Daniel Juan de Dios Aguiñaga Gómez

Problem:
A digital product needs to evaluate whether adding new traffic sources improves conversion-funnel performance and, in particular, the key payment event. The goal is to make an informed decision on whether to scale these sources without compromising profitability or user experience.

What do I do in this project?
– Build and analyze the event funnel
– Validate the experimental design using A/A
– Run A/B across multiple funnel events
– Rigorous statistical evaluation (including multiple comparisons)
– Interpret findings and final recommendation

Value added:
I don’t just report metrics: I translate statistical results into actionable decisions (when to scale an initiative, when to stop it, and what additional changes could create more impact). This reduces the risk of investing resources without proven benefit and strengthens evidence-based decision-making.

Key metrics:

• Evaluation of over 4,900 users distributed in three experimental groups.
• Behavior analysis across 5 key events conversion funnel events.
• Observed differences were not statistically significant at α = 0.05 nor under the multiple-testing adjusted level.

Problem:
A digital service needs to better understand user behavior to optimize marketing campaigns, improve retention, and estimate lifetime value (LTV) by cohort and acquisition channel.

What do I do in this project?
– Clean and prepare data so it is consistent and ready for analysis.
– Exploratory analysis (EDA) to understand usage, retention, and cohort behavior patterns.
– Build metrics and simple models to estimate LTV and compare acquisition channels.
– Synthesize findings and write recommendations in business-friendly language.

Value added:
I don’t just calculate metrics; I connect them to concrete decisions: which channels appear more profitable, how cohorts evolve over time, and what type of investment may yield better returns. The result is a set of insights that help prioritize marketing actions based on data.

Key metrics:

• Analysis of 207,051 users with an approximate retention of 22.5% at 30 days.
• Estimated projected average LTV of around $6.90 USD per user.
• Identified that the top 3 acquisition sources concentrate 75.8% of the traffic analyzed.

Problem:
Understand the drivers of user churn and design retention strategies based on segmentation and prediction.

What do I do in this project?
– Perform exploratory data analysis (EDA) to identify relevant patterns.
– Build a machine learning model to predict user churn probability (churn model).
– Segment users using clustering techniques to identify behavioral profiles (clustering).
– Integrate results into decision-oriented conclusions (conclusions).

Value added:
This project helps anticipate customer churn, understand different behavioral profiles, and prioritize evidence-based retention actions.

Key metrics:
– Model performance (Logistic Regression): accuracy = 0.925, precision ≈ 0.88, recall ≈ 0.83, AUC-ROC = 0.977.
– Segmentation: 5 clusters with differentiated profiles and contrasting churn rates.
– Churn range by segment: from ≈ 1.4 % (premium clients/Ideal) to ≈ 58.8 %(high-risk segment).

Project description:
MatchReport AI is an interactive web application that automatically generates professional football match reports using StatsBomb Open Data. The application transforms raw event data into tactical and analytical insights through a fully automated data pipeline.

What does this application do?
– Allows users to dynamically select competitions and matches.
– Generates advanced tactical and analytical visualizations.
– Automatically builds professional match reports.
– Exports reports in PDF and PNG formats.
– Allows users to download all outputs in ZIP format.

Visual analytics included:
– Shot maps (xG and goals).
– Cumulative xG by minute.
– Passing networks (home vs away).
– Progressive passes comparison.
– Defensive pressure heatmaps.

Key insights enabled:

• Compare the quality of scoring opportunities generated by each team.
• Identify possession structures and key player roles.
• Detect game progression patterns.
• Analyze defensive intensity and pressing zones.

Analytical value:
This application replicates workflows used by professional sports analytics departments, enabling non-technical users to generate tactical insights through a web-based interface.

Problem:
A large volume of Amazon product reviews makes it difficult to prioritize which feedback to review manually. The goal is to train a model that automatically classifies reviews into categories to speed up customer-sentiment understanding and detect potential issues.

What do I do in this project?
– Text cleaning and normalization, tokenization, TF-IDF vectorization, and training of multiclass classification models.
– Performance evaluation using metrics such as accuracy and F1, and interpretation of results to understand the model’s strengths and limitations.

Value added:
The model makes it easier to prioritize reviews and understand the overall tone of customer feedback without reading each text manually. It can be integrated into a product quality monitoring or customer support workflow, helping respond faster to recurring issues and detect dissatisfaction patterns.

Key metrics:

• Model training with 103,460 reviews and a vocabulary of 51,932 terms distintos.
• Best model (LinearSVC) with 89.2% accuracy on the test set.
• Weighted F1-score: 0.88, showing a good balance between precision and coverage across classes.

Problem:
An online retailer needs to better understand how, when, and what customers buy in order to optimize assortment, promotions, and logistics. The goal is to explore orders and products data to detect demand patterns and behavioral segments.

What do I do in this project?
– Explore and describe user, order, and product data.
– Preprocessing, duplicate removal, and creation of frequency, recency, and monetary metrics.
– Identify purchase patterns and customer segments based on the generated metrics and visualizations.

Value added:
The analysis helps answer questions such as how loyal customers are, what times they buy the most, and which demand patterns repeat. It can be used to design segmented campaigns, adjust inventory, and prioritize improvements in the shopping experience.

Key analysis metrics:

• Customers place an average of 3.4 orders per month, with a recurring segment that exceeds 6 orders per month.
• 35% of orders occur between 10:00 and 15:00, identifying the most relevant operational window for demand.
• RFM segmentation shows that only 18% of users generate 42% of all purchases, highlighting a high-value segment.