Hard Business Analytics Interview Questions

Show answer

WITH m AS (SELECT month, SUM(amount) AS rev FROM orders GROUP BY month) SELECT month, rev, ROUND(100.0 * (rev - LAG(rev) OVER (ORDER BY month)) / LAG(rev) OVER (ORDER BY month), 1) AS mom_pct FROM m;

A bank wants to reduce customer churn. What would you do?

Show answer

Define churn, analyse who leaves and why, segment customers, find drivers through data and surveys, design retention offers and measure uplift with a control group.

Revenue grew but profit fell. Why could that happen?

Show answer

Look at the cost side: heavier discounts, higher CAC, logistics costs, product mix shifting to low-margin items, one-time expenses or pricing changes. Break contribution margin down by segment.

How would you price a new subscription product?

Show answer

Combine cost-plus, competitor benchmarks and willingness to pay from research, test tiers with experiments and check margin and churn impact.

How do you estimate the annual revenue of a coffee shop chain?

Show answer

Number of outlets x daily customers x average bill x operating days, then validate using footfall, peak hours and similar chains.

How do you manage conflicting requirements from two senior stakeholders?

Show answer

Clarify the underlying goals, present data and trade-offs, facilitate a joint session, escalate to the sponsor if needed and document the agreed decision.

Should a food-delivery app launch in a new city?

Show answer

Check market size (population, online ordering share), competition, restaurant supply, unit economics per order, CAC and payback period. Propose a pilot with clear success metrics.

What is a KPI tree?

Show answer

It breaks a top metric into drivers: Revenue = Users × Conversion × AOV; Users = New + Returning, and so on. It shows which lever to pull and where a drop came from.

What is root cause analysis and which techniques do you know?

Show answer

Finding the underlying reason for a problem using the 5 Whys, fishbone diagrams or Pareto analysis, then fixing the cause not the symptom.

Explain row context and filter context in DAX.

Show answer

Row context means DAX is evaluating one row at a time, as in a calculated column or inside iterators like SUMX. Filter context is the set of filters coming from slicers, visuals, rows, columns and CALCULATE that decides which rows a measure sees. Row context does not filter other tables by itself; context transition through CALCULATE converts it into filter context. Most wrong-number bugs in DAX come from misunderstanding these two.

What are LOD expressions in Tableau and how do FIXED, INCLUDE and EXCLUDE differ?

Show answer

Level of Detail expressions compute an aggregate at a granularity different from the view. FIXED computes at the stated dimensions regardless of the view, for example { FIXED [Customer ID] : MIN([Order Date]) } for first purchase date. INCLUDE adds a dimension to the view level, useful for average sales per customer by region. EXCLUDE removes a dimension, useful for comparing each month with the yearly total. FIXED ignores normal dimension filters unless they are context filters.

Why should you avoid bidirectional relationships in a Power BI model?

Show answer

Relationships normally filter one way, from the dimension (one side) to the fact (many side). Bidirectional filtering lets facts filter dimensions too, which can create ambiguous filter paths, unexpected totals and slower queries, especially with multiple fact tables. Prefer a clean star schema with single-direction filters, and when you need a reverse filter for one measure, use CROSSFILTER or TREATAS inside CALCULATE instead of changing the model globally.

What is incremental refresh and when do you need it?

Show answer

Incremental refresh partitions a large table by date so each refresh reloads only recent data, such as the last 10 days, while older partitions stay stored. You define RangeStart and RangeEnd parameters in Power Query, filter the date column with them, and set the policy for how much history to keep. It is needed when full refreshes of millions of rows take too long or fail, and it reduces load on the source system.

Your Power BI report takes 30 seconds to load. How do you speed it up?

Show answer

Start with Performance Analyzer to see which visuals and DAX queries are slow. Then fix the model: remove unused columns, reduce high-cardinality text, use a star schema and correct data types. Rewrite heavy measures, avoiding FILTER over whole tables and nested iterators. Cut the number of visuals per page, use aggregations or Import instead of DirectQuery where possible, and push transformations upstream into SQL or Power Query.

Five departments each build their own Power BI reports with different revenue numbers. How would you design a better BI setup?

Show answer

Create a shared, certified semantic model with one definition per KPI, built on a governed warehouse layer. Departments build thin reports on top of it through live connection instead of copying data. Set up workspaces by environment (dev, test, prod), deployment pipelines, naming standards, a KPI glossary and owners for each measure. Keep a small central team for the core model and let departments self-serve on top, which balances control and speed.

A Tableau dashboard on a 50 million row data source responds slowly to filters. How do you fix it?

Show answer

Use an extract with only the needed columns and rows, aggregated to the visible grain where possible. Replace quick filters showing all values with fewer filters, use context filters wisely and avoid 'only relevant values' on large fields. Move complex calculations into the data source, reduce marks per view and the number of sheets per dashboard. Use the Performance Recorder to find the slowest queries and fix those first.

Write a DAX measure for a running total of sales by date.

Show answer

Running Sales = CALCULATE( [Total Sales], FILTER( ALL('Date'[Date]), 'Date'[Date] <= MAX('Date'[Date]) ) ) -- for a yearly reset use TOTALYTD([Total Sales], 'Date'[Date])

Write a DAX measure to count distinct active customers in the last 30 days.

Show answer

Active Customers 30D = CALCULATE( DISTINCTCOUNT(Sales[CustomerID]), DATESINPERIOD('Date'[Date], MAX('Date'[Date]), -30, DAY) ) -- MAX('Date'[Date]) makes it relative to the date in the current filter

What is Excel Solver and give a business example of using it.

Excel for Business AnalystsHard

Show answer

Solver is an optimisation add-in that finds the best value of a target cell by changing decision cells within constraints. Unlike Goal Seek, it handles many variables and constraints. Example: choose how many units of three products to make to maximise profit, subject to machine hours, raw material and minimum order limits. You set the objective, the variable cells, constraints and a method such as Simplex LP for linear problems.

Finance asks you to build a monthly cohort retention table in Excel from raw order data. How do you approach it?

Excel for Business AnalystsHard

Show answer

Add a helper column for each customer's first order month using MINIFS on customer ID, and another for the order month. Compute the month index as the difference in months between them. Then build a PivotTable or COUNTIFS grid with cohort month as rows, month index as columns and distinct customers as values. Divide each cell by the month-zero count to get retention percentages and add a colour scale to show decay.

How can LET and LAMBDA make complex Excel formulas easier to maintain? Give an example.

Excel for Business AnalystsHard

Show answer

// LET names intermediate values so they are calculated once =LET(rev, SUM(C2:C100), cost, SUM(D2:D100), IF(rev=0, 0, (rev - cost) / rev)) // LAMBDA saved in Name Manager as MARGIN =LAMBDA(rev, cost, IF(rev=0, 0, (rev - cost) / rev)) // then use =MARGIN(C2, D2) anywhere in the workbook

What is normalisation, and why are analytics tables often denormalised?

Show answer

Normalisation splits data into related tables to remove redundancy and update anomalies, which suits transactional systems that write often. Analytics workloads read far more than they write, so warehouses often denormalise into wide fact and dimension tables or star schemas. That reduces joins, makes queries simpler for analysts and speeds up aggregation, at the cost of extra storage and the need for controlled loading pipelines.

What is the logical order in which a SQL SELECT query is executed?

Show answer

Logically the database processes FROM and JOIN first, then WHERE, GROUP BY, HAVING, SELECT, DISTINCT, ORDER BY and finally LIMIT or TOP. This explains common errors: you cannot use a SELECT alias in WHERE because WHERE runs earlier, aggregate conditions must go in HAVING, and window functions cannot be filtered in WHERE, so you wrap them in a CTE or subquery. The optimiser may physically reorder steps, but the result follows this order.

A report query on a 200 million row table takes 15 minutes. How do you optimise it?

Show answer

Read the execution plan to find full scans and expensive joins. Filter early on partitioned or indexed columns such as date, select only needed columns instead of SELECT *, and avoid functions on filter columns. Pre-aggregate in a CTE before joining, replace correlated subqueries with joins or window functions, and check join keys have matching types. For repeated use, build a summary table or materialised view refreshed daily.

Your LEFT JOIN report should list all customers, but customers without orders are missing. Why?

Show answer

A condition on the right table was probably placed in the WHERE clause, for example WHERE o.order_date >= '2026-01-01'. For customers with no orders, o.order_date is NULL, so WHERE removes them and the query behaves like an INNER JOIN. Move the condition into the ON clause, as in LEFT JOIN orders o ON o.customer_id = c.id AND o.order_date >= '2026-01-01', or explicitly allow NULLs.

Write a SQL query for a 7-day rolling average of daily revenue.

Show answer

WITH daily AS ( SELECT order_date, SUM(amount) AS revenue FROM orders GROUP BY order_date ) SELECT order_date, revenue, AVG(revenue) OVER (ORDER BY order_date ROWS BETWEEN 6 PRECEDING AND CURRENT ROW) AS rolling_7d FROM daily ORDER BY order_date; -- fill missing dates from a calendar table first, or gaps skew the average

Write a SQL query for a signup to purchase funnel from an events table.

Show answer

WITH steps AS ( SELECT user_id, MAX(CASE WHEN event = 'signup' THEN 1 ELSE 0 END) AS s1, MAX(CASE WHEN event = 'add_to_cart' THEN 1 ELSE 0 END) AS s2, MAX(CASE WHEN event = 'purchase' THEN 1 ELSE 0 END) AS s3 FROM events WHERE event_date >= '2026-10-01' GROUP BY user_id ) SELECT SUM(s1) AS signups, SUM(s1 * s2) AS carts, SUM(s1 * s2 * s3) AS purchases, ROUND(100.0 * SUM(s1 * s2 * s3) / NULLIF(SUM(s1), 0), 2) AS conv_pct FROM steps;

What is a North Star metric and how do you choose one?

Show answer

A North Star metric is the single measure that best captures the value customers get from the product and predicts long-term revenue. Examples are weekly active learners for an ed-tech app or orders delivered on time for a delivery business. Choose one that reflects customer value, can be influenced by teams and leads revenue. Support it with input metrics such as activation, frequency and retention that teams can work on directly.

What are unit economics and why do investors and leaders focus on them?

Show answer

Unit economics measure revenue and cost for one unit of business, such as one order, one customer or one subscription. For an order you compute average order value minus product cost, delivery, payment fees, discounts and support cost to get contribution margin. If each unit loses money, scale only increases losses. Leaders use unit economics to decide pricing, which cities or channels to grow and where to cut cost.

Marketing and product teams define 'active user' differently and report conflicting numbers. How do you fix this?

Show answer

Bring both teams together and list each definition, its purpose and its numbers. Agree on one company-wide definition tied to real value, such as a user who completed a core action in the last 7 days, and allow clearly named secondary metrics like 'visitors' for marketing. Assign an owner, document it in a metric glossary, implement it once in the warehouse or semantic model, and retire the duplicate dashboards.

The sales team keeps hitting its revenue target, but company margins are falling. What would you recommend?

Show answer

The incentive rewards revenue regardless of profitability, so reps are likely discounting heavily or pushing low-margin products. Analyse deal-level discount, product mix and margin by rep and region to confirm. Then recommend changing the target to gross margin or contribution, or adding discount approval limits and a margin floor to commission. Pilot the new plan with one region and track revenue, margin and win rate together.

A lender is launching a new personal loan product. What KPIs would you set up?

Show answer

Cover the full funnel and risk. Acquisition: applications, approval rate, cost per disbursed loan and time to disbursal. Value: disbursed amount, average ticket size and net interest margin. Risk: delinquency rates at 30 and 90 days past due by cohort, and early default rate. Customer: drop-off by application step, complaint rate and repeat borrowing. Track risk by monthly disbursal cohort, because defaults show up months later.

Tell me about a time you challenged how a metric was defined.

Show answer

Situation: our team reported 'conversion' as orders divided by sessions, and it looked healthy while revenue was flat. Task: understand the gap. Action: I showed that repeated sessions from bots and app refreshes inflated the denominator inconsistently, and proposed user-level conversion excluding internal and bot traffic. I presented both definitions with a month of data to the product head. Result: the new definition was adopted, and it exposed a checkout drop that the team then fixed.

How do you influence stakeholders when you have no direct authority?

Show answer

Build credibility with accurate work and by understanding each stakeholder's goals and pressures. Frame proposals in terms of their outcomes, such as cost saved or targets met, and back them with data. Involve them early so ideas feel shared, find allies who already support the change, and pilot small to show results. Be transparent about trade-offs. Influence comes from trust built over time, not one meeting.

Your analysis contradicts a strongly held belief of the vice president. How do you handle it?

Show answer

Double-check the data and method first, and ask a peer to review it. Then share it privately with the VP before any group meeting, so they are not surprised in public. Present it neutrally: what we expected, what the data shows, and possible reasons. Invite their context, as they may know factors you missed. Suggest a small test to settle it. Focus on the decision, not on being right.

Developers say a requirement is not technically feasible, but the business insists on it. How do you mediate?

Show answer

Get to the underlying business goal behind the requirement, because the goal is often achievable in another way. Ask developers to explain the constraint and the cost of alternatives. Bring both sides together with options: a simpler version, a phased delivery, a manual workaround or a different design. Show effort, risk and value for each, and let the product owner or sponsor decide. Document the decision.

Finance and sales blame each other for wrong numbers in a monthly report. How do you resolve it as the analyst?

Show answer

Stay neutral and move the discussion from blame to facts. Trace the number end to end: source system, extraction, transformation and report logic, and find exactly where the values diverge. Share the findings with both teams together, agree on the fix and on a single owner for the definition. Then add automated reconciliation checks and a data dictionary, so the process, not people, prevents the next issue.

Describe a time you influenced an important decision without having authority.

Show answer

Situation: leadership planned to expand a loyalty programme company-wide. Task: as the analyst, I believed the pilot results were inflated. Action: I showed that pilot stores were already our best performers and built a matched control comparison, which cut the estimated uplift from 15 to 4 percent. I presented it to the sponsor first and proposed a larger randomised pilot. Result: the rollout was redesigned, saving significant cost on a weak programme.

Tell me about a time you led or mentored a team of analysts.

Show answer

Situation: I led three junior analysts on a pricing review with a six-week deadline. Task: deliver quality work while developing them. Action: I split the work by workstream, set up short daily check-ins and a shared query library, reviewed their logic rather than rewriting it, and let each present their section to the client. Result: we delivered a week early, two juniors later handled client calls alone, and the review practice became standard in our team.

You must gather requirements to replace a 15-year-old legacy system with no documentation. How do you approach it?

Requirements and AgileHard

Show answer

Combine several sources. Interview and observe users doing real tasks, and review screens, reports and outputs to infer business rules. Analyse the database and logs to see which data and functions are actually used. Work with long-serving staff and IT support for hidden rules and workarounds. Build AS-IS process maps, decide what to keep, drop or improve, and validate the findings with users. Plan data migration requirements early.

A release passed all system testing, but business users reject it during UAT. What went wrong and what do you do?

Requirements and AgileHard

Show answer

The team built what was written, but not what users needed, which points to gaps in elicitation, missing real-world scenarios or users not involved early. Capture each rejection and classify it as a defect, a missed requirement or a new request. Prioritise blockers with the product owner and agree on a fix plan and date. Going forward, involve users in demos every sprint and base UAT scenarios on real workflows.

Describe a time you led the requirements phase of a large project.

Requirements and AgileHard

Show answer

Situation: our company was replacing its loan origination system across five business units. Task: lead requirements with two other BAs in three months. Action: I set up a stakeholder map, ran structured workshops per unit, used shared templates and a traceability matrix, and held weekly review checkpoints with the sponsor to resolve conflicts quickly. Result: requirements were signed off on time, change requests in build stayed low, and the template was reused for later projects.

How do sample size and minimum detectable effect affect an A/B test design?

Show answer

The minimum detectable effect is the smallest lift worth detecting, such as a 2 percent relative increase in conversion. Smaller effects, lower baseline rates and higher required power need much larger samples, roughly four times the users to detect half the effect. Calculate sample size before starting using the baseline rate, MDE, significance level and power, usually 80 percent, then fix the test duration and avoid stopping early.

What is the novelty effect, and why should A/B tests run for full weeks?

Show answer

The novelty effect is a temporary lift because users are curious about something new, which fades after a few days; the opposite, change aversion, causes a temporary drop. Running tests for at least one or two full weeks covers weekday and weekend behaviour and lets these effects settle. You can also check whether the lift is stable over time and among new users who never saw the old version.

An A/B test shows a 5 percent higher conversion rate, but revenue per user is lower. Should you ship it?

Show answer

Not automatically. Find out why: the variant may push cheaper products, bigger discounts or smaller orders, so more people buy but each spends less. Check average order value, margin and long-term metrics like repeat purchase. Decide based on the metric tied to the business goal, usually revenue or contribution per user, which should have been the primary metric or a guardrail. Consider iterating to keep the conversion gain without lowering order value.

A product manager wants to stop an A/B test after three days because the variant is winning. What do you advise?

Show answer

Advise against it. Checking results repeatedly and stopping at the first significant result inflates false positives, often far beyond the planned 5 percent. Three days also misses weekly patterns and novelty effects. Run the test for the pre-calculated sample size and duration. If speed is critical, the team can use a sequential testing method designed for early looks, but that must be planned before the test starts.

A feature was launched to all users at once without an A/B test. How would you measure its impact?

Show answer

Use quasi-experimental methods. Compare before and after with enough history to account for trend and seasonality. If it rolled out at different times in regions or platforms, use difference-in-differences between early and late groups. Compare users who adopted with similar non-adopters using matching, while being careful about self-selection. For future launches, recommend keeping a small holdout group so impact can be measured cleanly.

Overall conversion fell this month, yet conversion improved for both mobile and desktop users. How is that possible?

Show answer

This is Simpson's paradox caused by a change in mix. If mobile converts lower than desktop and the share of mobile traffic grew sharply, the overall average can fall even though each segment improved. Show conversion by segment along with the traffic mix. The insight is not that conversion got worse but that traffic shifted, so the question becomes how to raise mobile conversion further.

Write a SQL query for Day-7 retention by weekly sign-up cohort.

Show answer

WITH cohort AS ( SELECT user_id, signup_date, DATE_TRUNC('week', signup_date) AS cohort_week FROM users ), ret AS ( SELECT DISTINCT c.user_id FROM cohort c JOIN events e ON e.user_id = c.user_id AND e.event_date = c.signup_date + INTERVAL '7 day' ) SELECT c.cohort_week, COUNT(*) AS users, ROUND(100.0 * COUNT(r.user_id) / COUNT(*), 1) AS d7_retention_pct FROM cohort c LEFT JOIN ret r ON r.user_id = c.user_id GROUP BY c.cohort_week ORDER BY c.cohort_week;

Tell me about an experiment that failed and what you learned.

Show answer

Situation: we tested a simplified pricing page expecting higher paid conversion. Task: I owned the experiment design and analysis. Action: after two weeks the variant showed no significant change, and segment checks revealed that heavy users missed the plan comparison table. I documented the result and shared it openly rather than searching for a winning segment. Result: the next version kept the simple layout with an optional comparison, and we learned to interview users before testing.

How do you use NPV and payback period to evaluate an investment in a case?

Show answer

Estimate the upfront investment and yearly incremental cash flows, then discount future cash flows at the cost of capital to compute Net Present Value. A positive NPV means the investment creates value. Payback period is how many years until cumulative cash flows recover the investment, which shows risk and liquidity but ignores later cash flows and time value. Present both, with a sensitivity check on key assumptions like volume and price.

Should a supermarket chain launch its own private-label products?

Show answer

Evaluate attractiveness and ability. Attractiveness: private labels usually earn higher margins, so check category size, how price-sensitive customers are, and whether brand loyalty is weak in categories like staples and cleaning. Ability: supplier availability, quality control, shelf space and brand trust. Estimate the margin gain against cannibalisation of branded sales and supplier relationships. Recommend starting with a few high-volume, low-loyalty categories and measuring repeat purchase.

A mid-size manufacturer must choose between building and buying a warehouse management system. How do you advise?

Show answer

Compare options on fit, cost, time and risk. Buying a proven product is faster with lower risk and includes upgrades, but may need process changes and has recurring licence fees. Building fits unique processes exactly but needs a strong tech team, longer time and ongoing maintenance. Estimate five-year total cost of ownership, check whether processes are truly unique, and usually recommend buying and configuring unless warehousing is a core differentiator.

Guesstimate: what is the annual revenue of one multiplex screen in a large Indian city?

Show answer

Assume 5 shows a day and 200 seats with 30 percent average occupancy, giving 300 tickets a day. At an average ticket of 250 rupees, ticket revenue is 75,000 rupees a day. Food and beverage adds about 40 percent of ticket revenue, around 30,000, and screen ads maybe 5,000. That is about 1.1 lakh a day, or around 4 crore a year. Weekends and blockbuster releases push occupancy higher.

What is an event-based gateway and when would you use it?

Show answer

An event-based gateway is a decision point where the path is chosen by whichever event happens first, not by a data condition. It is followed by intermediate catching events such as a message or a timer. Example: after sending a quotation, either the customer's acceptance message arrives, or a 7-day timer fires and a reminder is sent. It models real waiting situations that an exclusive gateway cannot express correctly.

What is a boundary event in BPMN, and how do interrupting and non-interrupting boundary events differ?

Show answer

A boundary event is attached to the edge of an activity and reacts to something that happens while that activity is running. An interrupting boundary event (solid double border) cancels the activity and takes the exception path, for example a 24-hour timeout cancelling manual approval. A non-interrupting one (dashed border) starts a parallel path while the activity continues, for example sending a reminder email after 12 hours without stopping the approval task.

What is value stream mapping and how is it different from BPMN?

Show answer

Value stream mapping is a Lean technique that maps the end-to-end flow of material and information to deliver value to a customer, along with timing data such as process time, wait time and inventory between steps. Its focus is finding waste and calculating lead time. BPMN focuses on precise logic: roles, decisions, events and messages, often detailed enough for automation. Many teams use VSM to find where to improve and BPMN to design the new process.

What is process mining and how does it help a business analyst?

Show answer

Process mining uses event logs from systems such as ERP or CRM, each record holding a case ID, activity name and timestamp, to automatically rebuild how a process really runs. It shows actual paths, rework loops, bottlenecks and how often cases deviate from the designed flow. For a BA it replaces opinion-based AS-IS workshops with evidence, for example showing that 22 percent of purchase orders are changed after approval. Data quality of the logs is the main limitation.

What is DMN and how does it relate to BPMN?

Show answer

DMN (Decision Model and Notation) is an OMG standard for modelling business decisions, mainly through decision tables that list input conditions, outputs and a hit policy. BPMN shows the flow of work, while DMN holds the decision logic. Instead of drawing ten nested gateways for loan eligibility, you place one business rule task in BPMN that calls a DMN table. This keeps diagrams clean and lets business users change rules without redrawing the process.

Leadership wants to automate an invoice approval process with RPA. What do you check in the process model before agreeing?

Show answer

Check that the process is stable, rule-based and high volume, with structured digital inputs. Look at the variant rate: if half the cases are exceptions handled by judgement, a bot will fail often. Verify that the underlying systems and screens will not change soon. Fix the process first, because automating a broken flow only makes errors faster. Finally estimate hours saved against bot build and maintenance cost, and define how exceptions get routed to people.

Your TO-BE process cuts the account-opening cycle time by half, but the compliance team rejects it. How do you handle it?

Show answer

Ask compliance which specific controls they believe are weakened, and map each regulatory requirement to steps in the old and new flows. Often the control can be kept in a different form, such as automated checks, risk-based sampling or a post-approval audit instead of a manual pre-check. Bring compliance into the redesign rather than presenting a finished model. Document the agreed controls in the TO-BE diagram so audit can trace them. Speed never overrides a mandatory control.

Describe a BPMN model for a purchase order process where credit check and stock check happen at the same time.

Show answer

Start event: purchase order received. Task: validate order details. A parallel gateway then splits into two paths: Finance lane runs Check customer credit, Warehouse lane runs Check stock. A second parallel gateway joins them, waiting for both to complete. An exclusive gateway then decides: credit OK and stock available goes to Confirm order and Schedule dispatch; otherwise go to Notify customer with reason. Using a parallel join here is correct because both results are always needed.

Describe a time when stakeholders resisted a process change you proposed. How did you handle it?

Show answer

Situation: I proposed removing a manual second approval for low-value expense claims, and team leads resisted, fearing fraud. Task: get agreement without weakening control. Action: I analysed six months of claims and showed the second approver rejected less than 0.5 percent below 5,000 rupees. I suggested a pilot with random audits of 10 percent of claims and agreed a reversal trigger. Result: the pilot ran for two months with no extra fraud, approval time fell by 3 days and the change was adopted.

Your A/B test was meant to split traffic 50/50, but the actual split is 52/48 on 200,000 users. What do you do?

Show answer

This is a sample ratio mismatch. With 200,000 users, a 52/48 split is far outside random chance, which a chi-square test will confirm. It means the assignment or logging is broken, for example the variant page crashing before the tracking event fires, bots filtered unevenly or a redirect losing users. Do not trust the results. Investigate by segment, platform and day to find the cause, fix it and rerun the test.

What is IRR and how is it different from NPV?

Show answer

IRR (internal rate of return) is the discount rate at which a project's NPV becomes zero, so it expresses the return as a percentage. NPV gives the value created in rupees at a chosen discount rate. Accept a project if IRR exceeds the cost of capital or NPV is positive. When comparing projects of different size or timing, NPV is more reliable, because IRR ignores scale and can give multiple answers when cash flows change sign more than once.

What is operating leverage and why does it matter in business analysis?

Show answer

Operating leverage describes how much of a company's cost base is fixed. With high operating leverage, such as software or airlines, a small rise in revenue produces a large rise in operating profit once fixed costs are covered, but a fall in revenue hurts profit just as sharply. It is measured as percentage change in EBIT divided by percentage change in sales. Analysts use it to judge risk, set break-even targets and stress-test forecasts.

Vendor A has a lower licence price, but Vendor B has lower running costs. How do you compare them?

Show answer

Compare total cost of ownership over the same period, usually 3 to 5 years. Include licence or subscription fees, implementation, data migration, integration, training, internal staff time, infrastructure, support and upgrade costs, and exit costs. Discount the yearly totals to present value if the timing differs. Then weigh the result against non-cost factors such as fit to requirements, scalability and vendor stability. Often the cheaper licence ends up costlier once customisation and support are added.

What is the difference between zero-based budgeting and incremental budgeting?

Show answer

Incremental budgeting takes last year's budget and adjusts it by a percentage, which is quick but carries forward old inefficiencies. Zero-based budgeting starts every cost line from zero, and each activity must be justified by its value each cycle. ZBB uncovers waste and suits cost-reduction programmes, but takes much more effort and can create friction. Many companies apply ZBB to discretionary costs such as marketing and travel and use incremental methods elsewhere.

A product line shows a loss after allocated overheads. Leadership wants to shut it down. How do you advise?

Show answer

Separate contribution from allocation. If the product's revenue exceeds its variable and directly attributable fixed costs, it contributes to covering shared overheads, which will not disappear if it is closed and will be reallocated to other products. Check whether the overhead allocation key, such as revenue share, is fair. Also consider strategic links, like customers who buy it alongside profitable products. Recommend shutting it only if contribution is negative or the freed capacity has a better use.

The CFO asks whether the company should lease or buy 50 delivery vans. How do you approach the analysis?

Show answer

Build a discounted cash flow comparison over the vehicle life. For buying: purchase price, financing cost, maintenance, insurance, tax benefit of depreciation, and resale value. For leasing: lease rentals, what maintenance is included, tax treatment and end-of-term conditions. Discount both at the company's cost of borrowing and compare present value of costs. Then add qualitative factors: flexibility if volumes drop, technology risk such as a shift to electric vans, and impact on the balance sheet.

A project's business case promised 30 percent cost savings, but after go-live only 10 percent is realised. What do you do?

Show answer

Run a benefits review against the original baseline and assumptions. Check whether savings depended on actions that did not happen, such as reducing manual roles, retiring the old system or full user adoption. Check whether the baseline or volumes changed. Separate delayed benefits from lost ones. Then produce a recovery plan with owners, for example decommissioning the legacy licence or driving adoption. Also feed the lessons into how future business cases estimate benefits.

What is the difference between bookings, billings and revenue?

Show answer

Bookings are the value of contracts signed in a period, showing future commitment. Billings are the amounts invoiced to customers in the period. Revenue is what is earned and recognised under accounting rules as the service is delivered. Example: a 3-year contract worth 36 lakh signed in April is 36 lakh in bookings, 12 lakh in billings if invoiced annually, and only 1 lakh in revenue for April. Confusing them is common when reporting sales performance.

What are MRR and ARR, and how do you calculate net revenue retention?

Show answer

MRR is monthly recurring revenue from active subscriptions; ARR is the annualised figure, usually MRR times 12. Net revenue retention measures revenue kept from an existing customer cohort over 12 months: (starting MRR + expansion - contraction - churned MRR) / starting MRR. Example: 100 lakh start, plus 15 expansion, minus 5 contraction, minus 8 churn gives 102 percent. Above 100 percent means existing customers grow revenue even without new sales.

Guesstimate: how many new two-wheeler helmets are sold in India each year?

Show answer

Drive it from vehicles. Assume about 2 crore new two-wheelers are sold a year, each with an average of 1.5 helmets bought (rider plus pillion), giving 3 crore. The existing fleet is around 25 crore vehicles; assume 60 percent are active and each active vehicle replaces one helmet every 3 years: 15 crore / 3 = 5 crore. Total is about 8 crore helmets a year. Sanity check: enforcement varies by state, so give a range of 6 to 9 crore.

How do you plan test data for UAT?

Show answer

Start from the test cases and list the data each needs: customer types, product variants, edge values and historical records. Decide whether to use masked copies of production data, synthetic data or both, because real personal data must be anonymised under data protection rules such as the DPDP Act. Prepare data for negative and boundary cases, keep a refresh plan for repeated cycles, and assign owners. Poor test data is one of the most common causes of blocked UAT days.

What is a cutover plan and a go-live readiness checklist?

Show answer

A cutover plan is an hour-by-hour runbook for moving from the old system to the new one: freezing old transactions, final data migration, validation checks, configuration, smoke testing, switch-over and a rollback plan with clear decision points. A go-live readiness checklist confirms that UAT is signed off, users are trained, support and hypercare are staffed, data is reconciled, and communications are sent. The go or no-go meeting reviews this checklist before the cutover starts.

What is operational acceptance testing and how is it different from UAT?

Show answer

Operational acceptance testing (OAT) checks whether the operations and support teams can run the system in production: backup and restore, failover, monitoring and alerts, batch job scheduling, performance under expected load, security patching and support procedures. UAT checks whether business users can do their work. Both are needed before go-live. A system can pass UAT and still fail in production if nobody tested that the nightly batch finishes before the business opens.

A critical defect is found on the last day of UAT, and the go-live date cannot move. What do you do?

Show answer

Confirm the impact and reproduce it quickly. Work with development to estimate a fix and its regression effort. Then present options to the business owner with risks: fix and do targeted retesting before go-live, go live with a temporary workaround or manual process, go live with the affected feature switched off, or delay. The decision belongs to the business, documented at the go or no-go meeting. Never hide a critical defect to protect a date.

Because of a tight deadline, the business sponsor wants to skip UAT and go live directly. How do you respond?

Show answer

Explain the risks in business terms: without UAT, there is no evidence that real processes work, and production defects cost far more to fix and can affect customers and compliance. Offer alternatives: a shortened UAT covering only critical business flows, parallel running with the old system, a phased rollout to one branch or user group, or extended hypercare. If the sponsor still decides to skip it, record the decision and accepted risks formally.

Write SQL checks to validate a customer data migration between the legacy and new systems during UAT.

Show answer

-- 1. Row counts should match SELECT (SELECT COUNT(*) FROM legacy.customers) AS legacy_rows, (SELECT COUNT(*) FROM target.customers) AS target_rows; -- 2. Records missing in the target SELECT l.customer_id FROM legacy.customers l LEFT JOIN target.customers t ON t.customer_id = l.customer_id WHERE t.customer_id IS NULL; -- 3. Financial totals should reconcile SELECT SUM(outstanding_balance) FROM legacy.customers; SELECT SUM(outstanding_balance) FROM target.customers;

Describe a time you recommended delaying a go-live.

Show answer

Situation: two days before the go-live of a loyalty points system, UAT showed points were doubled for refunded orders. Task: give a clear recommendation at the go or no-go meeting. Action: I presented the defect, estimated financial exposure at about 6 lakh rupees a month and showed the fix needed three days including regression. Result: the sponsor agreed to a one-week delay. The release went live cleanly, and the sponsor later cited the decision as an example of good risk management.

Write a SQL query to calculate monthly customer churn rate for a subscription business.

Show answer

WITH monthly AS ( SELECT DISTINCT customer_id, DATE_TRUNC('month', billing_date) AS month FROM subscriptions_billed ) SELECT prev.month AS month, COUNT(prev.customer_id) AS customers_at_start, COUNT(prev.customer_id) - COUNT(nxt.customer_id) AS churned, ROUND(100.0 * (COUNT(prev.customer_id) - COUNT(nxt.customer_id)) / COUNT(prev.customer_id), 2) AS churn_pct FROM monthly prev LEFT JOIN monthly nxt ON nxt.customer_id = prev.customer_id AND nxt.month = prev.month + INTERVAL '1 month' GROUP BY prev.month ORDER BY prev.month;

A query using NOT IN to find customers without orders returns zero rows, although you know such customers exist. Why?