Using AI to Check Your Maths and Build Estimates
A founder is staring at a whiteboard full of numbers: monthly orders, delivery riders, repeat rates, average ticket size. One wrong zero changes the story from “profitable pilot” to “cash-burning disaster” - and AI can catch that error only if you ask it like a manager, not like a calculator.
- AI is best used as an auditor, not the owner of the estimate. You build the logic; AI checks arithmetic, units, assumptions and missing drivers.
- A good estimate follows: define the metric, break it into drivers, calculate transparently, sanity-check the range, then communicate the answer.
- Never ask “solve this guesstimate” first. Ask: “Audit my approach, find unit errors, test assumptions and recalculate using my formula.”
- LLMs can hallucinate numbers. Use hypothetical inputs, company-provided case data or clearly stated assumptions unless a source is given.
- The strongest candidates show a number, a confidence range and the one assumption that matters most.
- The interviewer rewards judgement more than precision: clean logic beats false accuracy.
Big Picture
Think of AI as a second pair of eyes sitting beside your scratchpad. It should not replace your reasoning; it should pressure-test your reasoning before you speak with confidence.
Core Explanation: The AI-Aided Estimation Loop
In cases, market sizing, profitability, operations and product interviews, estimates usually fail for three reasons: the candidate solves the wrong question, misses a major driver, or makes a silent arithmetic error. AI helps most in the second and third problem - but only after the first is clear.
Start by defining the problem sharply. If you are unsure whether the interviewer wants annual market size, monthly revenue, unit capacity or profit impact, pause and clarify. This is the same discipline covered in defining the problem before solving it.
The Five-Step Method
The mental model is a funnel: wide business reality at the top, narrow defensible estimate at the bottom. AI is useful because it helps you see what leaked out of the funnel.
The Prompt That Actually Works
Do not ask AI: “Solve this market sizing problem.” That trains you to become a passenger. Use a manager-style audit prompt instead:
“I am preparing for a case interview. Do not solve from scratch. Audit my estimate below. Check arithmetic, unit consistency, missing drivers, double counting, unreasonable assumptions and the one assumption that most affects the result. Then recalculate only using my formula and show corrections clearly.”
This prompt is powerful because it separates thinking from checking. You remain the consultant; AI becomes the analyst reviewing your model.
Worked Example: Monthly Revenue of a Campus Chai Kiosk
Suppose you are asked to estimate monthly revenue for a chai kiosk inside a B-school campus. Use simple hypothetical assumptions:
- Students on campus = 1,000
- Daily buyers = 45 percent of students
- Average cups per buyer per day = 1.2
- Average price per cup = ₹20
- Operating days per month = 30
Formula: Monthly revenue = students × daily buyer share × cups per buyer per day × price × days.
Calculation: 1,000 × 0.45 × 1.2 × ₹20 × 30 = ₹3,24,000 per month.
Now ask AI to audit. A good AI check might confirm the arithmetic, verify that the output is monthly revenue, and flag the key sensitivity: if daily buyer share is 35 percent instead of 45 percent, revenue becomes 1,000 × 0.35 × 1.2 × ₹20 × 30 = ₹2,52,000. Your answer becomes stronger because you can say: “Base estimate is about ₹3.2 lakh per month, with buyer penetration as the biggest swing factor.”
How to Know Your Estimate Is Interview-Ready
Use these checks before you finalise the answer. They are simple enough to apply in two minutes and rigorous enough to impress.
Human vs AI: Who Should Do What?
The best candidates divide the work cleanly. Humans are better at context, business judgement and trade-offs. AI is faster at recomputation, pattern spotting and generating alternative checks.
When practising, first solve on paper. Then paste your approach into AI and ask for three things only: “Find one arithmetic error, one missing driver and one unrealistic assumption.” This keeps feedback focused.
Definitions You Can Say Aloud
- Estimate: A reasoned approximation built from assumptions when exact data is unavailable.
- Driver: A variable that directly moves the final number, such as price, volume, frequency, cost or conversion.
- Sanity check: A quick test that confirms whether an answer is directionally plausible.
- Sensitivity: The change in the final answer caused by changing one key assumption.
- MECE: Categories are non-overlapping and collectively cover the problem.
Urban Company: Using Estimates to Think Like an Operator
Urban Company shows why service businesses need clean estimates of demand, professional capacity, utilisation and repeat behaviour before scaling city operations.
Urban Company is a useful case because it is not just a consumer app story. It is an operations story. A customer sees a clean booking screen; behind it sits a harder question: are enough trained professionals available in the right locality, at the right time, at the right quality level?
Situation: In home services, demand is local and time-bound. A salon booking in Gurgaon on a Saturday evening cannot be served by an idle professional in another city. Growth therefore depends on estimating both sides of the marketplace: customer demand and service professional supply.
The move: Urban Company built a more standardised service model around trained professionals, service categories, app-based discovery, scheduling and quality control. The primary driver was creating a reliable supply network for fragmented home services. Supporting drivers included customer trust, standardised service packages, technology-enabled matching and repeat usage.
The lesson: If you were estimating city expansion potential, you would not stop at “number of households × service price.” You would also estimate active households, booking frequency, professional hours available, utilisation, travel time, cancellations and contribution margin. AI can help audit that driver tree and catch double counting, but the business logic must come from you.
The strategic takeaway: marketplace estimates must balance demand and capacity. AI is excellent at checking whether the two sides reconcile; it is poor at knowing which operational constraint matters unless you specify the business context.
How AI Changes Using AI to Check Your Maths and Build Estimates
AI is changing estimation practice in three concrete ways.
- From answer checking to assumption checking: Earlier, students used calculators to verify arithmetic. Now AI can ask whether penetration, frequency or capacity assumptions are unrealistic.
- From one estimate to scenario ranges: AI can quickly recompute base, low and high cases once your formula is clear. This helps you speak in ranges instead of pretending rough assumptions create exact answers.
- From passive practice to targeted feedback: AI can review your solved case and identify repeated weaknesses - unit errors, skipped sanity checks, poor rounding or missing cost drivers.
The caution is simple: a language model is not automatically a truth engine. It may produce fluent but wrong arithmetic if it is not using a calculation tool, and it may invent assumptions if you do not constrain it.
Solve a guesstimate on paper, then paste only your formula, assumptions and arithmetic into ChatGPT. Prompt: “Audit this like a case interviewer. Do not introduce external facts. Check units, recompute the maths, identify missing drivers and suggest one sensitivity test.” Then redo the estimate aloud without looking at the AI response.
If you want to practise full cases, use AI as a structured interviewer rather than a shortcut. The natural next step is practising cases with AI as a mock interviewer.
Interview Relevance
“Estimate the monthly revenue of a food delivery counter on our campus. You may make assumptions. How would you use AI afterwards to check your answer?”
In profitability or break-even cases, estimation quality matters even more because one driver error can flip the recommendation. If your estimate feeds a profit decision, revise contribution margin and break-even analysis in cases next.
Common Mistake
The biggest mistake is letting AI solve before you think. It costs candidates because they lose the ability to defend assumptions, explain trade-offs and recover when challenged. One-line fix: build your estimate first, then use AI only to audit maths, units, assumptions and sensitivity.