A comprehensive dataset of customer behavior in Latin American Fintech: 12-month transactional and demographic data for churn analysis

Abstract

This article introduces COFINFAD (Colombian Fintech Financial Analytics Dataset), a single-company 12-month dataset containing comprehensive behavioral information from 48,723 customers (representing the complete active customer base) of a Colombian fintech company. The dataset spans January 4 to December 29, 2023, capturing 3,159,157 transactions alongside demographic profiles, product usage patterns, customer satisfaction metrics, and digital app engagement behaviors. Data collection involved API integration with the company’s CRM system, transaction databases, and customer surveys (6,965 respondents, 14.3% response rate), while prioritizing customer privacy through multi-layered anonymization (pseudonymization, generalization, k-anonymity verification). The dataset comprises 57 variables including transaction frequency and value, product portfolio composition (savings, credit cards, loans, investments, insurance), customer satisfaction scores, Net Promoter Score (NPS), app login frequency, feature usage patterns, and support interaction metrics—all measured in Colombian Pesos (COP). COFINFAD is distinguished by its integrated multi-dimensional design, combining transactional, demographic, satisfaction, and digital engagement data unavailable in existing publicly available fintech datasets. The dataset is openly accessible via Mendeley Data (DOI: 10.17632/mhb4zn3258.1) under CC BY 4.0 license, adhering to FAIR (Findable, Accessible, Interoperable, Reusable) data principles.

Specifications Table

Subject	Computer Sciences
Specific subject area	Customer behavior analysis and churn prediction in fintech; digital financial inclusion in emerging markets
Type of data	Tables, CSV files
Data collection	API integration with CRM system and transaction databases (Python 3.9, pandas 1.3.5), customer surveys (Qualtrics), mobile analytics SDK (Firebase Analytics 9.1)
Data source location	Country: Colombia City/State: Medellín (Antioquia), Bogotá (Cundinamarca), Cali (Valle del Cauca), Barranquilla (Atlántico)
Data accessibility	Repository name: COFINFAD: Colombian Fintech Financial Analytics Data identification number: 10.17632/mhb4zn3258.1 Direct URL to data: https://data.mendeley.com/datasets/mhb4zn3258/1 Instructions for accessing these data: The data are accessible directly via the repository URL provided below without any special access requirements
Related research article	None

1. Value of the Data

• •Multi-dimensional behavioral and attitudinal integration: COFINFAD uniquely combines granular transactional data (3.1M transactions), comprehensive demographics, customer satisfaction metrics (satisfaction_score, NPS, feedback_sentiment), and digital engagement behaviors (app_logins_frequency, feature_usage_diversity).
• •Unique Granularity: Unlike existing datasets like LendingClub or MBD, no publicly available dataset integrates observed behaviors with self-reported satisfaction at this granularity.
• •Research Potential: Researchers can develop churn prediction models incorporating attitudinal variables, analyze satisfaction-behavior relationships, study multi-product adoption dynamics, and test customer engagement hypotheses in digital banking.
• •Emerging market financial services representation: This dataset addresses critical data scarcity for Latin American financial inclusion research. Colombia represents the third-largest fintech ecosystem in a region experiencing 112% growth (2018-2021). The data provides granular income-level information and digital engagement behaviors from a developing economy context underrepresented in existing financial datasets.
• •Complete population study: The 48,723 customers represent the complete active customer base for 2023, making this a population study eliminating sampling bias. Researchers access 57 observed behavioral features capturing actual customer actions rather than self-reported intentions.
• •Single-company methodological coherence: The focused single-company design provides uniform policies, UI/UX design, customer communication protocols, and regulatory environment across all observations, eliminating confounding variables from heterogeneous fintech environments.

2. Background

This dataset was compiled to provide empirical data for researchers analyzing fintech customer behavior in emerging markets, addressing the scarcity of real-world behavioral data from Latin America. Through collaboration with a Colombian fintech company, we curated and de-identified this comprehensive dataset for public access.

Colombia represents the third-largest fintech ecosystem in Latin America, with 394 companies operating as of Q1 2024. The single-company design provides a “revelatory case” with rare access to granular behavioral data typically inaccessible to researchers, and a “longitudinal case” enabling 12-month temporal analysis. This focused approach ensures contextual coherence eliminating confounding variables from heterogeneous fintech environments, behavioral granularity capturing 57 observed features of actual customer actions, and longitudinal depth tracking behavioral evolution.

The dataset follows the precedent of focused single-organization datasets providing unique depth, such as MIMIC-III healthcare data [[1], [2], [3], [4]]. Related datasets exist for banking and financial behavior, but they typically lack one or more of the core dimensions needed for churn research in fintech: (i) true transaction sequences over time, (ii) multi-product portfolio indicators, (iii) behavioral app analytics, and (iv) attitudinal measures such as satisfaction and NPS [[5], [6], [7], [8], [9], [10]]. For example, LendingClub-style public loan datasets are primarily focused on credit risk and applications rather than customer retention dynamics, while synthetic transaction datasets such as PaySim reproduce statistical properties but do not capture real customer-product relationships [[11], [12], [13], [14], [15], [16], [17]]. Recent banking datasets based on event sequences provide valuable multimodal signals, yet they are often limited in geographic scope and may not include satisfaction or support interactions. These limitations motivate COFINFAD’s integrated design for a Latin American fintech setting (Fig. 1, Fig. 2, Fig. 3, Fig. 4) [[18], [19]].

Fig. 1.

Churn Rate by Product Usage. The figure displays the relationship between the number of active products held by a customer and the observed churn rate.

Fig. 2.

Geographical Distribution of Customers. The figure displays customer distribution across Colombia, showing concentrations in Bogotá (CO-CUN), Medellín (CO-ANT), and Cali (CO-VAC).

Fig. 3.

Distribution of Customer Tenure. The figure shows customer tenure distributed across durations, with a concentration toward longer tenures (4.7-12.0 months).

Fig. 4.

Correlation Matrix of Key Variables. The figure displays correlations between key variables, including customer lifetime value, transaction frequency, app login frequency, and satisfaction scores.

3. Data Description

The dataset contains information on 48,723 customers of a Colombian fintech company, collected over a 12-month period from January 4, 2023, to December 29, 2023. The 48,723 customers represent the complete active customer base for calendar year 2023, making this a comprehensive population study rather than a sample.

The dataset comprises 57 variables categorized into demographics, product usage, transaction records, customer feedback, digital engagement, and derived metrics.

Data are organized in two CSV files with relational structure:

• 1.customer_data.csv: Contains customer-level information including demographics, product portfolio, satisfaction metrics, and aggregated behavioral features (one row per customer, 48,723 rows).
• 2.transactions_data.csv: Contains transaction-level data for all customers over the 12-month period (one row per transaction, 3,159,157 rows).

The two files are linked via ‘customer_id’ foreign key with 1:N cardinality. The primary key in ‘customer_data.csv’ is ‘customer_id’; ‘transactions_data.csv’ does not include an explicit transaction identifier; each row corresponds to one transaction event. Table 1, Table A1 provides a complete dictionary for the dataset variables.

Table 1.

Dataset variables with valid ranges, units, and examples.

No.	Variable Name	Definition	Data Type	Valid Range/Values
1	customer_id	Unique identifier	Integer	1-48723
2	age	Customer’s age	Integer	18-65
5	income_bracket	Income category	String	Low, Medium, High, Very High
17	active_products	Number of active products	Integer	0-5
18	app_logins_frequency	App logins per month	Integer	0-100
25	tx_count	Total transactions	Integer	10-77465
33	satisfaction_score	Overall satisfaction	Integer	1-6
34	nps_score	Net Promoter Score	Integer	0-10
53	churn_probability	Predicted churn probability	Float	0.0-1.0
56	amount	Transaction amount	Float	>0 (COP)

Note: Selected key variables shown for brevity; full dictionary available in repository

Missing Data: In ‘customer_data.csv’ (N=48,723), only ‘complaint_topics’ (49.97%), ‘credit_utilization_ratio’ (37.48%), and ‘feature_requests’ (33.07%) contain missing values when the corresponding product or interaction is not applicable; all transaction-level variables are complete.

Variable	Mean	Median	Std Dev	Min	Max
Age (years)	44.5	45.0	12.3	18.0	65.0
Monthly Transactions	64.8	18.0	611.9	1.0	77,465.0
Avg Transaction Value (COP)	3,564,851	1,763,107	3,959,442	318,215	51,670,302
Customer Tenure (months)	11.4	11.6	0.7	4.7	12.0
App Login Freq (per month)	22.4	22.0	11.8	0.0	100.0
Satisfaction Score (1-6)	4.2	4.0	0.6	1.0	6.0
NPS Score (0-10)	7.8	8.0	1.9	0.0	10.0

4. Experimental Design, Materials and Methods

Data were gathered from three integrated sources:

• 1.API Integration: Python scripts (Python 3.9) with the requests library performed daily automated API requests to the fintech company’s CRM system (PostgreSQL 13.2) and transaction databases (MongoDB 5.0). These requests retrieved customer profiles, transaction records, and product usage metrics. ETL (Extract, Transform, Load) pipeline utilized pandas 1.3.5 for data transformation and validation.
• 2.Customer Surveys: Quarterly surveys were administered through the company’s mobile application (Qualtrics platform). Customers received in-app prompts to complete 5-question surveys evaluating satisfaction levels (1-6 scale), likelihood of recommendation (NPS score, 0-10 scale), and providing qualitative feedback. Survey distribution occurred in February, May, August, and November 2023.
• 3.App Usage Tracking: Real-time monitoring of app usage was enabled through integration of Firebase Analytics SDK (version 9.1) into the mobile application. The SDK tracked feature utilization, login frequency, session duration, and in-app navigation patterns.

Comprehensive validation checks ensured data quality and integrity. Completeness validation verified all mandatory fields; range validation checked numeric variables against expected ranges (e.g., age 18-65); and consistency validation ensured cross-field logic. 99.7% of records passed all validation checks. Records failing validation (n=1,461) were excluded. Outlier detection flagged transaction amounts exceeding 3 standard deviations; erroneous entries were removed (n=347). Exact duplicate records were removed.

To ensure customer privacy while preserving analytical utility, a multi-layered anonymization approach was implemented following GDPR principles and Colombian data protection regulations (Law 1581 of 2012):

• 1.Direct Identifier Removal and Pseudonymization: All direct identifiers (names, IDs, emails) were removed. Customer records were assigned cryptographic pseudonyms using SHA-256 hashing.
• 2.Quasi-Identifier Generalization: Age was binned into 5-year groups; locations aggregated to city/department level; income categorized into brackets; timestamps converted to date-only. Rare attribute values (<1%) were suppressed.
• 3.K-Anonymity Verification: Analysis using ARX Data Anonymization Tool 3.9.0 confirmed that each combination of quasi-identifiers appears in at least 10 records (k=10).

The dataset adheres to FAIR principles. It is Findable via a persistent DOI (10.17632/mhb4zn3258.1) and rich metadata. It is Accessible via open HTTPS protocol on Mendeley Data. It is Interoperable using standard CSV format (RFC 4180) and ISO standards for currency and dates. It is Reusable under the CC BY 4.0 license with complete provenance documentation.

We verified the released CSV artifacts against the data dictionary. customer_data.csv contains 54 customer-level columns and transactions_data.csv contains 4 transaction-level columns. customer_id is the join key. All variable names match the CSV headers exactly.

Limitations

The dataset represents a single Colombian fintech company observed over 12 months (2023 only), limiting temporal scope and generalizability. The 48,723 customers represent the complete active customer base for 2023. Urban areas are over-represented (75% from Bogotá, Medellín, Cali), potentially overlooking rural regions. Income data are self-reported categorical brackets with 12% non-response rate. The dataset inherently excludes individuals without smartphone access. Satisfaction-related variables are subject to non-response bias (14.3% survey response rate). The single-year timeframe may not capture long-term trends or multi-year economic cycles. All monetary values are in Colombian Pesos; exchange rate fluctuations may affect international comparisons.

Ethics Statement

This research was conducted in full compliance with Colombian data protection laws (Law 1581 of 2012) and international ethical guidelines. The fintech company provided formal authorization to use anonymized customer data for research purposes. The dataset was rigorously anonymized using a multi-layered approach: (a) pseudonymization via SHA-256 hashing, (b) quasi-identifier generalization, and (c) k-anonymity verification (k≥10) validated with ARX Data Anonymization Tool 3.9.0.

The research protocol received approval from the fintech company’s internal ethics board (approval #ETH-2023-047, dated January 15, 2023) and an independent ethical review committee (approval #IND-ETH-2023-112, dated February 2, 2023). For the customer survey component, participants provided informed consent via in-app consent screen explaining survey purpose, voluntary nature, data use, and privacy protections. Raw unanonymized data are stored encrypted on secure servers within Colombia, with access restricted to authorized research staff only. The authors affirm compliance with ethical publication requirements in Data in Brief.

CRediT Author Statement

Luis Eduardo Muñoz Guerrero: Conceptualization, Data curation, Formal analysis, Writing - original draft; Yony Fernando Ceballos: Methodology, Software, Validation, Formal analysis, Writing - review & editing; Luis David Trejos Rojas: Project administration, Resources, Supervision, Writing - review & editing, Funding acquisition.

Appendix

Table A1.

The following table provides the complete data dictionary for the COFINFAD dataset. The variables are divided between the customer-level aggregations (`customer_data.csv`, Variables 1–54) and the granular transaction records (‘transactions_data.csv’, Variables 55–57).

No.	Variable Name	Definition	Data Type
1	customer_id	Unique identifier for each customer	Integer
2	age	Customer's age in years	Integer
3	gender	Customer's gender	String
4	location	Customer's city and state	String
5	income_bracket	Customer's income category	String
6	occupation	Customer's job or profession	String
7	education_level	Highest level of education attained by the customer	String
8	marital_status	Customer's marital status	String
9	household_size	Number of people in the customer's household	Integer
10	acquisition_channel	How the customer was acquired (e.g., organic, referral)	String
11	customer_segment	Category assigned to the customer based on their behavior	String
12	savings_account	Whether the customer has a savings account (True/False)	Boolean
13	credit_card	Whether the customer has a credit card (True/False)	Boolean
14	personal_loan	Whether the customer has a personal loan (True/False)	Boolean
15	investment_account	Whether the customer has an investment account (True/False)	Boolean
16	insurance_product	Whether the customer has an insurance product (True/False)	Boolean
17	active_products	Number of active financial products the customer has	Integer
18	app_logins_frequency	Number of times the customer logs into the app per month	Integer
19	feature_usage_diversity	Number of unique features used by the customer in the app	Integer
20	bill_payment_user	Whether the customer uses bill payment feature (True/False)	Boolean
21	auto_savings_enabled	Whether the customer has enabled auto-savings feature (True/False)	Boolean
22	credit_utilization_ratio	Ratio of credit used to credit available	Float
23	international_transactions	Number of international transactions made by the customer	Integer
24	failed_transactions	Number of failed transactions for the customer	Integer
25	tx_count	Total number of transactions made by the customer	Integer
26	avg_tx_value	Average value of customer's transactions	Float
27	total_tx_volume	Total value of all transactions made by the customer	Float
28	first_tx	Date of the customer's first transaction	Date
29	last_tx	Date of the customer's most recent transaction	Date
30	base_satisfaction	Base satisfaction score for the customer	Float
31	tx_satisfaction	Satisfaction score based on transaction history	Float
32	product_satisfaction	Satisfaction score based on product usage	Float
33	satisfaction_score	Overall customer satisfaction score	Integer
34	nps_score	Net Promoter Score for the customer	Integer
35	last_survey_date	Date when the customer last completed a survey	Date
36	support_tickets_count	Number of support tickets opened by the customer	Integer
37	resolved_tickets_ratio	Ratio of resolved support tickets to total tickets	Float
38	app_store_rating	Customer's rating of the app in the app store	Float
39	feedback_sentiment	Sentiment analysis of customer's feedback	String
40	feature_requests	Features requested by the customer	String
41	complaint_topics	Main topics of customer's complaints	String
42	clv_segment	Customer Lifetime Value segment	String
43	monthly_transaction_count	Average number of transactions per month	Float
44	average_transaction_value	Average value of customer's transactions	Float
45	total_transaction_volume	Total value of all transactions made by the customer	Float
46	transaction_frequency	Number of transactions per day	Float
47	last_transaction_date	Date of the customer's most recent transaction	Date
48	preferred_transaction_type	Most frequent type of transaction for the customer	String
49	first_transaction_date	Date of the customer's first transaction	Date
50	weekend_transaction_ratio	Ratio of transactions made on weekends	Float
51	avg_daily_transactions	Average number of transactions per day	Float
52	customer_tenure	Length of time as a customer in months	Float
53	churn_probability	Predicted probability of customer churn	Float
54	customer_lifetime_value	Estimated total value of the customer to the business	Float
55	date	Date of a specific transaction	Date
56	amount	Amount of a specific transaction	Float
57	type	Type of a specific transaction	String

Data Availability

• Mendeley DataCOFINFAD: Colombian Fintech Financial Analytics Dataset (Original data).