22 A/B Testing Interview Questions & How to Answer Them

Originally published on June 15, 2022

We’ve seen candidates ace A/B testing interviews as much as we’ve seen candidates answer interview questions in ways that made them look less qualified than they truly were. So, we decided to write this piece to help you do it right, whether you’re in product management or data science.

Our hiring team shared common questions they ask and what they’re looking for in the answers. Prepare answers to these A/B testing interview questions, and you’ll be well on your way to leaving the hiring manager with a favorable impression.

22 real-life examples of A/B testing interview questions—and how to answer them

Below is a representative sample of questions we have asked or have been asked in A/B testing interviews. As a digital optimization company, our questions are slightly biased toward the context of using A/B testing to build and grow a product.

You’ll find questions and answers for both product management and data scientist roles, as well as questions that assume that the hiring process has already verified basic knowledge of A/B testing.

Experiment design and setup

Most interviewers will start with foundational questions about A/B best practices. This is your chance to prove you know the fundamentals—after all, a solid foundation is necessary for a successful test. These questions are a bit of a warm-up and a chance for you to practice giving answers that show your thought process.

1. What are the ideal conditions for A/B testing?

A/B tests are the best tool for the job when you’ve just launched a major website or product update or when you have a specific metric you want to boost. They work best for features or elements many people are interacting with (to ensure a sufficient sample size).

2. What should you test?

Strictly speaking, A/B tests only involve one variable (multivariate tests are their own thing). An A/B test is run like a scientific experiment: First, you identify the metric you’re targeting, then use your knowledge of your customers to make an educated guess on what variables to change.

3. If A/B testing is not an available option, how would you answer a question instead?

Basic behavioral tracking can help show what customers respond to (or don’t like). Heatmaps or scroll maps are one simple example; you may opt for more detail with a full session recording. If you have the resources for it, running a product feature analysis helps your team understand how customers engage with your product. You can also collect direct feedback via customer surveys or interviews.

4. How long would you run an experiment for?

Two weeks is a sufficient minimum length for any A/B test, giving you enough time to gather data during the weekdays and weekends. Beyond that, it depends on the sample size (which is always determined during the experiment design phase). Tools like Amplitude’s duration estimator provide a starting point when designing an experiment. You can also use our statistical significance calculator to check whether or not the test results are real.

Well-designed tests account for user behavior. If you’re testing a feature, such as reporting, that some teams typically access once a month, you’ll want to extend the duration so you can include those infrequent users in your results.

A/B testing tools

Companies want to know whether you can hit the ground running, so proving your first-hand experience with common tools is necessary.

5. What A/B testing software do you recommend and why, based on your own experience?

We hope you’ll say, “Amplitude” here, but there’s no right answer to this question. When discussing the “why,” consider factors like usability and integrations alongside features.

6. How would you learn a newer A/B testing tool like Amplitude Experiment?

This is meant to explain how well you’ll adapt to how your new team works. Remember the first time you picked up whatever tool you are working with now? Pairing your methods with illustrative anecdotes will show you’re not just speaking hypothetically.

Resolving experimentation issues

Hiring managers want to see you have practical experience running A/B tests and that you’re capable of a measured response when things don’t go as expected.

7. How do you deal with small sample size issues?

Because we calculate sample size based on our desired baseline conversion rate, confidence level, and minimum detectable effect, these are the factors to look at when your sample size is much smaller than you’d like. We might decide that an A/B test with less certain results is better than no test at all.

You could look for a higher confidence level or a lower minimum detectable effect. You might also use Bayesian (rather than frequentist) statistics, especially if you already know how your customers tend to interact with your site or software.

8. What issues could impact your A/B test results in the development cycles of our product?

Timing matters for A/B tests: Testing too early might result in a small sample size, whereas testing too late may mean providing a suboptimal experience for months. Tests constrained by time may result in a small sample size or otherwise lower-quality data.

There’s also the question of whether tests might affect each other. Even if there’s no direct overlap between the features you’re testing and the metrics you’re tracking, the differentiation between customers’ experiences might skew your results.

9. How do you mitigate these issues?

Mitigating these technical issues requires effective communication between the data science and product management teams.

Data scientists will consider questions like: Can we do this with a smaller sample size? Can we have multiple tests going simultaneously and maintain confidence in our data?

PMs must ask: What is the lowest level of confidence I’d feel comfortable working with? Can we tweak our roadmap to enable a schedule that rules out potential interference?

Before proposing a new test, either party can ask: Are we running this test because we have a clear hypothesis we want to examine?

10. How do you design a test to minimize interference between control and treatment?

Minimizing interference between control and treatment groups in an A/B test means looking for (and avoiding) indirect and direct connections. Because direct connections involve an individual in the control group interacting with an individual in the treatment group, it’s best first to identify network clusters among your users. Then, assign customers not individually but as clusters.

Indirect interference is harder to spot—sometimes, it’s caused by shared resources, and other times, there are other variables that aren’t immediately obvious. The best way to avoid this problem is to use a different interval of time for the control and treatment groups.

Common A/B testing scenarios

Organizations are likely to ask about your experience with A/B testing, especially regarding their product. Because these answers are based on your specific situation, we’ll tell you what to focus on when crafting your answer.

11. Tell us about a successful A/B test you designed. What were you trying to learn, what did you learn, and how will the experience help you if you work for us?

Interviewers want to learn about your process, so start in the pre-experiment phase and take them all the way through to the data you found and how you interpreted it. Don’t focus on the test results when talking about how the experience will help you—trends among your customers may not apply to this company’s customers. Instead, share some things you think you could do differently or better in the future.

12. From your experience with using our product, what improvements would you suggest, and what experiments would you set up for them?

Prepare for this question by interacting with the company’s product for at least 10 minutes. Then, ask yourself what the company’s key business objectives likely are and what metrics relate to those. These are the metrics you’d be targeting in an A/B test; from there, it should be easy to find potential features to iterate.

13. Let’s say we want to compare Feature A and Feature B in an experiment for user flow. How would you go about designing this test, given what you know about our product?

Be ready to define a potential hypothesis and metrics that would matter to this test. From there, take your interviewers through your process: Describe variations you’d create (if necessary) and then share potential issues you’d want to watch for. Interviewers want to know you’re actively thinking about how to get useful data.

14. How do you deal with super long-term metrics where you have to wait two months to get your metric? For example, when you try to test how much money people spend during the two months after seeing a feature?

Long testing times can introduce complications, and this question addresses how you can handle them. Be ready to discuss potential shifts in data caused by novelty, primacy effects, or customer-side changes like deleted cookies and evolving needs. There’s also the threat of interference with overlapping tests, which you’re more likely to run the longer a test goes on. Don’t forget to address how you’d justify your decision to impatient PMs or other stakeholders pushing you for quicker results.

Data analysis and decision-making

Gathering valid data is one skill; gleaning useful insights from it is another. Interviewers want to understand your thought process when making sense of your A/B tests.

15. What would you do if your experiment is inconclusive and looks more like an A/A test? How would you analyze the test results, and what would you look into?

The first step after receiving an inconclusive test result is to look closer at the data to ensure it hasn’t been polluted. Also, make sure your audience was properly segmented and that no other tests or factors interfered with your experiment.

If your test was sound, look at secondary metrics—as long as they’re ones you previously defined, not ones you’ve cherry-picked. Then, segment data: Look at mobile users vs. desktop users and new vs. returning audiences. And make sure to segment any data that might have been affected by a simultaneous test.

Finally, ask yourself what an inconclusive test means: What have you disproved?

16. When you know there is a social network effect and the independence assumption doesn’t hold, how does it affect your analysis and decisions?

For social network tests where an independence assumption does not hold, the effect is amplified. The network effect brings the control and treatment closer together because one group affects the other.

Say the treatment group performed 2% better than the control—that data is what you saw after the control group’s behavior was affected, which means it’s skewed toward the treatment group’s results.

17. In our A/B test, the results were not statistically significant. What are some potential reasons for this?

It’s always possible the variable you were testing didn’t affect customers’ behavior, and you want to keep this in mind before you waste time going down rabbit holes. However, design issues like a small sample size or insufficient statistical power can lead to a statistically insignificant result. If you’re seeing a lot of variance in your key metric, it may be worth revisiting your metric or implementing stratification or Controlled-experiment Using Pre-Existing Data (CUPED).

18. What do you do when you’re testing for two metrics and aim to increase both, but one increases with statistical significance, and the other one decreases with statistical significance?

Deciding which metric to prioritize in this situation depends on the significance of each business. If the metric that’s more important to your bottom line was the one that decreased, it’s not a change worth making.

Workflows and resources

Finally, interviewers are likely to ask questions that dig into how you use resources. These questions are more likely to be aimed at product managers, not data scientists.

19. What software do you recommend for reporting on experiment results?

Whatever your choice, be sure you’re thinking beyond just your role. Talk about how the software (and its outputs) work for everyone, not just those trained in it.

20. What tools would you integrate with your A/B testing software in order to get more from the experiment data?

This question is designed to show how you think about building systems. It’s likely the company already has a tech stack to support their A/B tests. Still, they’ll want to see that you can identify important ancillary capabilities like advanced statistical analysis and segmentation.

21. What new hires would you suggest for your A/B testing team if you already have team members for roles X, Y, Z?

Hiring managers ask this to see if you understand the ins and outs of A/B testing. The best teams include a variety of specialists who have expertise in data analytics or machine learning, statistics, design, consumer psychology and behavior, and engineering.

22. Which roles on your product team should be involved in your tests, and how would you make it easy for them to be involved?

A thoughtful answer to this question addresses the importance of collaboration in A/B testing. Testing isn’t just about the experience (UX designers) and functionality (engineers). Marketers can share a wealth of information about your ideal customer profile, while product managers can speak to overall strategic goals and ensure your hypotheses align with long-term plans.

Common A/B testing question mistakes to avoid

Now that you have the answers, it’s time to talk about how you share them in the context of an interview. Common mistakes we’ve seen candidates make are:

• Showing their technical skills but not their creative side or analytical thought process
• Talking about their previous experience without making the answers pertinent to the context of the company that’s interviewing them
• Focusing on just one tool they used without showing interest in learning new tools

Your interviewers already know you have experience in A/B testing thanks to their screening process. During the interview, they want to hear how you approach problems. When you show your work, you’re letting them see your thought process and how you perform on a team.

Our top candidates haven’t just excelled at giving technical answers—they’ve included anecdotes and statements that show they’re aware of their impact on the organization as a whole. Whether you’ll be running experiments that guide product development or perfecting marketing campaigns, speak to the larger context of your work to prove you’d be an asset.

Pair the new A/B testing position with top tools

A new job will empower you to do great work if you also have the best software. We invite you to keep going and learn how to analyze A/B test results in Amplitude Analytics or how to run tests in Amplitude Experiment. You can also review our list of 11 top A/B testing tools.

References

• How to Do A/B Testing: 15 Steps for the Perfect Split Test. Hubspot
• 7 lessons learned from 5 years of product-led experimentation. Productboard
• Product Management Skills: A/B Testing. Product School.