Scrolls
2 min read
On this page

Back-of-Envelope Math

Back-of-envelope math is the art of doing quick, rough calculations to check whether something makes sense. You do not need a spreadsheet or a calculator — just round numbers, simple arithmetic, and a willingness to be approximately right rather than precisely wrong.

The Everyday Version

Can I Drive 500km in 4 Hours?

Quick calculation:
- Highway speed limit: about 120 km/h
- In 4 hours at 120 km/h: 4 x 120 = 480 km
- 500 km is more than 480 km
- Answer: Barely, and only if you never slow down,
  never stop for fuel, and hit no traffic

More realistic estimate:
- Average speed including stops and slowdowns: 100 km/h
- In 4 hours: 4 x 100 = 400 km
- Answer: No. Budget 5 hours.

You did not need a GPS route planner to get a useful answer. Thirty seconds of mental math told you the plan was unrealistic.

Can I Afford This Vacation?

One-week beach trip for a family of four:

Round numbers:
- Flights: $400 per person x 4 = $1,600
- Hotel: $150 per night x 7 = $1,050
- Food: $100 per day x 7 = $700
- Activities and transport: $500
- Total: roughly $3,850

Round up for surprises: about $4,500

You now know the order of magnitude. If your budget
is $3,000, you need to cut something. If your budget
is $10,000, you are comfortable. You did not need to
price every meal at every restaurant.

How Much Paint Do I Need?

Room: roughly 4m x 5m, walls are 2.5m high

Wall area:
- Two long walls: 2 x (5 x 2.5) = 25 square meters
- Two short walls: 2 x (4 x 2.5) = 20 square meters
- Total: 45 square meters
- Subtract a door and window: maybe 5 square meters
- Paintable area: about 40 square meters

Paint coverage: roughly 10 square meters per liter
Liters needed: 40 / 10 = 4 liters for one coat
Two coats: 8 liters

Go buy two 4-liter cans. Close enough.

How Long Will This Renovation Take?

Kitchen renovation estimate:

- Demolition: 2-3 days
- Plumbing: 3-4 days
- Electrical: 2-3 days
- Cabinets: 3-4 days
- Countertops: 2-3 days (plus waiting for fabrication: 2 weeks)
- Tile: 3-4 days
- Painting: 2-3 days

Add up the work days: about 20 days
Some tasks overlap, so maybe 15 working days
Add the countertop wait: 15 + 10 = 25 working days
Add buffer for surprises (30%): about 33 working days
That is roughly 6-7 weeks.

If a contractor says "2 weeks," be skeptical.
If they say "8 weeks," that is realistic.

The Power of Round Numbers

The key technique in back-of-envelope math: round aggressively.

Rules for rounding:
- Round to one significant digit
- 347 becomes 350 or 300
- 8,234 becomes 8,000
- $47.99 becomes $50

Why it works:
- Rounding errors tend to cancel out
- Some numbers round up, others round down
- The final estimate is usually within 2x of reality
- Being within 2x is useful for most decisions

When precision matters more:
- Multiply your estimate by 0.5 and 2
- The real answer is probably in that range
- Example: Estimate is $4,000
  Real cost is probably between $2,000 and $8,000
  Still useful for deciding if the project is feasible

Connecting to Technology

Estimating Storage Needs

Scenario: Your app stores user photos.
You expect 100,000 users, each uploading 2 photos per week.

Quick estimate:
- Average photo size: about 3 MB
- Photos per week: 100,000 users x 2 = 200,000 photos
- Storage per week: 200,000 x 3 MB = 600,000 MB = 600 GB
- Storage per month: 600 x 4 = 2,400 GB ≈ 2.4 TB
- Storage per year: 2.4 x 12 ≈ 29 TB

Now you know:
- You need terabytes, not gigabytes
- A single hard drive will not cut it
- Cloud storage costs roughly $0.02 per GB per month
- Monthly cost: 2,400 x $0.02 = $48/month (first month)
- But storage accumulates, so month 12: 29,000 x $0.02 = $580/month

This took 2 minutes and gives you a budget range.

Estimating Server Capacity

Scenario: Your web service handles user requests.
You expect 1 million daily active users.

Quick estimate:
- Users are not all active at once
- Peak hours: roughly 8 hours of heavy usage
- Peak requests: 1,000,000 users / 8 hours = 125,000 per hour
- Per second: 125,000 / 3,600 ≈ 35 requests per second
- But each user makes multiple requests per session: say 10
- Adjusted: 35 x 10 = 350 requests per second at peak

A single web server can handle roughly 500-1000
simple requests per second.

Conclusion: You need 1-2 servers for normal operation,
maybe 4-5 for safety margin and redundancy.
You do NOT need 100 servers. This quick math
prevents over-provisioning by 20x.

Estimating Bandwidth Requirements

Scenario: A video streaming service.

Quick estimate:
- Video stream: roughly 5 Mbps per viewer (HD quality)
- Concurrent viewers at peak: 10,000
- Total bandwidth: 10,000 x 5 Mbps = 50,000 Mbps = 50 Gbps

A standard server network connection: 1-10 Gbps
You need: 5-50 network connections at minimum
Plus CDN (Content Delivery Network) in front

This estimate tells you immediately:
- This is a serious infrastructure problem
- You cannot host this on a single server
- CDN costs will be a major expense
- Budget for hundreds of thousands per year in bandwidth

Estimating Database Query Performance

Scenario: You have a database with 10 million rows.
How long will a search take?

Without an index (scanning every row):
- Reading speed: roughly 1 million rows per second
- Time: 10,000,000 / 1,000,000 = 10 seconds
- That is way too slow for a user-facing query

With an index (binary search):
- Binary search on 10 million rows: log2(10,000,000) ≈ 23 steps
- Each step: roughly 0.1 milliseconds (disk read)
- Time: 23 x 0.1 = 2.3 milliseconds
- That is fast enough

Conclusion: You need an index on this column.
The back-of-envelope math made the answer obvious
without running any benchmarks.

Common Quick Estimates for Tech

Useful numbers to memorize:

Time:
- 1 day ≈ 100,000 seconds (actually 86,400)
- 1 year ≈ 30 million seconds (actually 31,536,000)

Data:
- 1 KB ≈ a short email
- 1 MB ≈ a photo or a minute of compressed audio
- 1 GB ≈ a movie or 1,000 photos
- 1 TB ≈ 1,000 movies

Network:
- Typical home internet: 100 Mbps
- Downloading 1 GB at 100 Mbps: about 80 seconds
- Downloading 1 TB at 100 Mbps: about 22 hours

Scale:
- A thousand (10^3): a small school
- A million (10^6): a city
- A billion (10^9): a large country or global platform

The Sanity Check

The most valuable use of back-of-envelope math is not planning — it is catching nonsense.

Someone claims: "Our app will have 500 million users
in the first year."

Sanity check:
- World population: 8 billion
- Smartphone users: about 4 billion
- 500 million = 12.5% of all smartphone users worldwide
- Facebook took 6 years to reach 500 million
- This claim is almost certainly unrealistic

Someone claims: "This project will cost $5,000."

Sanity check:
- Project requires 3 developers for 2 months
- Developer salary: roughly $100,000/year = $8,500/month
- 3 developers x 2 months x $8,500 = $51,000
- $5,000 is off by 10x. Something is wrong with the estimate.

How to Get Better at This

Practice tips:

1. Estimate before you look up the answer
   - How many people live in your city?
   - How far is it to the next town?
   - How much does a new car cost?

2. Work from what you know
   - You know your rent, your salary, your commute time
   - Build estimates from these anchors

3. Break big questions into small ones
   - "How much will this cost?" is hard
   - "How many hours?" x "Cost per hour?" is easier

4. Check your answer against reality
   - Does this number feel reasonable?
   - Is it the right order of magnitude?
   - Would I bet money on this being within 2x?

Common Pitfalls

  • False precision. Saying "we need 2,847 GB of storage" when your inputs were rough estimates. Say "about 3 TB" instead. False precision gives false confidence.
  • Forgetting to sanity-check. Always ask: "Does this number make sense?" If your estimate says a project takes 10,000 hours, check — that is 5 work-years.
  • Not accounting for growth. Storage, users, and data accumulate. An estimate for day one is not an estimate for month twelve.
  • Rounding all in the same direction. If you round every number up, your estimate will be way too high. Mix your rounding for better accuracy.
  • Skipping the estimate entirely. Many expensive mistakes happen because nobody spent 2 minutes checking whether the plan was even feasible.
  • Treating the estimate as a commitment. Back-of-envelope math gives you a range, not a promise. Use it to guide decisions, not to set exact budgets.

Key Takeaways

  • Back-of-envelope math takes minutes and prevents hours of wasted effort on infeasible plans.
  • Round aggressively. One significant digit is usually enough. Rounding errors tend to cancel out.
  • Break big questions into smaller ones you can estimate, then multiply.
  • The most important use is the sanity check: catching numbers that are off by 10x before you commit to them.
  • In tech, quick estimates for storage, bandwidth, and server capacity prevent both over-provisioning and under-provisioning.
  • Being approximately right is far more valuable than being precisely wrong.