← Citi Interview Insights

Citi·Data Scientist·Technical Phone Screen·Senior

SeniorPrefer not to say
Jun 2026

Summary

This was a case-style interview for a data scientist role at Citi, basically a product analytics deep dive on a crypto trading scenario. The structure was pretty demanding and they clearly wanted you to think out loud through measurement, segmentation, and recommendations rather than just listing ideas.

Questions Asked (5)

Q1

BTC market volatility is elevated but retail trading volume on the platform has dropped. Walk through how you'd investigate this with data and translate your findings into a concrete recommendation.

Product Analytics & MetricsRoot Cause AnalysisAdaptability & Ambiguity
Author's notes

This is the kind of question where you can really embarrass yourself by jumping straight to 'users are scared of volatility' without checking whether the metric is even real.

Create a free account to read the full note

AI HintsAI Generated

Suggested Approach

Start by clarifying the metrics and defining the scope, then systematically segment the data to isolate the root cause—whether it's a drop in user activity, a shift in user mix, or a product issue. Finally, translate the findings into a business recommendation that addresses the underlying driver and aligns with Citi's strategic priorities.

Pro tip: Acknowledge that volatility can both attract and deter retail traders; consider that the drop might be due to risk aversion or a shift to other assets. Also, mention the importance of checking data quality and external factors like market events or platform changes before jumping to conclusions.

1. Clarify the question and define metrics

Confirm what 'retail trading volume' means (e.g., number of trades, notional value, active traders) and the time frame. Ensure you understand the platform's definition of 'retail' and how volatility is measured.

2. Segment the data to find patterns

Break down volume by user cohorts (new vs. existing, high-value vs. low-value), product types (stocks, options, crypto), and demographics. Look for disproportionate drops that indicate a specific segment driving the decline.

3. Investigate potential causes

Analyze internal factors (e.g., platform outages, fee changes, UX issues) and external factors (e.g., market sentiment, competing platforms, macroeconomic news). Use funnel analysis to see where users drop off.

4. Validate findings and quantify impact

Run statistical tests or A/B analysis if possible to confirm hypotheses. Quantify the contribution of each factor to the overall volume drop to prioritize actions.

5. Formulate a recommendation

Based on the root cause, propose a concrete action (e.g., targeted re-engagement campaign, product improvement, risk education) with expected impact and success metrics. Align with business goals.

Key Points to Mention

  • Segmentation by user cohorts and product types to identify the source of the decline
  • Consideration of both internal (platform changes) and external (market conditions) factors
  • Use of funnel analysis to pinpoint where retail users disengage
  • Quantification of the impact of each factor to prioritize recommendations
  • Alignment of the recommendation with Citi's business objectives and risk appetite
  • Acknowledgment of data limitations and the need for iterative investigation

AI-generated suggestions, not part of the candidate's original notes. May be inaccurate — verify before relying on them.

Q2

If session counts are stable but executed trade volume is falling, what does that tell you and what would you do next?

Product Analytics & MetricsRoot Cause Analysis
Author's notes

Liked this one because it's actually a sharp diagnostic.

Create a free account to read the full note

AI HintsAI Generated

Suggested Approach

Start by acknowledging that stable sessions with falling trade volume indicates a drop in trading activity per session, not a traffic issue. Then systematically break down the possible causes—from user behavior changes to product or market factors—and propose a data-driven investigation plan to pinpoint the root cause and recommend actions.

Pro tip: Frame your answer around the 'why' and 'so what'—don't just list hypotheses; prioritize them by business impact and feasibility of testing. Mention that you'd validate findings with A/B tests or holdout groups before making recommendations.

1. Clarify the metric definitions and scope

Confirm what 'session counts' and 'executed trade volume' mean (e.g., unique users, trades per session, notional value) and the time frame. Ensure you're comparing apples to apples and rule out data pipeline issues.

2. Segment the data to localize the drop

Break down trade volume by user segments (new vs. existing, retail vs. institutional), product types, geographies, and device/platform. Identify whether the decline is broad-based or concentrated in specific segments.

3. Generate and prioritize hypotheses

List potential causes: changes in user behavior (e.g., fewer trades per session), product changes (e.g., fee structure, UI), market conditions (e.g., low volatility), or external factors (e.g., competitors). Prioritize by likelihood and impact.

4. Analyze and validate with data

Use funnel analysis, cohort analysis, and statistical tests to confirm which hypotheses hold. Check for correlations with deployments, marketing campaigns, or market events. Quantify the impact of each factor.

5. Recommend actions and next steps

Based on findings, propose actionable recommendations—e.g., A/B test a new feature, adjust pricing, or target a specific segment. Define success metrics and a plan to monitor the impact.

Key Points to Mention

  • Distinguish between session count and session quality—stable sessions don't guarantee stable engagement.
  • Consider the possibility of a shift in user mix (e.g., more low-value users) or a change in trading behavior (e.g., fewer trades per session).
  • Check for external factors like market volatility, interest rates, or competitor promotions that could affect trading volume.
  • Use segmentation to identify if the drop is driven by a specific product, channel, or customer segment.
  • Validate hypotheses with statistical rigor (e.g., hypothesis testing, confidence intervals) before concluding causality.
  • Propose a test-and-learn approach (e.g., A/B test) to confirm the root cause and measure the impact of any intervention.

AI-generated suggestions, not part of the candidate's original notes. May be inaccurate — verify before relying on them.

Q3

How would you test whether a change in risk policy caused the trading volume drop?

A/B Testing & ExperimentationRoot Cause Analysis
Author's notes

Blanked for a second on the cleanest way to frame this.

Create a free account to read the full note

AI HintsAI Generated

Suggested Approach

Start by clarifying the business context and data available, then outline a structured causal inference approach that combines experimental and observational methods. Emphasize the importance of ruling out alternative explanations and validating assumptions before concluding causality.

Pro tip: In a bank like Citi, policy changes are rarely randomized, so you'll often rely on quasi-experimental designs like difference-in-differences or synthetic control. Mention that you'd also check for pre-existing trends and consider the possibility of anticipatory effects.

1. Clarify the policy change and data

Understand exactly what the policy change entailed, when it was implemented, and which trading desks or products were affected. Identify available data on trading volume, policy application, and potential confounders.

2. Define the counterfactual

Determine what trading volume would have been in the absence of the policy change. This could involve a control group (e.g., unaffected desks), a pre-post comparison, or a model-based prediction.

3. Choose appropriate causal inference methods

Select methods such as difference-in-differences, synthetic control, interrupted time series, or instrumental variables, depending on data structure and whether randomization occurred. If a natural experiment exists, use it.

4. Test assumptions and robustness

Check for parallel trends, placebo tests, sensitivity to model specifications, and potential confounders like market volatility, seasonality, or concurrent events. Validate that the policy change is the most plausible cause.

5. Quantify and communicate impact

Estimate the effect size with confidence intervals, and translate findings into business insights. Acknowledge limitations and suggest further validation if needed.

Key Points to Mention

  • Difference-in-differences (DiD) and parallel trends assumption
  • Synthetic control method for case studies
  • Interrupted time series analysis
  • Confounders: market conditions, seasonality, other simultaneous changes
  • Placebo tests and robustness checks
  • Causal inference vs. correlation, and the importance of counterfactuals

AI-generated suggestions, not part of the candidate's original notes. May be inaccurate — verify before relying on them.

Q4

During a volatility spike, what metrics would signal that a reliability or infrastructure issue is causing the volume drop rather than a demand issue?

Product Analytics & MetricsRoot Cause Analysis
Author's notes

Shorter answer than they probably wanted.

Create a free account to read the full note

AI HintsAI Generated

Suggested Approach

Start by distinguishing demand-driven volume drops from supply-side failures using a set of reliability and infrastructure metrics. Focus on metrics that capture system health, error rates, latency, and user behavior anomalies, and compare them against historical baselines during volatility spikes. Emphasize that a combination of these signals, rather than a single metric, provides strong evidence of an infrastructure issue.

Pro tip: In a trading context like Citi, correlate infrastructure metrics with market data timestamps to rule out demand-side explanations; for example, if error rates spike exactly when market volatility increases, it's likely a system issue. Also, consider that demand-driven drops often show gradual declines, while infrastructure issues cause sudden, sharp drops.

1. Establish baseline behavior

Define what normal looks like for key metrics during volatility spikes, including expected volume patterns and system performance thresholds. This helps identify anomalies that deviate from typical demand-driven fluctuations.

2. Monitor system health metrics

Track infrastructure metrics such as error rates (e.g., HTTP 5xx, API failures), latency (e.g., p99 response times), throughput, and resource utilization (CPU, memory, network). Spikes in these metrics during volume drops suggest reliability issues.

3. Analyze user behavior and funnel metrics

Examine user-level metrics like session duration, page load failures, transaction abandonment rates, and retry attempts. If users are trying to engage but failing, it points to infrastructure problems rather than lack of demand.

4. Correlate with external factors and logs

Cross-reference volume drops with deployment logs, incident reports, and external market events. If the drop coincides with system changes or errors, it's likely infrastructure-related; if it aligns with market-wide trends, it may be demand-driven.

5. Validate with A/B testing or canary analysis

If possible, compare affected vs. unaffected user segments or regions to isolate infrastructure impact. Consistent drops across all user segments during system issues indicate a reliability problem.

Key Points to Mention

  • Error rates (e.g., 5xx errors, API failures) spiking during volume drops
  • Latency metrics (e.g., p95/p99 response times) increasing beyond normal thresholds
  • Infrastructure resource utilization (CPU, memory, network) hitting limits
  • User behavior anomalies: increased retries, session timeouts, abandonment rates
  • Correlation with deployment or incident timelines
  • Comparison with historical volatility events to distinguish demand patterns

AI-generated suggestions, not part of the candidate's original notes. May be inaccurate — verify before relying on them.

Q5

If the volume decline is isolated entirely to new users, how does your recommendation change?

Product Analytics & MetricsProduct Sense & Ideation
Author's notes

This reframe is actually kind of interesting because it shifts the problem from retention or market sentiment to onboarding and activation.

Create a free account to read the full note

AI HintsAI Generated

Suggested Approach

First, acknowledge that isolating the decline to new users fundamentally changes the problem from a broad retention issue to a top-of-funnel acquisition or onboarding issue. Then, systematically segment new users by acquisition channel, cohort, and early behavior to pinpoint where the drop-off occurs, and adjust your recommendation to focus on fixing that specific stage.

Pro tip: Quantify the impact of the new-user decline on overall metrics (e.g., what percentage of total volume do new users typically contribute?) to show you understand business context and prioritize accordingly.

1. Confirm and quantify the new-user decline

Validate that the decline is indeed isolated to new users by comparing new vs. returning user trends, and calculate the contribution of new users to total volume to assess severity.

2. Segment new users to find the drop-off point

Break down new users by acquisition channel, geography, device, and cohort to identify if the decline is concentrated in specific segments or across the board.

3. Analyze the new-user journey

Map the new-user funnel from acquisition to first transaction, and measure conversion rates at each step to locate where the biggest drop occurs.

4. Form and test hypotheses

Generate hypotheses for the decline (e.g., marketing campaign changes, onboarding friction, product changes) and propose A/B tests or further analyses to validate them.

5. Revise recommendation

Based on findings, tailor your recommendation to address the root cause, such as optimizing acquisition channels, improving onboarding, or adjusting targeting.

Key Points to Mention

  • Cohort analysis to compare new user behavior over time
  • Funnel analysis to identify drop-off points in the new-user journey
  • Segmentation by acquisition channel, campaign, and user demographics
  • Impact of external factors (e.g., seasonality, competitor actions, marketing budget changes)
  • Onboarding experience and time-to-first-value for new users
  • Prioritization based on potential impact and ease of implementation

AI-generated suggestions, not part of the candidate's original notes. May be inaccurate — verify before relying on them.