298 questions from 39 interviews · updated from reports 2017-2024
About
ZS Associates is a consulting and technology firm that helps companies use data, analytics, and software, mainly in healthcare and life sciences. In India, it hires Business Technology Analysts, Business Technology Solutions Associates, and related data analytics, technology consulting, and software roles.
The roles that come up most are Business Technology Analyst, Business Technology Solutions Associate and BTSA. This covers 39 candidate interviews reported from 2017 to 2024. Most sat it at entry level (38 of 39 that recorded a level). Among the 36 that recorded either route, arrivals split between campus drives (31, 86%) and off-campus applications (5, 14%). Most questions fall under CS fundamentals and Aptitude.
Ans. Read the title, units, axes, labels and time period first. Identify exactly what is being asked, then locate only the relevant values. Use ratios, percentages, differences or averages as needed. Watch for scale changes, rounded figures and combined categories. Estimate only when options allow it, and check whether the answer is reasonable.
Ans. There is not enough information to give a numerical answer. The method is to convert each person or machine’s speed into “task per unit time”, add the rates when they work together, and use time equals remaining work divided by combined rate. If conditions change, split the work into stages and track the fraction completed.
Ans. The minimum time is 17 minutes. Send 1 and 2 across, taking 2. Bring 1 back, total 3. Send 5 and 10 across, total 13. Bring 2 back, total 15. Send 1 and 2 across again, total 17. The key is using the two fastest people to shuttle the torch.
Ans. Estimate it by approximating the city boundary as a simple shape, usually a circle or rectangle, then calculate the area. Use known or guessed dimensions, such as the distance across the city north to south and east to west. Adjust for irregular edges, water, parks or suburbs, then state your assumptions clearly.
Ans. I would first identify the required output columns, source tables, join keys, filters, grouping, and ordering, then form the query using SELECT, FROM, JOIN, WHERE, GROUP BY, HAVING, and ORDER BY as needed. The key detail is applying filters before aggregation with WHERE and after aggregation with HAVING.
Ans. About one million trees for a city of one million people. I would estimate land area first, say 500 square kilometres. If 20% is parks, gardens and street verges, that is 100 square kilometres of tree-bearing land. At an average of 10,000 trees per square kilometre, the estimate is about 1,000,000 trees.
Ans. React.js is a UI library focused on building components, while Angular.js is a full front-end framework with routing, dependency injection, forms, and application structure built in. The key practical difference is control: React gives more flexibility and needs extra libraries, while Angular.js is more opinionated and uses two-way data binding.
Ans. Start by identifying entities from the database tables, then draw them as rectangles with primary keys and main attributes. Connect related entities using relationship lines, marking cardinality such as one-to-one, one-to-many, or many-to-many. The key detail is to show primary keys, foreign keys, and participation clearly so the schema relationships are unambiguous.
Ans. I would build the query by selecting the required columns, joining the relevant tables on their key relationships, applying filters in the where clause, then using group by, having, order by, or limit as needed. The important detail is using indexed join and filter columns, so performance avoids full table scans where possible.
Ans. About 4 million tennis balls fit in a large passenger airplane, such as a Boeing 747. Estimate usable internal volume at around 1,000 cubic metres. A tennis ball is about 6.7 cm across, with volume 0.000157 cubic metres. Allowing 64% packing efficiency, 1,000 divided by 0.000245 gives roughly 4 million.
Ans. Around 10 million footballs are sold in India each year. Start with about 300 million people aged 5 to 25. If 15% play football at least occasionally, that is 45 million players. Assuming one ball is shared by five players and replaced yearly gives 9 million, plus schools and clubs, so about 10 million.
Ans. I would first restate the rules, define the unknowns, and look for constraints that eliminate possibilities. I would test simple cases, keep track of assumptions, and check for contradictions. If one path fails, I backtrack and explain why. I would give the final answer only after verifying it satisfies every condition.
Ans. An index in SQL is a data structure, usually a B-tree, that speeds up searching, sorting and joining rows by keeping ordered references to table data. Common types include clustered, non-clustered, unique, composite, primary key, full-text and spatial indexes. Indexes improve reads but add storage cost and slow inserts, updates and deletes.
Ans. You should switch. There are three doors, one prize and two goats. Your first pick has a 1 in 3 chance of winning, so the other two doors together have 2 in 3. Monty knows the doors and opens a goat among those two. That 2 in 3 probability moves to the remaining unopened door.
Ans. Precompute, for every cell, how many valid border cells extend right and down, then try square sizes from largest to smallest. For a top left cell and size k, the square is valid if top and bottom have k right-counts and left and right have k down-counts. This uses two integer matrices and runs in O(nm min(n,m)).
Ans. About 50,000 street lights for a city of one million people. Assume 400,000 homes, around 2.5 people each. If streets serve about 40 homes per kilometre, that gives 10,000 kilometres of streets. With lights every 30 to 40 metres on alternating sides, estimate about 5 lights per kilometre, giving roughly 50,000.
Ans. Use SELECT to retrieve data, with WHERE, JOIN, GROUP BY and ORDER BY to filter, combine, aggregate and sort results. Use INSERT, UPDATE and DELETE to manipulate rows, ideally inside transactions when changes must be atomic. The key detail is to target rows precisely, especially for UPDATE and DELETE, to avoid unintended changes.
Ans. Use a generic sort function that accepts a collection of elements and a comparator defining their order. Store the elements in an array or list, then apply a comparison-based algorithm such as merge sort or quicksort. The key detail is that sorting any data type requires a comparator. Average time complexity is O(n log n).
Ans. I would identify the required output columns, the tables that contain them, and the relationships between those tables. Then I would join on primary and foreign keys, apply filters in the where clause, aggregate with group by when needed, and use order by or limits only after the result set is correct.
Ans. About 3 million litres of petrol per day, or roughly 1 billion litres a year. I would estimate 20 million people, about 4 million vehicles, with maybe 3 million using petrol. If each averages 20 km daily at 20 to 25 km per litre, consumption is around 2.5 to 3 million litres daily.
Ans. SQL does not have Windows commands as part of the language, but Windows-based SQL tools often use commands such as sqlcmd to run queries, bcp to import or export data, and net start or net stop to control SQL Server services. xp_cmdshell can run Windows shell commands, but it is usually disabled for security.
Ans. Make two straight vertical cuts through the centre at right angles, dividing the cake into four equal quarters. Then make one horizontal cut halfway up the cake, parallel to the base, through all four quarters. Each quarter is split into two equal layers, giving 4 × 2 = 8 equal pieces.
Ans. Estimate it by breaking the city into road junctions. Take the city area, estimate major-road density, then estimate how many junctions per kilometre are signalised. For example, 300 square kilometres, 4 km of main road per square kilometre, one signal every 0.8 km gives about 1,500 traffic signals.
Ans. There is no single answer without a specific puzzle, so I would solve it by defining the facts, listing constraints, removing impossible cases, and checking what remains. For example, in truth-teller puzzles I assign true or false to each statement, test consistency, and keep only the arrangement that satisfies every condition.
Ans. Networking fundamentals cover how data moves between devices using layered protocols, addressing, routing and reliable delivery. The key detail is understanding the TCP/IP model: IP handles addressing and routing, TCP adds reliable ordered delivery, UDP favours speed, and DNS maps names to IP addresses.
Ans. Use joins when you need columns from related tables, and use subqueries when you need to filter or aggregate before the main query. Start from the required result, identify keys, choose inner or outer joins carefully, and push filtering into WHERE or HAVING depending on whether it happens before or after aggregation.
Ans. Start with the business objective, then break the problem into simple drivers such as market size, users, frequency, price, cost, and margin. State clear assumptions, use round numbers, calculate step by step, and sanity check the result against real-world benchmarks. Highlight the key sensitivity and explain what data would improve accuracy.
Ans. Use expand around centres: for each index, expand once for an odd-length palindrome and once between indices for an even-length palindrome, tracking the best start and length. The key detail is handling both centre types. This uses only a few variables, runs in O(n squared) time, and uses O(1) extra space.
Ans. Read the paragraph for its main point, then separate stated facts from assumptions. Note keywords showing cause, contrast, quantity, or certainty. Test each option against the passage only, not outside knowledge. Choose the conclusion that must follow or is best supported, and reject choices that exaggerate, contradict, or introduce new information.
Ans. Choose a real situation where your behaviour set a visible standard, such as staying calm under pressure, owning a mistake, supporting a junior colleague, or following values when it was difficult. Emphasise the example you set, how others responded, and the positive outcome. Interviewers listen for consistency, humility, influence, and integrity.
Ans. A machine learning model should be retrained with fresh, representative data when its performance drops, the data distribution changes, or new labelled data becomes available. The key is to monitor production metrics and data drift continuously, then retrain, validate on a holdout set, compare against the current model, and deploy only if it improves safely.
Ans. Use faster hardware, parallel or distributed training, mini-batching, efficient data pipelines, transfer learning, early stopping, smaller models, feature selection, and hyperparameter search methods such as random or Bayesian search. The most important detail is to remove bottlenecks first, often data loading or overly large models, rather than only adding compute.
Ans. Clarify what is fixed, what is missing, and what assumptions are allowed. Convert the data into equations, estimates, or bounds rather than trying to calculate everything exactly. Use rounding to check scale, test edge cases, and state any assumption clearly. If multiple answers are possible, give the range or the most reasonable interpretation.
Ans. Database indexing creates an additional data structure, often a B-tree or hash index, to find rows faster without scanning the whole table. Database partitioning splits a large table or index into smaller physical parts, usually by range, list, or hash. Indexes speed reads but cost writes; partitioning improves manageability and targeted queries.
Ans. Use a window function such as DENSE_RANK over salaries ordered descending, then select rows where the rank equals n. Use DENSE_RANK if duplicate salaries should share the same rank, or ROW_NUMBER if every row must be ranked separately. The database typically sorts the salary column, giving about O(m log m) time for m rows.
Ans. Use Excel’s Data or Power Query tools to connect to the database, write the required SQL SELECT queries, and load the results into a worksheet or pivot table. The key is to translate each business scenario into filters, joins, grouping, and aggregations, then validate totals against the source data.
Ans. Read the title, units, labels and time period first, then identify exactly what is being asked. Extract only the needed values from the chart or table. Use the correct operation, such as difference, percentage change, ratio, average or total. Estimate only when allowed, and check that your answer matches the units.
Ans. About 6,000 red traffic signals in a city of one million people. I would estimate 2,000 signalised junctions, based on roughly one major controlled junction per 500 residents. Each junction has about three red signal heads per approach across four approaches, so 12 heads. Half may be smaller junctions, giving about 6,000 overall.
Ans. Merge Sort splits the array, sorts each half, then merges them, while Quick Sort partitions around a pivot and recursively sorts the partitions. Merge Sort is stable and always O(n log n) but needs extra space. Quick Sort is usually faster in practice and in-place, but can degrade to O(n²) with poor pivots.
Ans. Select the best model by comparing candidates against the right metric on validation data, not by training accuracy. Use cross validation where possible, tune hyperparameters systematically, check bias and variance, and keep a separate test set for final evaluation. Optimal performance also depends on clean data, good features, regularisation, and monitoring after deployment.
Ans. Join the related tables on their key columns, group the joined rows by the required dimension, and apply aggregate functions such as count, sum, average, minimum, or maximum to each group. Use where to filter rows before grouping and having to filter groups after aggregation. Ensure non-aggregated selected columns appear in the group by clause.
Ans. Estimate route length, stations, depot size, rolling stock and land needs, then apply unit costs. For Delhi, use benchmarks such as cost per kilometre for elevated and underground track, cost per station, trains, signalling, power and contingencies. Multiply by quantities, adjust for inflation, land acquisition and project risk.
Ans. Bias can arise when text data overrepresents some groups, viewpoints, dialects, topics, or historical stereotypes, causing models to learn unfair associations. Mitigation involves auditing data, improving sampling balance, removing or labelling harmful content, using debiasing techniques, testing across demographic groups, and documenting dataset sources and limitations.
Ans. I would first clarify the input, output, constraints, and edge cases, then choose an algorithm that fits those constraints and implement it in Java or JavaScript. The key detail is matching the data structure to the problem, such as arrays, hash maps, stacks, queues, or heaps, and explaining the time and space complexity clearly.
Ans. The wall area is 400 square feet, so paint required depends on the paint’s coverage and number of coats. If one litre covers 100 square feet for one coat, one coat needs 4 litres. For two coats, it needs 8 litres, plus a little extra for wastage or rough surfaces.
Ans. Start by identifying entities, their attributes, and relationships, then create tables where each fact is stored once. Use primary keys, foreign keys, unique constraints, and junction tables for many-to-many relationships. Aim for third normal form: no repeating groups, no partial dependency, and no transitive dependency. Denormalise only for proven performance needs.
Ans. About £20,000 for the paint materials for a narrow-body airliner. Approximate the aircraft as 700 to 800 square metres of external surface. With primer, base coat and clear coat, that is about 2,400 square metres of coverage. At roughly 8 square metres per litre, use 300 litres. At £60 per litre plus wastage, cost is about £20,000.
Ans. Start with the recommendation, then support it with the key numbers, assumptions, risks and trade-offs. Break the case into objective, data, analysis, options and decision. For multiple questions, answer each directly, link it back to the business goal, and state any missing information that could change the conclusion.
Ans. Pick a real deadline where the stakes were clear and you personally influenced the outcome. Emphasise how you prioritised, communicated trade-offs, stayed organised, and protected quality. Interviewers listen for calm decision making, ownership, teamwork, and evidence that you delivered or learned something useful rather than simply working long hours.
Ans. Fill the 9-litre bucket and pour into the 4-litre bucket twice, emptying the 4-litre bucket each time. This leaves 1 litre in the 9-litre bucket. Pour that 1 litre into the 4-litre bucket. Fill the 9-litre bucket again, then pour 3 litres into the 4-litre bucket, leaving exactly 6 litres.
Ans. Identify what the table, chart, or graph is measuring, then note the units, totals, and time periods. Extract only the figures needed, convert units if required, and apply the correct formula for percentage, ratio, average, growth, or difference. Estimate first to spot errors, then calculate carefully and compare options.
Ans. Start by identifying the base table, required joins, filters, grouping, and output columns, then build the query incrementally and validate row counts at each step. Use primary and foreign key relationships to avoid duplicate rows, apply aggregation only after joins are correct, and check indexes on join and filter columns for performance.
Ans. Use a SELECT query that names the required columns, reads from the relevant tables with FROM, joins related tables on their key columns, filters rows with WHERE, groups with GROUP BY when aggregating, and orders with ORDER BY if needed. SQL uses table indexes where available, and performance mainly depends on joins, filters, and indexing.
Ans. Use GROUP BY with COUNT to aggregate rows, JOIN to combine related tables on key columns, and UNION or UNION ALL to combine result sets with matching columns. UNION removes duplicates, while UNION ALL keeps them and is usually faster. Always group by every non-aggregated selected column and join on indexed keys where possible.
Ans. Break the problem into simple drivers, make clear assumptions, estimate each driver, then multiply or add to reach a rough answer. Use round numbers, state units, and sanity check the result against real-world intuition. For example, estimate market size as population times relevant percentage times usage frequency times price.
Ans. Use SQL by selecting the required columns, filtering rows with WHERE, combining tables with JOIN, grouping with GROUP BY, and changing data with INSERT, UPDATE, or DELETE. The key detail is indexing: queries usually scan tables without indexes, but indexed filters or joins can reduce lookup cost from linear to logarithmic or near constant.
Ans. Pick a real example where the problem could not be solved by one team alone, such as a delivery issue, customer incident, or process breakdown. Emphasise your role in aligning people, clarifying ownership, managing trade-offs, and keeping communication clear. Interviewers listen for collaboration, influence without authority, accountability, and a measurable outcome.
Ans. Build it by loading notebook content, splitting it into chunks, embedding those chunks, storing them in a vector index, retrieving the most relevant chunks for a user query, and passing them with the query to an LLM. The key detail is preserving notebook structure, such as markdown, code cells, outputs, and execution order.
Ans. I would package the trained model as a versioned service, expose it through an API, and deploy it via CI/CD to a controlled environment such as Kubernetes or a managed ML platform. The most important consideration is reliable monitoring, covering latency, errors, data drift, model performance, rollback, security, and reproducibility.
Ans. A login page sends the user’s credentials from the frontend to the backend, which validates them against user data in the database and returns an authenticated session or token. The key detail is that passwords are never stored or compared as plain text. The backend hashes the submitted password and compares it with the stored password hash.
No question matches those filters. Reset filters.
Showing 60 of 298 questions. Ranked by how often the same question came back across interviews.
A spoken interview built from these questions, scored when you finish; the feedback is yours.
Start practisingWhen you are ready, record The One: a single interview hiring teams watch, so you stop repeating first rounds.
Candidate interviews most often cover CS fundamentals (36%) and Aptitude (27%).
Candidate interviews show an average of 4.4 rounds per experience, with a typical sequence of Online test → Technical → Technical. Individual interview paths can vary.
Among questions with a recorded difficulty, the mix is easy 33%, medium 62%, hard 5%.
