American Express Data Analytics Interview Questions and Answers (2026 Guide)

Quick answer: American Express Data Science and Analytics interviews generally test SQL, Python, statistics and Machine Learning fundamentals alongside a company or domain specific case discussion. This guide covers the likely process, core technical questions, and how to prepare.

Data Science and Analytics now sit at the centre of how large organisations make decisions. Companies use SQL, Python, statistics and Machine Learning to understand customers, reduce risk, forecast demand and improve day to day operations.

American Express is a global financial services company best known for its payments and card business, along with travel and merchant services. Data science and analytics sit close to the core of the business, supporting credit risk, fraud detection, customer engagement and marketing decisions.

If you are preparing for a American Express Data Science or Data Analytics interview, understanding the likely structure of the process and the topics that commonly come up will help you prepare with far more focus.

About American Express

American Express works across areas such as:

• Payments and card services

• Consumer and commercial banking

• Merchant services

• Travel services

• Global risk management

Data and analytics teams typically support work such as:

• Credit risk modelling

• Fraud detection and prevention

• Customer lifetime value analysis

• Marketing and offer targeting

• Churn and retention modelling

• Portfolio and pricing analytics

Roles that commonly open up in this space include:

• Data Analyst

• Data Scientist

• Risk Analyst

• Decision Scientist

• Machine Learning Engineer

Interview Process

Hiring processes vary by role, team and experience level. A data role at a company of this size generally moves through several stages.

1. Online Assessment

Often covers:

• Aptitude and logical reasoning

• SQL queries

• Python programming

• Statistics fundamentals

2. Technical Interview

Topics commonly covered include:

• SQL joins, aggregations and window functions

• Python and Pandas for data manipulation

• Statistics and probability

• Machine Learning fundamentals

• Data visualisation and reporting

Case or Business Round

You may be asked to reason through an open ended business problem, explain your approach, and justify the metrics you would track.

Managerial and HR Round

Focus areas usually include project experience, communication, stakeholder management, and how you handle ambiguity.

SQL Interview Questions

What is the difference between WHERE and HAVING?

What is the difference between INNER JOIN and LEFT JOIN?

INNER JOIN returns only the rows that match in both tables. LEFT JOIN returns every row from the left table, and fills unmatched columns from the right table with NULL.

SELECT c.customer_id,
       c.name,
       o.order_id
FROM customers c
LEFT JOIN orders o
  ON c.customer_id = o.customer_id;

How do you find duplicate records in a table?

SELECT email, COUNT(*) AS record_count
FROM customers
GROUP BY email
HAVING COUNT(*) > 1;

What are window functions?

Window functions perform a calculation across a set of rows while still returning every individual row. They are widely used for ranking, running totals and period on period comparisons.

SELECT region,
       sales_amount,
       RANK() OVER (
         PARTITION BY region
         ORDER BY sales_amount DESC
       ) AS sales_rank
FROM regional_sales;

What is a Common Table Expression?

A Common Table Expression, written with the WITH clause, creates a named temporary result set that exists only for the duration of the query. It makes long queries far easier to read and debug.

WITH monthly_totals AS (
  SELECT customer_id,
         SUM(amount) AS total_spend
  FROM transactions
  GROUP BY customer_id
)
SELECT *
FROM monthly_totals
WHERE total_spend > 10000;

Python Interview Questions

Why is Python widely used for data work?

Python combines readable syntax with a mature ecosystem of libraries. Commonly used ones include:

• Pandas for data manipulation

• NumPy for numerical computing

• Scikit-Learn for Machine Learning

• Matplotlib and Seaborn for visualisation

What is the difference between a list and a tuple?

How do you handle missing values in Pandas?

import pandas as pd

# Inspect how much is missing
df.isnull().sum()

# Drop rows where a critical field is missing
df = df.dropna(subset=['customer_id'])

# Fill numeric gaps with the median
df['income'] = df['income'].fillna(df['income'].median())

The right choice depends on why the data is missing and how much of it is missing. Removing rows is safe only when the missing share is small and not systematic.

What is the difference between merge, join and concat?

merge combines DataFrames on key columns, similar to a SQL join. join combines on the index by default. concat stacks DataFrames along an axis without matching keys.

Statistics Interview Questions

What is the difference between mean, median and mode?

The mean is the arithmetic average, the median is the middle value in sorted data, and the mode is the most frequent value. The median is preferred when the data contains outliers, because it is not pulled by extreme values.

What is standard deviation?

Standard deviation measures how far values typically fall from the mean. A low value means the data is tightly clustered, a high value means it is spread out.

What is a p-value?

A p-value is the probability of observing a result at least as extreme as the one measured, assuming the null hypothesis is true. A small p-value gives evidence against the null hypothesis. It does not tell you the size of the effect or that the result matters commercially.

What is the Central Limit Theorem?

The Central Limit Theorem states that the distribution of sample means approaches a normal distribution as sample size increases, regardless of the shape of the population distribution. It is the reason so many statistical tests work on large samples.

What is the difference between Type I and Type II error?

Machine Learning Interview Questions

What is the difference between supervised and unsupervised learning?

What is overfitting and how do you prevent it?

Overfitting happens when a model learns noise in the training data and fails to generalise to new data. Common remedies include cross validation, regularisation, simplifying the model, pruning features, and gathering more representative data.

Explain precision and recall

Precision is the share of predicted positives that are actually positive. Recall is the share of actual positives that the model successfully identified. There is usually a trade off between them, and which one matters more depends entirely on the cost of each type of mistake.

Why is accuracy a poor metric for imbalanced data?

If only one percent of cases are positive, a model that predicts negative every single time is ninety nine percent accurate and completely useless. Precision, recall, F1 score and ROC AUC give a far more honest picture.

What is cross validation?

Cross validation splits the data into several folds, trains on some and validates on the rest, then rotates. K-Fold Cross Validation is the most common form. It gives a more reliable estimate of performance than a single train and test split.

Risk, Fraud and Financial Analytics Questions

How would you build a fraud detection model?

Fraud data is heavily imbalanced, so accuracy is misleading. Useful features include transaction amount relative to the customer's own history, location and device signals, merchant category and velocity of recent transactions. Evaluate with precision, recall and the cost of each error type, since a false positive blocks a genuine customer and a false negative lets fraud through.

How do you handle class imbalance?

Options include resampling, class weights, and choosing threshold and metrics deliberately. In practice, tuning the decision threshold against business cost often matters more than the sampling technique.

What is a credit risk scorecard?

A scorecard converts customer attributes into a score that estimates the probability of default. Scorecards are widely used in lending because they are interpretable and can be explained to both customers and regulators, which a black box model often cannot.

Why does model explainability matter in financial services?

Lending and risk decisions are regulated. If a customer is declined, the institution may need to explain why. This is a major reason logistic regression and tree based models with clear feature attribution remain common in credit decisions.

What is customer lifetime value?

Customer lifetime value estimates the total profit expected from a customer over the length of the relationship. It informs how much can rationally be spent on acquiring and retaining a customer.

Case Study Questions

Card spending has dropped in one customer segment

Separate the drop into fewer customers transacting, or the same customers transacting less. Check acquisition, attrition, merchant mix, seasonality and any recent policy or pricing change. Only after locating where the drop sits should you propose a cause.

Reducing false positives in fraud alerts

Quantify the cost of a blocked genuine transaction against the cost of missed fraud, then move the threshold with that trade off explicit. Better features and customer level baselines usually reduce false positives more than swapping the algorithm.

HR and Behavioural Questions

Tell me about yourself

A clear structure works well: your education, your technical skills, one or two projects you can defend in depth, any work experience, and what you are looking for next. Keep it under two minutes.

Why American Express?

A credible answer usually connects the scale and quality of transaction data to the kind of problems you want to solve. Fraud and credit risk are areas where a model decision has an immediate, measurable financial consequence, which makes the analytical work unusually concrete.

Describe a project you are proud of

Use a simple arc: the business problem, the data you had, what you built, how you evaluated it, and what changed as a result. Interviewers are far more interested in your reasoning than in the algorithm you picked.

Tell me about a time a project did not work

Answer honestly. Describe what went wrong, what you learned, and what you would do differently. A candidate who can discuss failure clearly usually reads as more experienced, not less.

Preparation Tips

Build genuine SQL fluency

Practise joins, aggregations, subqueries, window functions and CTEs until you can write them without hesitation. SQL is the single most commonly tested skill in data interviews.

Get comfortable with Pandas

Focus on merging, grouping, reshaping, handling missing values and cleaning messy real world data rather than memorising the entire library.

Revise the statistics that actually come up

Hypothesis testing, distributions, correlation, sampling and experiment design appear far more often than advanced theory.

Prepare two projects properly

Two projects you can discuss deeply beat six you can only describe superficially. Be ready to explain your choices and the limitations of your work.

Practise explaining technical work to non technical people

Almost every data role sits between a technical system and a business decision. The ability to translate between them is repeatedly tested.

Final Thoughts

Interviews at American Express reward candidates who combine solid technical foundations with clear business reasoning. Strong SQL, practical Python, dependable statistics, and the ability to explain your thinking usually matter more than knowing an unusually advanced algorithm.

Prepare systematically, build projects you can genuinely defend, and practise speaking about your work out loud. That combination is what separates candidates who pass from candidates who freeze.