Timed Modelling Tests: Build a Clean Model Under Pressure
Two candidates receive the same messy case pack at 10:00 AM. By 11:30, one has a beautiful workbook with half the outputs missing; the other has a simple, checked model, a clear recommendation and three assumptions they can defend.
- A timed modelling test is not a test of fancy Excel; it is a test of commercial logic under time pressure.
- Start with the required output, not the input data. Build backwards from decision, metric and format.
- Use a simple flow: question - drivers - model - checks - recommendation.
- A strong model is complete, auditable, formula-consistent and sanity-checked before it is polished.
- Time-box aggressively: 10 minutes to understand, 15 minutes to structure, 35 minutes to build, 15 minutes to check, 15 minutes to write the story.
- The best candidates separate assumptions, calculations and outputs so the reviewer can follow the logic fast.
- The biggest trap is overbuilding. A correct 80 percent model beats an elegant unfinished one.
Big Picture: What a Timed Modelling Test Really Measures
A timed modelling test measures whether you can convert an ambiguous business question into a decision-ready model before the clock beats you. Firms are not looking for a museum-quality workbook; they are looking for a reliable thinking machine.
Use these practical measures to judge your model before submitting it. They are not accounting standards; they are test-room quality signals.
Core Explanation: How to Attack the Test Without Freezing
The hidden skill is prioritisation. A modelling test gives you more possible analysis than time available, so your first job is to decide what not to model.
The Clean Workbook Structure Firms Like
A reviewer should understand your workbook in under a minute. That happens when the model has a predictable architecture.
Worked Example: A 10-Minute Profit and Cash Flow Model
Assume a simplified case gives you these inputs: current revenue is ₹100 crore, next-year growth is 20%, gross margin is 45%, operating expenses are 30% of revenue, tax is 25%, capex is 5% of revenue and net working capital investment equals 10% of incremental revenue. Ignore depreciation and debt for this quick test-room example.
The answer is not just “FCF is ₹5.5 crore.” The stronger answer is: “Growth creates ₹20 crore of incremental revenue, but capex and working capital absorb much of the operating profit, so cash conversion is the key sensitivity.” That is the commercial story.
Definitions You Should Be Able to Say Cleanly
Financial model: A structured, assumption-driven representation of a business used to forecast financial outcomes and support decisions.
Driver: A variable that materially influences an output, such as volume, price, margin, churn, utilisation, capex or working capital.
Sensitivity analysis: Testing how an output changes when one or more key assumptions change.
Sanity check: A quick test that confirms model outputs are directionally reasonable, internally consistent and commercially plausible.
Case Study: Lenskart and the Timed Model Mindset
Lenskart is a strong Indian example for timed modelling because its business forces you to connect online demand, offline stores, product margins, supply chain and repeat purchase behaviour.
Situation: Eyewear in India has long been a fragmented category with a mix of local opticians, offline discovery and trust-heavy purchases. A pure online model would struggle because customers often want fit, eye tests and confidence before buying.
The move: Lenskart built an omnichannel model: online discovery and convenience, supported by physical stores, eye-testing infrastructure, private-label products and a controlled supply chain. The primary driver is the integration of digital demand generation with offline fulfilment and trust-building. Supporting drivers include product assortment, store expansion, repeat purchase potential, supply chain control and brand recall.
How this becomes a timed modelling test: A firm may ask you to model whether opening new stores in a city is attractive. The weak candidate builds a giant retail model. The strong candidate identifies the few drivers that matter: store-level revenue, gross margin, rent and staff cost, customer acquisition, repeat purchase and working capital.
Outcome or lesson: The business lesson is that an omnichannel company cannot be judged using only website metrics or only store metrics. The modelling lesson is sharper: when time is limited, model the economic engine, not every operational detail.
How AI Changes Timed Modelling Tests
AI raises the bar because basic formula generation is becoming easier. What remains valuable is judgement: choosing assumptions, detecting nonsense, explaining trade-offs and defending the model.
- AI improves preparation, not permission: Many live tests restrict external tools. Treat ChatGPT, Claude or Excel Copilot as practice partners unless the test explicitly allows them.
- AI can generate cases instantly: You can ask ChatGPT to create a 90-minute FP&A, valuation or unit-economics case, then solve it in Excel without assistance.
- AI can audit your thinking: After the test practice, paste your assumptions and outputs into Claude or ChatGPT and ask: “Which assumptions are most fragile, and what sanity checks are missing?”
- AI makes communication more important: If many candidates can build formulas, the differentiator becomes the written recommendation and the ability to defend sensitivities.
Load a company annual report, investor presentation or DRHP into NotebookLM. Ask it to extract revenue drivers, cost drivers, working-capital items and risks. Then ask ChatGPT to turn those drivers into a 90-minute modelling prompt. Solve it manually, and use AI only after completion to critique your model.
Interview Relevance
“Suppose we give you 90 minutes to build a model on whether a retail brand should open 50 new stores. How would you approach the test?”
Say this line if you get stuck: “I would rather submit a simpler checked model that answers the decision than an over-detailed model with unverified outputs.” That signals maturity.
Common Mistake
The single most costly error is starting with detailed Excel building before defining the final decision output. It costs candidates because they spend precious time modelling irrelevant detail and leave no time for checks or recommendation. Fix it with one line: write the final output table first, then build only the drivers needed to populate it.
What to Revise Next
This is the natural capstone. Do one full 90-minute mock test, then spend 30 minutes reviewing only three things: whether your model answered the decision, whether your checks caught errors, and whether your final recommendation was defensible.