Explore: Interview experience · HR interview questions · Overview · Placement papers · Preparation guide
Fractal Analytics hires freshers into its Imagineer program through resume screening, timed online tests (in 2021 reports, 70 aptitude questions in 75 minutes and then 8 SQL and Python or R questions in 90 minutes), then two to three interview rounds - technical, Apex or business understanding, and HR - followed by a multi-week pre-onboarding training period before the formal offer or LOI.
Candidate-reported campus filters vary by college: 7.0+ CGPA (some drives 8.0+, e.g. multi-campus VIT Imagineer 2026), circuit B.E./B.Tech (CS, IT, ECE, EEE, AI, etc.) plus M.Sc/M.A Economics or Statistics on some drives, no active backlogs, and 2026 pass-out on referral postings. After selection, Imagineers typically rotate exposure across four tracks before specialization - Data Engineering, Data Science, Decision Science, and Data Visualization - with track placement driven by pre-onboarding performance and business need (public campus job descriptions, 2026).
Interview experience
Fractal Analytics interview experiences covering Online Test I, Online Test II, Technical / Problem Solving Round, Business Understanding Round, +1 more. Round-wise notes, common questions, and what candidates report clears each stage.
HR interview questions
Fractal Analytics HR interview questions with sample answer frameworks - motivation, relocation, STAR stories, and culture-fit prompts.
| Round | Duration | What they test |
|---|---|---|
| Online Test I | ~75 min (2021 reports) | Data analysis, quantitative ability, reasoning, verbal (~70 questions) |
| Online Test II | ~90 min (2021 reports) | 4 SQL + 4 Python or R questions |
| Technical / Problem Solving Round | 30-45 min | Case studies (market entry, profitability), project discussion |
| Business Understanding Round | 30 min | Guesstimates, probability, structured thinking |
| HR Round | 20-30 min | Strengths, values, why Fractal |
These answers are compiled from public candidate reports and campus job posts, not from an official Fractal notice, and drives differ. The instruction screen of your own test is the final word.
| Question | Answer |
|---|---|
| How many questions and how long? | Two 2021 GfG campus reports describe 70 aptitude questions in 75 minutes (data analysis, quant, reasoning, verbal), then 8 coding questions in 90 minutes (4 SQL, 4 Python or R, language chosen before the test). A later Glassdoor report cites 75 questions in 75 minutes, and a 2024 report describes 12 Python and SQL MCQs plus 2 SQL queries. One unconfirmed 2026 campus post describes aptitude, reasoning and communication items merged into one screen. |
| Does it have negative marking? | Not published, and reports do not settle it. Check the instruction screen. |
| Is there a sectional cutoff? | Not published. Reports conflict on weight: one 2021 report says shortlisting used the combined result of both tests, another says the coding test carried little weight. Do not rely on aptitude alone. |
| How much is the cutoff? | No score cutoff is published. 2026 campus job posts commonly ask for 7.0+ CGPA (8.0+ on some drives) and no active backlogs; the GfG reports list no criteria. |
| Which platform is it on? | No public source reliably names one; your invite mail carries the link. |
| Is there coding? | Yes. 2021 reports describe SQL questions built mostly on joins and Python questions on NumPy and Pandas, written in an editor with sample test cases. A Feb 2026 report describes two online coding rounds (coding plus SQL) before the behavioural and HR rounds. |
| Is the offer immediate after HR? | No. Imagineers report a 10-12 week pre-onboarding training period between selection and the final offer letter. |
Questions below are ones candidates report for Fractal’s Imagineer rounds and job posts. They show the level and style, not a fixed list.
| Track | Reported questions and topics |
|---|---|
| Imagineer - Decision Science and Data Science | Profitability and market-entry cases, guesstimates (smartphones sold in India in a year, Domino’s pizza volume, cricket balls in an aeroplane, Zomato orders in your area), probability of at least one six in four rolls, correlation vs causation, p-values. |
| Imagineer - Data Engineering | UNION vs UNION ALL (2024 report), BFS vs DFS and what code versioning is (May 2025 report), Python list vs tuple, second-largest number without sorting, top-selling product per category with a window function, SQL joins and aggregates, plus a project walk-through. |
| Experienced engineering roles (for example front end) | A take-home assignment, then onsite rounds on framework internals and two DSA problems in 45 minutes, closing with a senior leader on pressure handling. |
Asked across every track: tell me about yourself and why Fractal, why analytics over pure software development, what other offers you hold, and whether you are comfortable with the pre-onboarding training.
2021 GfG campus accounts describe 70 questions in 75 minutes across data analysis, quantitative ability, reasoning, and verbal sections. One unconfirmed 2026 campus post describes aptitude, logical reasoning and communication-style items combined into one screen before interviews. Off-campus Women-in-Tech style drives may split quant, data insights, verbal, and personal sections across timed buckets. This is the broad filter before SQL/Python depth, and given Fractal’s low acceptance rate, speed and accuracy both matter.
Common questions
A technical test of 8 questions in 90 minutes in 2021 reports: 4 SQL questions (mostly joins) and 4 Python or R questions (NumPy and Pandas), with the language chosen before the test.
Common questions
Combines a walkthrough of your resume projects (especially any analytics, ML, or data project) with business case questions in the market-entry or profitability-estimation style. Interviewers evaluate how you structure an ambiguous business problem, not just your final answer.
Common questions
Campus job posts label one round Tech / Apex (Managerial); candidate reports map this to guesstimates, probability, and resume-linked problem attitude. VIT Chennai 2026 technical panels asked airplane cricket-ball and Domino’s pizza volume estimates; off-campus accounts asked Zomato orders in your area with explicit instruction to segment population and cuisine, not blurt a number. Fractal scores structure and assumptions more than precision.
Common questions
Full round-by-round accounts are on the Fractal Analytics interview experience page.
A closing 20-30 minute conversation on strengths, values alignment, and motivation for choosing Fractal and an analytics career.
Common questions
Sample answer frameworks for each of these are on the Fractal Analytics HR interview questions page.
Unlike most companies where clearing HR means an offer letter within days, Fractal’s Imagineer program inserts a multi-week (commonly cited as 10-12 weeks) training period between “you’re selected” and the actual offer letter/joining date. This is a real structural quirk worth planning around - don’t assume you have an offer in hand the moment interviews end, and don’t be surprised if the formal paperwork takes noticeably longer than at other companies running the same campus season.
Imagineers are hired into one program but end up in different jobs. Campus job descriptions list four tracks: Data Engineering (pipelines, SQL, cloud data platforms), Data Science (models and statistics), Decision Science (business problem framing and analytics for clients) and Data Visualization (dashboards and reporting). Track placement is driven by pre-onboarding performance and business need, not only your preference.
The interview loop is the same for all of them, which is why it mixes Python and SQL with cases and guesstimates. If you want a particular track, say so in the HR round and back it with a matching project. This page covers the shared Imagineer loop.
A list is mutable and a tuple is immutable, which drives everything else. Because a tuple cannot change after creation its hash is stable, so tuples can be dictionary keys and set members while lists cannot. Tuples are slightly smaller and faster to construct, which matters when you are creating millions of records, and immutability protects against accidental mutation when a value is passed into a function. A classic gotcha Fractal’s Python section likes is the mutable default argument: writing def f(x, acc=[]) reuses the same list across every call, because defaults are evaluated once at definition time - the fix is to default to None and create the list inside the function.
Use a window function: SELECT category, product, revenue FROM (SELECT category, product, SUM(amount) AS revenue, ROW_NUMBER() OVER (PARTITION BY category ORDER BY SUM(amount) DESC) AS rn FROM sales GROUP BY category, product) t WHERE rn = 1; PARTITION BY restarts the numbering within each category, so filtering on rn = 1 gives one winner per category. The subquery is required because a window function cannot be referenced in the WHERE clause of the same SELECT - windows are evaluated after WHERE and GROUP BY. Swap ROW_NUMBER for RANK if you want every product tied for first place returned rather than one arbitrary row.
Compute the complement rather than the direct sum, because at least one is far easier to handle that way. The probability of not rolling a six on a single roll is 5/6, and rolls are independent, so the probability of no six across four rolls is (5/6) to the fourth power, which is 625/1296 or roughly 0.482. The answer is therefore 1 minus that, about 0.518 or 51.8 percent. This is the classic de Mere problem, and the interviewer is checking that you reach for the complement instead of trying to add the overlapping cases of exactly one, two, three and four sixes.
Decompose profit into revenue minus cost and work down a tree rather than guessing causes. Revenue splits into volume times price, where volume breaks into market size times market share, and price into list price minus discounts and mix shift. Cost splits into fixed and variable, then into inputs, labour, logistics and overhead. Next isolate the drop along dimensions - by product line, geography, customer segment and time - so you learn whether the decline is broad or concentrated, usually the single most informative cut. Then check whether the cause is internal or external by benchmarking against competitors and the category, and close with a recommendation tied to whichever branch actually moved.
Work top-down from population to smartphone users to replacement cycle, stating assumptions aloud with round arithmetic - Fractal scores the structure. Start from a population of about 1.4 billion and a smartphone-user base of roughly 700 million, then apply an average replacement cycle of about 3 years, which gives roughly 230 million replacement units per year. Add new first-time buyers from population growth and feature-phone upgrades, perhaps 20 to 30 million, for a total in the range of 250 to 260 million - close to published shipment figures of around 150 million, so you should note the gap and say your replacement-cycle assumption is the most likely culprit. Naming the assumption your answer is most sensitive to, and sanity-checking against any known anchor, matters more than the number.
Correlation measures how two variables move together; causation means changing one actually produces a change in the other. Correlation appears without causation for three reasons: a confounding variable driving both, such as ice-cream sales and drowning deaths both rising with temperature; reverse causation, where the outcome drives the supposed cause; and pure coincidence when many variables are searched at once. Establishing causation requires either a randomised controlled experiment, which balances confounders by design, or a quasi-experimental method such as difference-in-differences, instrumental variables or regression discontinuity when randomisation is impossible. In client work the distinction decides whether a finding is actionable: correlation supports a prediction, but only a causal estimate justifies pulling a lever.
A p-value is the probability of observing data at least as extreme as yours under the assumption that the null hypothesis is true. Below the chosen threshold, usually 0.05, you reject the null. It is not the probability that the null hypothesis is true, nor the probability that your result is a fluke, and a p-value above 0.05 is not evidence that the null is true - it just means you lack the evidence to reject it. The threshold is also a Type I error rate, so testing twenty hypotheses at 0.05 gives roughly a 64 percent chance of at least one false positive unless you apply a correction such as Bonferroni. Report effect size and confidence interval alongside it, since a large enough sample makes trivial differences significant.
Track two variables in a single pass - largest and second_largest, both initialised to negative infinity. For each value, if it exceeds largest, push the old largest down into second_largest and update largest; otherwise if it is strictly less than largest but greater than second_largest, update second_largest only. That is O(n) time and O(1) space, versus O(n log n) for sorting. The strict inequality matters: without it a list of identical values reports the same number twice when the interviewer meant the second-largest distinct value, so clarify that requirement before coding. Also handle the edge case of a list with fewer than two distinct elements rather than returning negative infinity.
WHERE filters rows before aggregation; HAVING filters groups after GROUP BY. Fractal interviewers often ask this right after you write a query on a live editor - explain that you cannot put COUNT(*) > 5 in WHERE because the aggregate does not exist yet. Example: filter active orders in WHERE, then HAVING SUM(amount) > 10000 for high-value customers per region. Mention execution order (FROM → WHERE → GROUP BY → HAVING → SELECT) to show consulting-grade clarity.
Relational databases store structured tables with schemas, ACID transactions, and SQL - good for joins and reporting. Non-relational stores (document, key-value, column, graph) trade flexible schemas and horizontal scale for weaker ad-hoc join patterns. VIT 2026 Imagineer reports asked this after project discussion - tie your answer to your project: “We used Postgres for transactional consistency; Redis for session cache” beats a textbook list.
Use median when outliers skew the mean - revenue per user with a few whale accounts, or housing prices. Use mean when the distribution is stable and the client cares about totals (average order value with bounded tickets). Mention that reporting mean alone without checking skew is a common dashboard mistake; Fractal’s stats probes in OA and tech rounds expect you to name the failure mode.
Segment Chennai population (~10M metro), estimate pizza-eating households or youth share, multiply by meals per week converted to daily, apply market share for Domino’s vs competitors, and adjust for weekday vs weekend. State which assumption moves the answer most (penetration vs frequency). This mirrors a reported VIT 2026 guesstimate - interviewers reward explicit branches, not precision.
Fractal does not publish a negative-marking rule, and public candidate reports do not settle it, so read the instruction screen before guessing. Two 2021 GfG campus reports agree on the shape: an aptitude test of 70 questions in 75 minutes, then a technical test of 8 questions in 90 minutes (4 SQL, 4 Python or R). Neither mentions a penalty.
No test score cutoff is published; shortlists are relative to the batch. Candidate-reported campus filters are 7.0+ CGPA (8.0+ on some drives such as VIT 2026), circuit branches plus M.Sc or M.A Economics or Statistics on some drives, and no active backlogs.
Fractal hires freshers into its Imagineer program through 4-5 stages: 1. Online Test I (75 minutes in 2021 reports) - 70 questions across data analysis, quantitative ability, reasoning, and verbal sections. 2. Online Test II (90 minutes in 2021 reports) - 8 questions, 4 SQL and 4 Python or R. 3. Technical/Problem Solving Round(s) (30-45 minutes) - project discussion plus case-study questions (market entry, profitability style). 4. Business Understanding Round (30 minutes) - guesstimates and probability questions, judged mainly on structure and approach rather than the exact number. 5. HR Round (20-30 minutes) - strengths, values, and why Fractal. Each interview round is a straight elimination round.
Imagineer is Fractal’s fresher hiring track for its analytics/AI consulting business. A distinctive part of it is a multi-week pre-onboarding training period - candidates report roughly 10-12 weeks of training - between being selected and receiving the final offer letter/joining date, rather than an immediate offer right after the HR round.
Expect Python and SQL coding questions in the online tests, business case questions modeled on market-entry and profitability problems, and guesstimate/probability questions in the Business Understanding round where interviewers care more about how you structure the problem than whether you land on the ‘right’ number. HR questions focus on why you want an analytics career and why Fractal specifically.
Most candidates go through two online tests (aptitude, then Python/SQL) followed by two to three interview rounds - technical/problem-solving, business understanding, and HR - each of which is an elimination round. Fractal is known to be highly selective, historically hiring under 2% of applicants.
Practice Python and SQL basics for the second online test, revise data interpretation and quantitative aptitude for the first, and drill structured approaches to guesstimates and business cases (market sizing, profitability trees) since Fractal explicitly evaluates your reasoning process. Be ready to discuss any analytics/ML project you’ve done in detail, and expect the pre-onboarding training period if selected.
| Stage | Strong signal | Weak signal |
|---|---|---|
| Online tests | Quants + SQL + Python accuracy | Skipping SQL to chase hard DSA |
| Technical | Live SQL with execution-order explanation + project depth | Query that “works” but cannot explain HAVING |
| Business / Apex | Metric → hypotheses → data → caveat in ~8 minutes | 15-minute ML lecture when asked why conversion dropped |
| HR | Why Fractal + pre-onboarding (~10-12 week) awareness | Shock at training gate after selection |
Imagineer loops reward client-ready caveats. Prefer a clean boundary over fake CPG expertise.
Good fit: You like messy business questions, you can write a JOIN without Stack Overflow, and you can explain a chart in two minutes. Consulting-analytics culture means they care whether a client would trust your caveat.
Harder path: You only practiced LeetCode hard; you cannot write GROUP BY; you freeze when asked “so what should the VP do?” A high CGPA with zero SQL usually dies in OA. A mid CGPA with clean SQL and one honest case often survives.
Use this as a default four-week loop before a Fractal Analytics drive. Adjust to your invite timeline.
10-day crash version: Days 1-4 timed DSA, days 5-7 resume projects + CS fundamentals, days 8-9 mock interviews out loud, day 10 HR STAR stories on the HR page.
Candidate-reported fresher/early-career packages for Fractal Analytics often cite ₹8-14 LPA. Treat forum numbers as ranges, not guarantees - confirm the written offer letter for your team and location.
Interview process details and sample questions on this site are editorially organized from public candidate reports and hiring write-ups, with round structure, prep notes, and company-specific framing added by our team. For round-by-round stories, see the Fractal Analytics interview experience page. For OA papers and coding drills, use Fractal Analytics on placementpapers.app. OA quick answers and track-wise questions draw on two GfG Fractal Imagineer interview experiences (2021 campus drives), a GfG Fractal engineering interview experience, Fractal Glassdoor interview snippets (2024-2026) seen via search, and the campus job posts and candidate reports already summarised on this page.
