Part 1: Demand Estimation

Objective

The primary objective is to determine if there is sufficient demand from Wise existing user base to justify the investment in launching a new currency route from Mexican Pesos (MXN) to US Dollars (USD). This analysis will provide a defensible, data-driven answer to the question: “Is there a core group of Wise users who would immediately and materially benefit from this route?”

Primary Methodology: Proxy Route Analysis

To provide the most effective and defensible demand estimate, I recommend one focused methodology: analyzing the behavior of existing Wise users whose actions serve as a strong proxy for demand. This approach is the most reliable because it is based on observed, transactional behavior rather than speculative intent.

Our single primary demand metric will be Proxied Transfer Volume (PTV), calculated from the total volume and unique user count of this core segment over the last 12 months. This establishes a conservative, data-backed “floor” for the immediate market opportunity.

Methodology Steps:

Identify the Core Segment: The analysis will query Wise’s transfer data to isolate a specific cohort: Wise users with a registered country of Mexico who are currently sending any currency (e.g., EUR, GBP, CAD) to a USD-denominated bank account.
Quantify the Opportunity: From this segment, we will calculate the total volume sent to USD accounts and the number of unique users performing these transfers over the past year. This will be our PTV.

Justification for Chosen Methodology

I have chosen Proxy Route Analysis as the single primary methodology over other potential options for three key reasons:

It Measures Behavior, Not Intent: Unlike analyzing search logs or running surveys, this method analyzes actual, completed transactions. A user who navigates the friction of a multi-step workaround (e.g., converting MXN externally to fund their Wise account) has a proven, painful need. This is a far stronger signal of demand than a simple search query or a “yes” on a survey.
It is Immediately Actionable: This analysis can be conducted immediately using existing historical data. It does not require waiting for an experiment to run (like a “Notify Me” test). This aligns with the 80/20 principle of generating the maximum impact with the most efficient effort.
It Provides a Conservative Financial Baseline: The calculated PTV provides a concrete, minimum-viable financial figure. We can confidently state to the product and finance teams that “at a minimum, we can expect to capture X volume from Y users who are already on our platform,” which is a powerful starting point for any business case.

Supplementary Analysis & Validation Steps

While the Proxy Route Analysis provides the core, high-confidence estimate, a complete business case should also understand the long-term growth potential. The following steps are proposed as supplements and follow-ups.

Total Addressable Market (TAM) Sizing

Method: This analysis uses external data from sources like the World Bank and industry reports to estimate the total annual remittance volume from Mexico to the US, focusing specifically on the digital portion of that market.
Purpose: Its purpose is not to provide an immediate demand signal, but to frame the long-term growth opportunity and the potential for new user acquisition.
Trade-offs: This methodology is supplementary, not primary, due to significant trade-offs. The data is often imprecise, aggregating diverse user segments into a single number. It is also a lagging indicator, often published 6-12 months late. Therefore, it is less reliable for making immediate product decisions than our internal behavioral data.

Follow-up Validation: Internal Search Log Analysis

Method: If the PTV is promising, we would validate this finding by analyzing search queries. The primary signal to query is a user searching for “MXN” or “Mexican Peso” within the currency selection dropdown. This indicates a direct intent to initiate a transfer from Mexico.
Rationale: This validates our findings by showing a high level of intent that complements the behavioral data from the primary analysis. It confirms that users are actively looking for the service on our platform.

Final Validation: “Notify Me” Waitlist Experiment

Method: As a final validation step before committing significant engineering resources, we could run a low-risk A/B test for a small cohort of users in Mexico. We would add “MXN” to the “send from” currency dropdown, which, when selected, would trigger a modal to sign up for a waitlist.
Rationale: This directly measures user interest and provides a list of engaged customers to contact upon launch.

By sequencing the analysis this way—starting with the most robust behavioral data (PTV) and layering on market context and intent-based validation—we can build a comprehensive and highly defensible case in an efficient, iterative manner.

Part 2: Launch Evaluation

TL;DR:

The new route launch shows a stable overall conversion (~26% created-to-transferred) over Jan–Mar 2024, but internal funnel dynamics shifted in early February. Created-to-funded dropped while funded-to-transferred improved, offsetting each other. Around Feb 1st, transfer volume—especially from Europe—spiked sharply, coinciding with these funnel changes. Regional analysis suggests two main factors: (1) Europe’s growing share (with lower created-to-funded but higher funded-to-transferred rates) altered the overall funnel mix; and (2) Europe’s created-to-funded rate itself declined notably (~38% → ~30%) in February. These findings point to regional performance shifts, possibly tied to marketing activity or data attribution issues, warranting further investigation.

Investigation

The objective is to evaluate the launch of the new route with the wise_funnel_events_regional data. The table provided contains the transfer events triggered by each user, allowing us to evaluate based on: volume of transactions, and funnel conversion.

The overall funnel shows that from 43.1k (100%) transfers created, only 19.4k (45%) were funded, and 11.1k (26%) were transferred. It is not trivial to evaluate if the funnel is performing as expected without a benchmark of other currency routes or more granular data on why the drops happened. However, it’s important to monitor the launch performance over time during the two months of data from 2024-01-01 to 2024-03-01.

If we look at the end-to-end conversion, from transfer created-to-transferred, it maintained stable through the whole period at around 26%. However, if we breakdown the funnel into two steps: (1) created-to-funded and (2) funded-to-transferred, there is a sudden change in the first days of February. Created-to-funded had a drop and stayed in a lower baseline for the rest of the period; while funded-to-transferred had a bump and stayed in a higher baseline, compensating the former drop to keep the created-to-transferred stable through the whole period.

Looking at the volume of transfers in the period, we see a fast ramp-up until Jan 10th where the three event types maintained stable with a slight upwards trend. However, the number of transfers created had a sudden increase in Feb 1st and maintained stable in a higher baseline. This behavior matches the dates in which there were the funnel conversion anomalies, likely being associated with the disturbance.

Breaking down the volume of transfer events by region, there is a clear sudden increase in events from Europe and a sudden decrease in Other and NorthAm (to a smaller degree). This raises the questions: (1) Are these movements expected?; (2) Could it be the effect from marketing campaigns; (3) Could it be a misattribution of regions?

That’s definitely worth investigating, but we can go further. Even if the volume/mix of regions is changing, it doesn’t necessarily cause changes in the funnel conversion. For example, if the conversion rates were similar between regions, the funnel conversion would have remained stable through the whole period even if the mix changed. This raises two hypothesis for the funnel behavior: (1) the mix of regions change caused the conversion rates to change because of unequal performance across regions; (2) the conversion rates from the regions changed in the beginning of Feb.

Breaking down the conversion rates for each step of the funnel by region, we see that the two hypothesis were true. (1) Users from Europe have a worse created-to-funded rate and a better funded-to-transferred rate than the other regions; (2) The created-to-funded conversion rate in Europe had the biggest change in February going from ~38% to ~30%.

Complete Code Investigation

Setup and Data Loading

Loading the libraries and the dataset.

library(tidyverse)
library(plotly)

df <- read_csv("wise_funnel_events_regional - wise_funnel_events_regional.csv")

Data Inspection

Inspecting the dataframe to check the possible values in each column.

73,440 rows
0 NULL values
event_name: Transfer Created, Transfer Funded, Transfer Transferred
region: Europe, NorthAm, Other
platform: Android, iOS, Web
dt: date from 2024-01-01 to 2024-03-01

head(df)

## # A tibble: 6 × 6
##   event_name           dt         user_id region  platform experience
##   <chr>                <date>       <dbl> <chr>   <chr>    <chr>     
## 1 Transfer Funded      2024-02-15 1513280 NorthAm iOS      Existing  
## 2 Transfer Funded      2024-02-29 1270052 NorthAm iOS      Existing  
## 3 Transfer Transferred 2024-02-29 1270052 NorthAm iOS      Existing  
## 4 Transfer Funded      2024-02-25 1399835 NorthAm iOS      Existing  
## 5 Transfer Transferred 2024-02-25 1399835 NorthAm iOS      Existing  
## 6 Transfer Funded      2024-02-01 1972860 NorthAm iOS      Existing

dim(df)

## [1] 73440     6

sum(is.na(df))

## [1] 0

df %>% 
  group_by(event_name) %>% 
  summarise(event_count = n(),
            unique_users = n_distinct(user_id))

## # A tibble: 3 × 3
##   event_name           event_count unique_users
##   <chr>                      <int>        <int>
## 1 Transfer Created           43070        40223
## 2 Transfer Funded            19290        18195
## 3 Transfer Transferred       11080        10557

funnel_order <- c("Transfer Created", "Transfer Funded", "Transfer Transferred")

df %>% 
  group_by(region) %>% 
  summarise(event_count = n(),
            unique_users = n_distinct(user_id))

## # A tibble: 3 × 3
##   region  event_count unique_users
##   <chr>         <int>        <int>
## 1 Europe        25665        15695
## 2 NorthAm       19468         9393
## 3 Other         28307        15704

df %>% 
  group_by(platform) %>% 
  summarise(event_count = n(),
            unique_users = n_distinct(user_id))

## # A tibble: 3 × 3
##   platform event_count unique_users
##   <chr>          <int>        <int>
## 1 Android        27593        15955
## 2 Web            20243        11681
## 3 iOS            25604        13172

df %>% 
  group_by(experience) %>% 
  summarise(event_count = n(),
            unique_users = n_distinct(user_id))

## # A tibble: 2 × 3
##   experience event_count unique_users
##   <chr>            <int>        <int>
## 1 Existing         33628        15808
## 2 New              39812        25234

df %>% 
  summarise(first_dt = min(dt), 
            last_dt = max(dt),
            unique_users = n_distinct(user_id))

## # A tibble: 1 × 3
##   first_dt   last_dt    unique_users
##   <date>     <date>            <int>
## 1 2024-01-01 2024-03-01        40223

Funnel Overall

overall_funnel_data <- df %>% 
  group_by(event_name) %>% 
  summarise(event_count = n())
  
overall_funnel_plot <- plot_ly(
  overall_funnel_data,
  x = ~event_count,
  y = ~event_name,
  type = 'funnel',
  textposition = 'inside',
  textinfo = 'value+percent initial+percent previous'
) %>%
  layout(
    title = "Overall Conversion Funnel by Event Count",
    yaxis = list(title = "Funnel Step"),
    xaxis = list(title = "Number of Unique Users")
  )

overall_funnel_plot

overall_funnel_data <- df %>% 
  group_by(event_name) %>% 
  summarise(event_count = n_distinct(user_id))
  
overall_funnel_plot <- plot_ly(
  overall_funnel_data,
  x = ~event_count,
  y = ~event_name,
  type = 'funnel',
  textposition = 'inside',
  textinfo = 'value+percent initial+percent previous'
) %>%
  layout(
    title = "Overall Conversion Funnel by Unique Users",
    yaxis = list(title = "Funnel Step"),
    xaxis = list(title = "Number of Unique Users")
  )

overall_funnel_plot

Funnel Over Time

daily_funnel <- df %>%
  mutate(event_name = str_to_lower(str_remove(event_name, "Transfer "))) %>% 
  group_by(dt, event_name) %>%
  summarise(distinct_users = n_distinct(user_id), .groups = 'drop') %>%
  pivot_wider(
    names_from = event_name,
    values_from = distinct_users,
    values_fill = 0 # Fills in 0 if an event type didn't occur on a given day
  ) %>%
  # Calculate conversion rates for each step
  mutate(
    created_to_funded_rate = funded / created,
    funded_to_transferred_rate = transferred / funded,
    overall_completion_rate = transferred / created
  ) %>%
  # Reshape data for plotting
  pivot_longer(
    cols = c(created_to_funded_rate, funded_to_transferred_rate, overall_completion_rate),
    names_to = "metric",
    values_to = "rate"
  )

# Create the time series plot
p_time_series <- ggplot(daily_funnel, aes(x = dt, y = rate, color = metric, group = metric)) +
  geom_line() +
  labs(
    title = "Daily Funnel Conversion Rates Over Time",
    x = "Date of Transfer Event",
    y = "Conversion Rate",
    color = "Conversion Metric"
  ) +
  scale_y_continuous(labels = scales::percent, limits = c(0, 1)) +
  scale_x_date(date_breaks = "1 week", date_labels = "%b %d") +
  theme_minimal() 

ggplotly(p_time_series)

p_time_series <- df %>% 
  group_by(dt, event_name) %>% 
  summarise(event_count = n(),
            unique_users = n_distinct(user_id)) %>% 
  ggplot(aes(x = dt, y = unique_users, color = event_name)) + 
    geom_line() +
    labs(
      title = "Daily Unique User Over Time",
      x = "Date of Transfer Event",
      y = "Unique Users",
      color = "Conversion Metric"
    ) +
    scale_x_date(date_breaks = "1 week", date_labels = "%b %d") +
    theme_minimal() 

ggplotly(p_time_series)

plot_daily_funnel_by_group <- function(data, grouping_var) {
  
  # Convert the string grouping_var to a symbol for use in dplyr
  grouping_sym <- sym(grouping_var)
  
  daily_funnel_by_group <- df %>%
    mutate(event_name = str_to_lower(str_remove(event_name, "Transfer "))) %>% 
    group_by(dt, event_name, !!grouping_sym) %>%
    summarise(distinct_users = n_distinct(user_id), .groups = 'drop') %>%
    pivot_wider(
      names_from = event_name,
      values_from = distinct_users,
      values_fill = 0 # Fills in 0 if an event type didn't occur on a given day
    ) %>%
    # Calculate conversion rates for each step
    mutate(
      created_to_funded_rate = funded / created,
      funded_to_transferred_rate = transferred / funded,
      overall_completion_rate = transferred / created
    ) %>%
    # Reshape data for plotting
    pivot_longer(
      cols = c(created_to_funded_rate, funded_to_transferred_rate, overall_completion_rate),
      names_to = "metric",
      values_to = "rate"
    )

  plot_title <- paste("Daily Funnel Conversion Rates Over Time by", tools::toTitleCase(grouping_var))
  
  p <- ggplot(daily_funnel_by_group, aes(x = dt, y = rate, color = !!grouping_sym, group = !!grouping_sym)) +
    geom_line() +
    labs(
      title = plot_title,
      x = "Date of Transfer Event",
      y = "Conversion Rate",
      color = tools::toTitleCase(grouping_var) # Label color by the grouping variable
    ) +
    scale_y_continuous(labels = scales::percent, limits = c(0, 1)) +
    scale_x_date(date_breaks = "2 weeks", date_labels = "%b %d") +
    theme_minimal() +
    facet_wrap(~ metric, ncol = 1) # Facet by the conversion metric now
  
  return(ggplotly(p))
}

plot_daily_funnel_by_group(user_summary, "region")

plot_daily_funnel_by_group(user_summary, "experience")

plot_daily_funnel_by_group(user_summary, "platform")

User Volume Over Time

plot_daily_counts_by_group <- function(data, grouping_var) {
  
  # Convert the string grouping_var to a symbol for use in dplyr
  grouping_sym <- sym(grouping_var)
  
  daily_counts_by_group <- df %>%
    mutate(event_name = str_to_lower(str_remove(event_name, "Transfer "))) %>% 
    group_by(dt, event_name, !!grouping_sym) %>%
    summarise(distinct_users = n_distinct(user_id), .groups = 'drop')
  
  plot_title <- paste("Daily Unique Users Over Time by", tools::toTitleCase(grouping_var))
  
  p <- ggplot(daily_counts_by_group, aes(x = dt, y = distinct_users, color = !!grouping_sym, group = !!grouping_sym)) +
    geom_line() +
    labs(
      title = plot_title,
      x = "Date of Transfer Event",
      y = "Number of Unique Users",
      color = tools::toTitleCase(grouping_var)
    ) +
    scale_x_date(date_breaks = "2 weeks", date_labels = "%b %d") +
    theme_minimal() +
    # Facet by funnel step, allow y-axis to scale independently for clarity
    facet_wrap(~ event_name, ncol = 1, scales = "free_y") 
  
  return(ggplotly(p))
}

plot_daily_counts_by_group(user_summary, "region")

plot_daily_counts_by_group(user_summary, "experience")

plot_daily_counts_by_group(user_summary, "platform")

User Mix Over Time

plot_daily_counts_by_group <- function(data, grouping_var) {
  
  # Convert the string grouping_var to a symbol for use in dplyr
  grouping_sym <- sym(grouping_var)
  
  daily_counts_by_group <- df %>%
    mutate(event_name = str_to_lower(str_remove(event_name, "Transfer "))) %>% 
    group_by(dt, event_name, !!grouping_sym) %>%
    summarise(distinct_users = n_distinct(user_id), .groups = 'drop') %>% 
    group_by(dt, event_name) %>% 
    mutate(percentage = distinct_users / sum(distinct_users)) %>% 
    ungroup
  
  plot_title <- paste("Share of Unique Users Over Time by", tools::toTitleCase(grouping_var))
  
  p <- ggplot(daily_counts_by_group, aes(x = dt, y = percentage, fill = !!grouping_sym)) +
    geom_area(alpha = 0.5) + 
    labs(
      title = plot_title,
      x = "Date of Transfer Event",
      y = "Share of Unique Users",
      color = tools::toTitleCase(grouping_var)
    ) +
    scale_y_continuous(labels = scales::percent, limits = c(0, 1)) +
    scale_x_date(date_breaks = "2 weeks", date_labels = "%b %d") +
    theme_minimal() +
    # Facet by funnel step, allow y-axis to scale independently for clarity
    facet_wrap(~ event_name, ncol = 1, scales = "free_y") 
  
  return(ggplotly(p))
}

plot_daily_counts_by_group(user_summary, "region")

plot_daily_counts_by_group(user_summary, "experience")

plot_daily_counts_by_group(user_summary, "platform")

Wise - Product Analytics Case Study

Eduardo Yada

2025-10-14