceAtlas2Based = list(gravitationalConstant = -60, springLength = 120),
stabilization = list(iterations = 200)
) |>
visLegend(width = 0.15, position = "right", main = "Node Type") |>
visInteraction(tooltipDelay = 100) |>
visLayout(randomSeed = 42)
```
### Regional Volume Heatmap
```{r regional-volume}
#| fig-height: 5
regional_stats <- claims |>
group_by(region) |>
summarise(
n_claims = n(),
pct_steered = mean(steering_flag),
avg_cost = mean(repair_cost),
.groups = "drop"
)
claims |>
count(region, claim_type) |>
ggplot(aes(x = claim_type, y = region, fill = n)) +
geom_tile(colour = "white", linewidth = 0.5) +
geom_text(aes(label = scales::comma(n)), size = 2.8, colour = "white", fontface = "bold") +
scale_fill_gradient(
low = "#66B5E8",
high = "#003781",
name = "Claims"
) +
labs(title = "Claim Volume by Region × Type",
subtitle = "2021–2024",
x = "Claim Type", y = NULL) +
theme_allianz(grid = "none")
```
## Process Funnel Analysis
```{r process-funnel}
#| fig-height: 5
event_durations <- proc_events |>
arrange(claim_id, timestamp) |>
group_by(claim_id) |>
mutate(
prev_ts = lag(timestamp),
step_days = as.numeric(difftime(timestamp, prev_ts, units = "days"))
) |>
ungroup() |>
filter(!is.na(step_days), step_days >= 0) |>
group_by(event_type) |>
summarise(
median_days = median(step_days),
p25 = quantile(step_days, 0.25),
p75 = quantile(step_days, 0.75),
.groups = "drop"
) |>
mutate(event_type = factor(event_type,
levels = c("FNOL","Inspection","RepairStart",
"RepairEnd","Payment","Closed")))
# Waterfall-style funnel using lollipop
ggplot(event_durations, aes(x = event_type, y = median_days)) +
geom_segment(aes(xend = event_type, yend = 0),
colour = "#66B5E8", linewidth = 2.5, alpha = 0.6) +
geom_point(colour = "#003781", size = 5) +
geom_errorbar(aes(ymin = p25, ymax = p75),
colour = "#0066CC", width = 0.25, linewidth = 0.8) +
geom_text(aes(label = sprintf("%.1fd", median_days)),
vjust = -1.2, size = 3.2, colour = "#003781", fontface = "bold") +
scale_y_continuous(expand = expansion(mult = c(0, 0.2))) +
labs(title = "Median Duration at Each Process Step",
subtitle = "Error bars = IQR (25th–75th percentile) · d = days",
x = "Process Step", y = "Days (median)") +
theme_allianz(grid = "y")
```
### Funnel by Claim Type
```{r funnel-by-type}
#| fig-height: 6
proc_with_type <- proc_events |>
left_join(claims |> select(claim_id, claim_type, steering_flag), by = "claim_id")
proc_with_type |>
arrange(claim_id, timestamp) |>
group_by(claim_id, claim_type) |>
mutate(
days_since_fnol = as.numeric(difftime(timestamp,
timestamp[event_type == "FNOL"],
units = "days"))
) |>
ungroup() |>
filter(event_type %in% c("RepairEnd", "Closed")) |>
group_by(claim_type, event_type) |>
summarise(
median_days = median(days_since_fnol, na.rm = TRUE),
.groups = "drop"
) |>
ggplot(aes(x = claim_type, y = median_days, fill = event_type)) +
geom_col(position = "dodge", width = 0.65) +
scale_fill_manual(
values = c(RepairEnd = "#0066CC", Closed = "#003781"),
labels = c(RepairEnd = "Repair completed", Closed = "Claim closed"),
name = NULL
) +
scale_y_continuous(labels = scales::number_format(suffix = "d")) +
labs(title = "Days from FNOL to Key Milestones",
subtitle = "By claim type · Median across 2021–2024",
x = "Claim Type", y = "Days since FNOL (median)") +
theme_allianz()
```
## KPI Heatmap
```{r kpi-heatmap}
#| fig-height: 9
kpis <- compute_kpis(claims, partners)
kpis_cost <- cost_benchmark(kpis)
speed_df <- speed_benchmark(claims, partners)
kpi_long <- kpis_cost |>
left_join(speed_df, by = "partner_id") |>
select(partner_id, name, type,
cost = cost_zscore,
csat = avg_csat,
reopen = reopen_rate,
volume = n_claims) |>
# Standardise all KPIs to z-scores for comparability
mutate(
cost = -cost, # flip: lower cost = better (positive z)
csat = scale(csat)[,1],
reopen = -scale(reopen)[,1], # flip: lower reopen = better
volume = scale(volume)[,1]
) |>
pivot_longer(c(cost, csat, reopen, volume),
names_to = "kpi", values_to = "z_score") |>
mutate(kpi = factor(kpi,
levels = c("cost","csat","reopen","volume"),
labels = c("Cost","CSAT","Quality\n(1-Reopen)","Volume")))
# Show top-30 partners by avg z-score
top30 <- kpi_long |>
group_by(partner_id, name) |>
summarise(avg_z = mean(z_score, na.rm = TRUE), .groups = "drop") |>
slice_max(avg_z, n = 30) |>
pull(partner_id)
kpi_long |>
filter(partner_id %in% top30) |>
plot_kpi_heatmap()
```
## Top / Bottom 10 Partners
```{r top-bottom}
#| fig-height: 5
#| layout-ncol: 2
kpis_ranked <- kpis_cost |>
arrange(cost_zscore)
top10 <- kpis_ranked |>
slice_head(n = 10) |>
mutate(highlight = "Top 10 (lowest cost)")
bot10 <- kpis_ranked |>
slice_tail(n = 10) |>
mutate(highlight = "Bottom 10 (highest cost)")
plot_ranked <- function(df, title) {
ggplot(df, aes(x = avg_cost, y = reorder(name, -avg_cost))) +
geom_segment(aes(xend = mean(kpis$avg_cost), yend = reorder(name, -avg_cost)),
colour = "#66B5E8", linewidth = 1.5, linetype = "dashed") +
geom_point(aes(colour = highlight), size = 4, show.legend = FALSE) +
geom_vline(xintercept = mean(kpis$avg_cost),
linetype = "dashed", colour = "#6B7280", linewidth = 0.5) +
scale_x_continuous(labels = scales::label_number(prefix = "CHF ", big.mark = ",")) +
scale_colour_manual(values = c(
"Top 10 (lowest cost)" = "#00A9CE",
"Bottom 10 (highest cost)" = "#FF6600"
)) +
labs(title = title, x = "Avg. Repair Cost", y = NULL) +
theme_allianz(grid = "y")
}
plot_ranked(top10, "Top 10 — Lowest Cost Partners")
plot_ranked(bot10, "Bottom 10 — Highest Cost Partners")
```
```{r top10-table}
#| echo: false
kpis |>
arrange(avg_cost) |>
slice_head(n = 15) |>
select(name, type, region, n_claims, avg_cost, avg_duration, avg_csat, reopen_rate) |>
gt() |>
fmt_number(columns = avg_cost, decimals = 0, sep_mark = ",") |>
fmt_number(columns = avg_duration, decimals = 1) |>
fmt_number(columns = avg_csat, decimals = 1) |>
fmt_percent(columns = reopen_rate, decimals = 1) |>
data_color(columns = avg_cost,
palette = c("#003781", "#66B5E8"),
domain = range(kpis$avg_cost, na.rm = TRUE)) |>
data_color(columns = avg_csat,
palette = c("#FF6600", "#00A9CE"),
domain = c(6, 10)) |>
tab_header(title = "Top 15 Partners — Lowest Avg. Cost",
subtitle = "Inline bars show relative cost and duration") |>
tab_style(
style = cell_fill(color = "#003781"),
locations = cells_column_labels()
) |>
tab_style(
style = list(cell_text(color = "white", weight = "bold")),
locations = cells_column_labels()
) |>
tab_options(table.font.size = px(12))
```
## Steering Rate by Claim Type and Region
```{r steering-overview}
#| fig-height: 4
steering_summary <- claims |>
group_by(claim_type, region) |>
summarise(pct_steered = mean(steering_flag), .groups = "drop")
p_steer <- ggplot(steering_summary,
aes(x = region, y = pct_steered, fill = claim_type)) +
geom_col(position = "dodge", width = 0.75) +
scale_fill_manual(
values = c(glass="#003781", body="#0066CC", engine="#00A9CE", total_loss="#FF6600"),
name = "Claim Type"
) +
scale_y_continuous(labels = scales::percent, limits = c(0, 1)) +
labs(title = "Steering Rate by Claim Type × Region",
subtitle = "Proportion of claims routed to network partner",
x = NULL, y = "Steering Rate") +
theme_allianz() +
theme(axis.text.x = element_text(angle = 30, hjust = 1))
ggplotly(p_steer) |>
layout(legend = list(orientation = "h", y = -0.2))
```
