A VP at a mid-size SaaS company approved a $2M marketing budget last quarter. The reasoning? Their biggest competitor had spent roughly the same and tripled their pipeline. Seemed logical. But 30 seconds of napkin math would have killed the idea on the spot. The competitor had 12x the brand recognition, a 4% organic conversion rate (compared to this company's 0.6%), and a sales team three times the size. The same $2M spend, filtered through those numbers, projected to roughly $180K in new revenue. Not $2M. Not even close. Nobody did the math. They approved the budget in a 45-minute meeting where the prettiest slide deck won.
This happens constantly. Not because people are bad at math, but because they never learned the one math skill that actually matters in professional life: estimation skills for business, the ability to get a rough answer to a hard question in under a minute, using nothing but your brain and maybe the back of an envelope.
What Is Fermi Estimation (and Why Should You Care)?
Enrico Fermi was an Italian physicist who won a Nobel Prize and helped build the first nuclear reactor. He was also famous for asking impossible-sounding questions and then answering them with surprising accuracy. "How many piano tuners are in Chicago?" That's the classic Fermi problem. It sounds like a trick question, but it's not. It's a method.
Fermi estimation (also called back-of-envelope calculation) is the practice of breaking a big, unknowable question into smaller pieces you can estimate, then multiplying those estimates together. You won't get the exact answer. You don't need to. The goal is to land within an order of magnitude, within a factor of 10 of the real number. That's almost always enough to make or kill a decision.
This isn't just a physics party trick. Management consultants at McKinsey and BCG use Fermi problems in interviews because they test exactly what matters in advisory work: structured thinking under uncertainty. Venture capitalists use napkin math to evaluate pitch decks in real time. Product managers use it to size markets before committing engineering resources. Founders use it to figure out whether a business model can work before writing a single line of code.
If you've ever studied probability applied to real-world data, you already have the right instincts. Estimation is probability's scrappier cousin: less precise, faster, and often more useful in a meeting.
5 Mental Math Anchors Worth Memorizing
Every good estimator carries a small set of reference numbers in their head. These are anchors: rough-but-useful facts that let you build estimates quickly without Googling. You don't need hundreds of them. Five good ones cover a shocking amount of ground.
| Anchor | The Number | Why It's Useful |
|---|---|---|
| World population | ~8 billion | Base for any global market sizing. "What fraction of humans would buy this?" |
| US population | ~335 million | The default denominator for US market estimates. ~130M households. |
| Working hours per year | ~2,000 | 50 weeks x 40 hours. Converts hourly rates to annual salary instantly. $50/hr = $100K/yr. |
| Seconds in a day / year | ~86,400 / ~31.5 million | Crucial for throughput, server load, and capacity calculations. |
| Rule of 72 | 72 / growth rate = doubling time | At 8% annual growth, your revenue doubles in 9 years. At 20%, it doubles in 3.6. |
That last one, the Rule of 72, is worth highlighting. If someone tells you their startup is growing 15% month-over-month, the Rule of 72 says it doubles every 4.8 months. Which means in a year, they'd be at roughly 5x their current size. If they're at $100K MRR, they're projecting $500K MRR in 12 months. Does that pass the smell test given their market, team, and product? Now you have a frame for the conversation. Understanding how roots and powers work under the surface makes anchors like the Rule of 72 feel intuitive rather than memorized.
Memorize these five. Tattoo them on the inside of your eyelids if you have to. They're the foundation of every quick mental math estimate you'll ever do.
How to Solve a Fermi Problem in 4 Steps
Fermi estimation is not about guessing. It's about decomposition: breaking a question into smaller sub-questions where each piece is individually estimable, then combining the pieces. Here's the method.
Restate the question in precise, measurable terms. "How big is the market?" becomes "How many people in the US would pay $20/month for this product?" Vague questions produce vague answers.
Find a chain of multiplications or divisions that connects your anchors to your answer. Population x percentage x price x frequency. Each factor should be something you can make a reasonable guess about.
For each sub-estimate, ask: am I sure this is within 2x of the real number? Within 5x? If you're wildly uncertain about one factor, that's where you should focus more thought (or, later, actual research).
Combine your factors. Round to one significant figure. Then ask: does this answer feel plausible? Compare it to something you already know. If your estimate says there are 50 million restaurants in the US, something went wrong (there are about 1 million).
That's it. Four steps. The whole process takes 30 to 90 seconds once you've practiced it a few times. The key insight is that errors in your individual estimates tend to cancel out: you'll overestimate some factors and underestimate others. The final product is usually more accurate than any single guess.
5 Worked Fermi Problems (with Full Reasoning)
Theory is nice. Let's do some actual napkin math. These five problems cover the kinds of estimation you'll actually face in business contexts.
Problem 1: How much revenue does a single Starbucks location make per year?
Start with what you can observe. A busy Starbucks serves maybe one customer every 30 seconds during morning rush (let's say 3 hours), one per minute during the rest of the open hours (about 12 hours). That's roughly 360 during rush + 720 during off-peak = about 1,000 customers per day. Average ticket is probably $5 to $6 (some people just get drip coffee, some get a latte and a sandwich). Call it $5.50. So daily revenue is about $5,500. Multiply by 365 days: roughly $2 million per year.
The actual average is about $1.5 to $2 million for US locations. We're right in the zone.
Problem 2: How many SaaS companies are there in the United States?
There are about 6 million employer businesses in the US (anchor worth knowing). The tech sector is roughly 10% of the economy. Of tech companies, maybe half are software-focused. Of software companies, maybe 60-70% are now SaaS (subscription model). So: 6M x 10% x 50% x 65% = about 195,000. Round to 200,000.
Various industry reports estimate 17,000 to 30,000 SaaS companies. We're off by about 7-10x. What happened? Most "tech sector" companies aren't building software products. IT services, consulting, hardware resellers all count as tech. The real fraction of tech companies building SaaS products is closer to 1-2%, not the 5% I implied. Adjusted: 6M x 1.5% x 30% = 27,000. Much better. This is why the sanity check matters.
Problem 3: Is it worth hiring a salesperson who costs $120K/year?
Your average deal size is $15K. Your current close rate is 20%. A salesperson can probably handle 8 qualified calls per week, or about 400 per year. But only half of those will be truly qualified leads. So: 200 qualified leads x 20% close rate = 40 deals. 40 deals x $15K = $600K in revenue. If your gross margin is 70%, that's $420K in gross profit from $120K in cost. The salesperson pays for themselves roughly 3.5x over. Yes, hire them.
But what if your close rate is only 10% and average deal size is $8K? Then: 200 x 10% x $8K = $160K revenue. At 70% margin, that's $112K. Now the salesperson barely breaks even. Same question, very different answer, and you got there in 45 seconds. This is the kind of thinking that strong business strategy is built on.
Problem 4: How many tennis balls fit in this room?
The classic interview question. A typical conference room might be 20 feet long, 15 feet wide, 10 feet tall. Volume: 3,000 cubic feet. A tennis ball is about 2.6 inches in diameter, so roughly 0.22 feet. Cube that: about 0.01 cubic feet per ball. But spheres don't pack perfectly. Random packing fills about 64% of the space. So: 3,000 / 0.01 x 0.64 = about 192,000 tennis balls.
This one sounds useless, but the skill it tests (spatial reasoning plus unit conversion plus knowing about packing efficiency) shows up constantly in logistics, warehouse planning, and capacity estimation.
Problem 5: What's the market size for a premium dog food delivery service in Austin, Texas?
Austin metro population: about 2.3 million. US average: roughly 50% of households have a dog. Austin is dog-friendly, so maybe 55%. Average household size is about 2.5 people, so roughly 920,000 households. 55% with dogs = about 506,000 dog-owning households. Premium dog food? That's an upper-middle-class product. Maybe 25% of dog owners would consider premium. Of those, maybe 10% would actually pay for delivery (the rest buy at the pet store). So: 506,000 x 25% x 10% = about 12,650 potential customers. At $80/month, that's roughly $12 million per year in addressable market. Enough to build a real business, but not enough to attract serious VC money.
Being off by 2x is a rounding error. Being off by 10x kills companies. If your estimate says a project will cost $500K and it actually costs $800K, you can adjust. If it costs $5M, your career has a problem. Fermi estimation protects you from the 10x mistakes, not the 2x ones. That's where its value lives.
Two practical implications: (1) if two of your sub-estimates are each off by 3x in the same direction, your final answer is off by 9x and you're approaching the danger zone. Diversify your estimation approach. (2) Always do a sanity check by estimating the same thing from a completely different angle. If both methods land in the same ballpark, you're probably safe.
Why 2x Off Is Fine but 10x Off Kills You
Most business decisions are not sensitive to small errors. If you estimate a project takes 6 months and it takes 9, that's painful but survivable. You adjust timelines, reallocate resources, maybe have an awkward conversation with a client. The company lives.
But if you estimate 6 months and it takes 5 years? That's a different category of mistake entirely. Budgets get blown. Opportunity costs stack up. Teams burn out. Sometimes the company folds. The 10x error is not a bigger version of the 2x error. It's a qualitatively different kind of failure.
This is why the order of magnitude matters more than precision. A quick estimate that gets you within 2-3x of the real answer is infinitely more valuable than a detailed spreadsheet that takes two weeks to build but is based on flawed assumptions. The spreadsheet gives you false confidence. The napkin math gives you a gut check.
Estimated $500K project cost, actual was $900K. Budget overrun requires reallocation but the project delivers value. Team adjusts scope for Phase 2. Lesson learned, nobody fired.
Estimated $500K project cost, actual trajectory is $5M. Project cancelled after $1.2M spent. Team reassigned. VP replaced. Competitor captures the market window. Recovery takes 18 months.
The VP from the opening story made a 10x error. Not because the competitor lied about their numbers, but because nobody bothered to check whether those numbers transferred to a different company with different fundamentals. Thirty seconds of napkin math would have caught it.
Common Estimation Mistakes (and How to Avoid Them)
Even after you learn the method, a few traps catch people repeatedly.
Anchoring too hard on the first number you think of. If someone says "this market is worth $50 billion" before you estimate, your brain will unconsciously orbit around $50 billion. Estimate first, then compare to external claims.
Confusing addressable market with total market. There are 330 million Americans, but your product is for left-handed vegan rock climbers in cities with populations over 500,000. Your actual addressable market is maybe 14,000 people. The chain of filters matters enormously.
Forgetting that percentages multiply, they don't add. If you need customers to see your ad (30% of traffic), click it (2% CTR), visit the page (60% load without bouncing), and buy (3% conversion), the final number is 0.30 x 0.02 x 0.60 x 0.03 = 0.000108. About 1 in 10,000. Not 95% (30 + 2 + 60 + 3). People make this mistake more often than you'd think.
Ignoring base rates. "Our new feature will increase retention by 20%." Sounds great. But if your current retention rate is 5%, a 20% improvement means 6%. The absolute change is 1 percentage point. Is that worth $300K in engineering time? Maybe. But the napkin math reframes the conversation.
How to Practice Estimation (a 30-Day Approach)
Estimation is a skill, not a talent. It gets better with practice. Here's how to build the habit without making it feel like homework.
Week 1: Estimate stuff you can verify. How many steps from your front door to the nearest coffee shop? How many emails did you get today? How many cars are in the parking lot? Estimate, then count. Calibrate your intuition against reality.
Week 2: Estimate business numbers from public companies. How much revenue does McDonald's make per store? How many employees does your favorite SaaS company have? Estimate, then look up the annual report. Public company data is free and abundant.
Week 3: Practice decomposition. Take questions you can't look up and practice the 4-step method. How many dog groomers are in your city? How much does your office building cost to heat per month? Write out the chain of logic. The goal isn't the answer, it's the breakdown.
Week 4: Start using it at work. Next time someone proposes a project, estimate the ROI before you look at the business case. When a vendor gives you a price, estimate what the service actually costs them. When someone cites a statistic, estimate whether it's plausible. This is where it becomes a superpower.
You're in a product meeting. Someone suggests adding a referral program. "If just 5% of our users refer a friend, we'll double our user base." You pause. You have 10,000 users. 5% referring = 500 referrals. You'd need 10,000 new users to double. That means each referrer needs to bring in 20 friends. That's not a referral program, that's a pyramid scheme. The 5% number sounded reasonable, but the math behind it was off by 20x. You caught it in 10 seconds because you practiced.
When to Stop Estimating and Start Spreadsheet-ing
Napkin math is for the first 5 minutes. It tells you whether an idea is plausible, whether a number passes the smell test, whether you should keep talking or move on. It's a filter, not a final answer.
You should move to a proper model when:
The stakes are high and the estimate is borderline. If your napkin math says a project returns 1.5x the investment, that's too close to 1x to trust without more detail. Build the spreadsheet. If the napkin says 10x or 0.2x, you probably don't need one.
Multiple scenarios could play out. Estimation gives you a point estimate. When you need to model best case, worst case, and most likely case, a spreadsheet with variable inputs is the right tool.
You need to communicate the reasoning to stakeholders. "I did some napkin math" doesn't fly in a board presentation. But "I started with a rough estimate that suggested this was viable, then built a detailed model to confirm" makes you look thorough and efficient.
Sensitivity matters. When you need to know which variables your result is most sensitive to (does a 10% change in price matter more than a 10% change in volume?), you need a model you can actually toggle.
The mistake most people make is jumping straight to the spreadsheet. They spend three days building a financial model for an idea that 30 seconds of napkin math would have killed. The estimation is the screening round. The spreadsheet is the deep dive. Use them in that order.
Estimation as a Competitive Edge
Here's something nobody tells you in school: the people who rise fastest in business are almost never the best at detailed analysis. They're the best at quick analysis. They can sit in a meeting, hear a proposal, and within a minute have a rough sense of whether the numbers work. They ask the questions that cut through the noise. "What's our cost per acquisition on this channel?" "If we assume a 3% conversion rate, how many leads do we need to hit that target?" "What's the payback period on this hire?"
These questions sound basic. They are basic. But most people in the room can't answer them off the top of their head, and most people won't bother to estimate before opening their mouths. The person who does the quick mental math before speaking has an enormous advantage. They redirect entire conversations. They prevent six-figure mistakes. They look like strategic thinkers because, functionally, they are.
Fermi estimation is not a math skill. It's a thinking skill wrapped in math. It forces you to ask what the key variables are, which ones matter most, and whether the whole thing passes a basic sanity check. That's the same process behind every good strategic business decision.
The takeaway: You don't need a better spreadsheet. You need a faster bullshit detector. Fermi estimation gives you one. Memorize five anchors, practice the 4-step decomposition for a month, and you'll catch bad assumptions before they become bad decisions. The math is simple. The habit is what separates the people who approve $2M mistakes from the people who prevent them.
