Analytics Day 0 Cheat Sheet: Formulas, Frameworks and Interview Answers
A payments app can be “99% accurate” by simply calling every transaction safe - and still let fraudsters walk through. That is the whole point of analytics: the winner is not the person who remembers the most formulas, but the one who chooses the right formula for the business decision.
- Analytics is the scientific process of transforming data into insight for making better decisions.
- Every analytics answer should move from business question - data - method - metric - decision.
- Use descriptive analytics to explain what happened, predictive analytics to estimate what will happen, and prescriptive analytics to recommend what to do.
- For classification models, never quote accuracy alone. Add precision, recall, F1, AUC and explain the cost of false positives versus false negatives.
- For A/B tests, report conversion lift, sample size, statistical significance, practical significance and whether the test was clean.
- The most interview-safe analytics structure is: define objective - form hypothesis - choose data - select method - evaluate - recommend action.
- The biggest trap is jumping to a model before clarifying the business problem and success metric.
The Big Picture: Analytics Is a Decision Machine, Not a Formula List
Think of analytics as a left-to-right decision pipeline. Data is only useful when it changes a decision - whom to target, which price to set, which customer to approve, which warehouse to replenish, or which campaign to stop.
Core Explanation: The Day 0 Analytics Playbook
Use this section as your mental operating system. When a case, resume probe, guesstimate follow-up, product question, marketing question or finance analytics question appears, first classify the problem. Then select the method and metric.
1. Match the Question to the Analytics Type
Most business analytics problems fall into three buckets:
- Descriptive analytics: What happened? Example: sales fell by region, channel or cohort.
- Diagnostic analytics: Why did it happen? Example: churn rose because first delivery delays increased in a specific city cluster.
- Predictive analytics: What is likely to happen? Example: forecasting demand for next week.
- Prescriptive analytics: What should we do? Example: optimize discount allocation subject to margin and inventory constraints.
2. The Formula Card You Actually Need
You do not need to recite every formula in statistics. You need the formulas that help you explain a business result clearly, defensibly and fast.
3. Metrics That Make Your Answer Sound Like an Analyst
Always define what “good” means before evaluating a model. A high number is not automatically good - it depends on business cost, base rate and actionability.
4. Classification Metrics in One Picture
If the model predicts a class - fraud or not fraud, churn or no churn, convert or not convert - use a confusion matrix. It forces you to speak in business costs: which mistake is worse?
5. Worked Example: A/B Test in 90 Seconds
A product team tests a new checkout page.
- Control: 10,000 visitors, 500 purchases, conversion rate = 5.0%
- Variant: 10,000 visitors, 560 purchases, conversion rate = 5.6%
Step 1 - Calculate lift: Lift = (5.6% - 5.0%) / 5.0% = 12%.
Step 2 - Check statistical evidence: Pooled conversion = 1,060 / 20,000 = 5.3%. Standard error is approximately sqrt(0.053 x 0.947 x (1/10,000 + 1/10,000)) = 0.317 percentage points.
Step 3 - Calculate z-score: Difference = 0.6 percentage points. z = 0.6 / 0.317 = about 1.89. The two-tailed p-value is roughly 0.06, so it is not significant at the 5% level.
Business answer: “The variant shows a promising 12% lift, but evidence is just short of 95% confidence. I would check test hygiene, segment-level effects and commercial impact before rollout.”
Definitions You Should Be Able to Say Cleanly
Analytics, per INFORMS: “The scientific process of transforming data into insight for making better decisions.”
Razorpay: Analytics as a Real-Time Risk Decision System
Razorpay shows how analytics moves from dashboards to real-time decisions in digital payments, where speed, trust and risk control must work together.
Situation: Indian digital payments operate at enormous transaction velocity across UPI, cards, net banking and wallets. A payment gateway must approve genuine payments quickly while detecting suspicious patterns, protecting merchants and staying compliant with RBI-linked payment norms and data protection expectations.
The move: Razorpay, like leading payment platforms globally, uses analytics across the payment journey - merchant onboarding, transaction monitoring, fraud-risk scoring, success-rate optimization and reconciliation. The primary driver is real-time risk scoring using transaction patterns. Supporting drivers include rule-based controls, anomaly detection, merchant category intelligence, historical dispute feedback, human review for edge cases and continuous monitoring of payment success rates.
Outcome or lesson: The strategic lesson is not “use machine learning.” The lesson is to combine business rules, statistical signals, operational review and feedback loops so that the model improves decisions without blocking too many genuine customers.
So what: This case proves that analytics value comes from connecting model output to operational action. A technically strong model that slows checkout or blocks good users can destroy business value.
How AI Changes Analytics in 2026
AI does not remove analytics fundamentals. It raises the bar because routine data work is faster, so interviewers expect sharper problem framing and better judgment.
- Natural-language analytics and GenBI: Tools can turn plain English questions into SQL, charts and summaries. The risk is that users may accept a chart without checking data definitions, joins or leakage.
- AutoML and model copilots: AI can test models, tune parameters and generate feature ideas quickly. The analyst still owns the objective, metric, validation design and ethical constraints.
- Unstructured-data analytics: LLMs make call transcripts, support tickets, reviews, policy documents and earnings-call text analyzable at scale. This expands analytics beyond neat spreadsheets.
Use NotebookLM for revision: upload your analytics notes, one company annual report or investor presentation, and this cheat sheet. Ask it to generate 10 interview questions linking that company to metrics, A/B testing, forecasting and classification. Then answer aloud using the business question - data - method - metric - decision flow.
Interview Relevance
“Suppose an e-commerce company wants to reduce customer churn. How would you use analytics to solve the problem?”
Always add one sentence on implementation: “I would pilot the intervention on a test group, compare it with a holdout group and track incremental retention, margin and customer experience.” That turns an analytics answer into a business answer.
Common Mistake
The mistake: quoting a model or formula before defining the business objective and success metric. It costs candidates because it sounds like classroom statistics, not managerial analytics. Fix: start every answer with “The decision we need to improve is…” and only then choose data, method and metric.
What to Revise Next
This is the final lesson, so do a capstone review rather than opening a new topic. Pick one company you care about - for example, HDFC Bank, Zomato, Tata 1mg, Indigo, Nykaa or Razorpay - and prepare a one-page analytics teardown.