Estimation questions
Reusable building blocks for market sizing questions
In this lesson11 sections
An unofficial lesson drawing on the methods in the Estimation Questions chapter of Cracking the PM Interview by Gayle Laakmann McDowell and Jackie Bavaro (2014). PracHub is not affiliated with the authors or publisher. Buy the book
Market sizing questions in PM interviews reuse a handful of sub-estimates: who actually uses the product, how often they buy, what they pay and who ends up with the money. Know these parts before you walk in and your time goes into choosing numbers, not inventing method.
The question:
"How much do people in the US spend each year fixing cracked phone screens?"
Scope it first, as in Structuring an estimate. I'll take consumer spending in one year, on screens only, counting what people pay at the counter. In words:
Spending = phones in use × share that crack in a year × share of cracks that get fixed × average price of a fix
Start from the people who use it
The US has about 340 million people, but not 340 million phones. Toddlers don't carry one. Many older people have no smartphone, or leave it on the kitchen table where it rarely falls. Apply your own crack rate to everyone and you charge children and light users for drops they never make.
So before any rate, remove non-users and give light users a lower rate. Count phones in use, not phones sold: sales miss the three-year-old phone in someone's pocket, which cracks just as easily.
"I'll count phones people actually carry, one per person, and leave out young children. Older users get a lower crack rate, since their phones mostly stay home."
Size an age group from population and lifespan
Two numbers you already carry give a head count for any age group: about 340 million people and a life expectancy of about 79 years. Spread people evenly across ages and each year of age holds 340 ÷ 79 ≈ 4.3 million. A band is that times its width in years.
| Group | People | Share with a smartphone (my guess) | Phones in use |
|---|---|---|---|
| Under 8 | 8 × 4.3 = 34 million | Treat as zero | 0 |
| 8 to 17 | 10 × 4.3 = 43 million | 65% | 28 million |
| 18 to 64 | 47 × 4.3 = 200 million | 95% | 190 million |
| 65 and over | 14 × 4.3 = 60 million | 75% | 45 million |
That's about 263 million phones, from bands that add to 337 million people. The shortcut runs high for children, since fewer babies are born now than the average implies (Building numbers you can defend measures the gap).
For a stage of life rather than an age band, divide the years people spend in the stage by the years it could happen in. Phones crack most in hands-on jobs such as delivery and warehouses. If a typical worker spends 5 of about 47 working years in one, about 1 working-age adult in 10 is in one now: 10% of 190 million is 19 million phones.
To practise this step, try Sizing the homeowner population within the US millennial generation and stop once you've sized the group.
Turn group rates into one rate, counting each person once
These yearly crack rates are my assumptions:
Ages 8 to 17: 25%. 28 million × 25% = 7 million.
Hands-on workers: 30%. 19 million × 30% = 5.7 million.
Other working-age adults: 12%. 171 million × 12% ≈ 20.5 million.
65 and over: 5%. 45 million × 5% ≈ 2.3 million.
That's about 35 million cracks a year, an overall rate of 35 ÷ 263 ≈ 13%, weighted by how many phones each group holds. The plain average of the four rates is 18%, which gives 47 million cracks, a third too many, because it lets two small, clumsy groups count as much as 171 million careful adults.
The hands-on workers came out of the working-age group. Leave all 190 million working-age phones at 12% and add the workers on top, and 19 million people sit in two groups: 282 million phones against 263 million. Before you multiply, add up your groups. A total above your starting population means someone is counted twice.
Keep a job split on top of an age split only if it moves the total. This one adds about 3.4 million cracks (19 million phones at 30% rather than 12%), a tenth of the answer.
Removing non-users mattered too: 13% of all 340 million people would be about 45 million cracks, nearly a third more.
Repair, replace or live with it
Some people live with a cracked phone, and some buy a new one. Another block helps: for anything owned and replaced on a cycle, yearly purchases are about owners ÷ years between purchases, plus first-time owners.
Assume people keep a phone about three and a half years. Then over a quarter of cracks hit a phone in its last year, and many of those owners will upgrade instead. I'll say a quarter of cracks end in a new phone and 30% in living with it, leaving 45% fixed: 35 million × 45% ≈ 16 million repairs.
Check the replaced share with the same block. Per 100 phones in use, 100 ÷ 3.5 ≈ 29 are replaced each year. My model has about 13 cracks per 100, a quarter of them replaced, so about 3: cracks cause about 1 upgrade in 10, which is believable. Had it said 25 of the 29, I'd revisit the crack rate.
To practise turning ownership into yearly purchases, try Market-sizing estimate of yearly vacuum cleaner purchases in the United States or Market Sizing: Estimate Yearly Television Buyers in New York for a TV Maker.
Market-sizing estimate of yearly vacuum cleaner purchases in the United States
A market-sizing exercise asking how many vacuum cleaners are bought in the United States each year.
Market Sizing: Estimate Yearly Television Buyers in New York for a TV Maker
A market-sizing question that asks how many TV buyers New York has in a year, framed for a company that makes televisions and semiconductors.
When prices vary, split them into tiers
A screen fix can cost a small plan fee or several hundred dollars at the maker's counter. No single price is typical, so split repairs into tiers and weight each tier's price by its share. My assumptions:
| Where the fix happens | Share of repairs | Price |
|---|---|---|
| Protection plan fee | 30% | $30 |
| Maker or authorized shop, no plan | 20% | $280 |
| Independent shop | 50% | $120 |
Weighted: 30% × $30 + 20% × $280 + 50% × $120 = $9 + $56 + $60 = $125. Spending is 16 million × $125 = $2 billion a year.
The plain average of the prices, about $143, would give $2.3 billion; the maker's $280 alone, $4.5 billion. The shares carry the answer, so justify them: half the work goes to independent shops because they're cheaper and in most strip malls. I counted plan fees at the counter, not monthly premiums: a scope call to say aloud.
"About $2 billion a year. The mix between tiers matters more than any price, and I'd check the independent shops' share first."
To practise this step, try Market-Sizing Estimate of Total Yearly Consumer Spend at US Cinemas and settle your tiers before your first multiplication.
Whose revenue is it?
Spending is what customers pay; the $2 billion passes through several hands. Ask which one the interviewer means:
"By revenue, do you mean what customers pay, or what the phone maker books? I'll size customer spending, then split it by who keeps it."
Then strip out what each layer in between keeps:
Plan fees and official repairs: 4.8 million × $30 + 3.2 million × $280 ≈ $1 billion. That's a ceiling for the phone maker, since authorized shops keep part of each repair and carrier plans run by insurers pay the fee to the insurer.
Independent shops take in 8 million × $120 = $960 million. Say $45 of each $120 pays for the part, so the shops keep 8 million × $75 = $600 million.
The $45 splits again. Parts sellers often quote their cut as a markup on cost. At 25%, a seller pays $36 and charges $45, keeping $9: a fifth of the price, not a quarter. Screen makers book about 8 million × $36 ≈ $290 million.
Ad revenue starts from active users
A possible follow-up: what would in-app ads earn for a free app that books repairs? Registered accounts are the wrong start, since many people install an app, open it twice and never return. Start from monthly active users:
Yearly ad revenue = monthly active users × sessions per user × ads per session × price per thousand ads ÷ 1,000 × 12
Say 20 million accounts, 5 million of whom open the app in a month, 6 times each, seeing 2 ads a visit: 5 million × 6 × 2 = 60 million ads a month. Ads in apps and feeds now sell mostly per thousand impressions (CPM) or per result, such as an install, rather than per click, so price per thousand. At $5, that's 60,000 × $5 = $300,000 a month, about $3.6 million a year. Starting from all 20 million accounts would give $14 million, four times too much.
To practise this step, try Revenue guesstimate for Google's AdSense publisher ad network and write the equation before any numbers.
Headcount has a floor
Most candidates size a team by dividing the work by what one person does. That's half of it. Take technicians at independent shops. The work: 8 million repairs × 45 minutes = 6 million hours, and at about 1,800 hours a year each, that's 3,300 technicians.
Now the floor. Someone must be there whenever the service is open, busy or not, so the minimum is people on duty at once × hours open per week ÷ hours one person works, plus cover for leave. Assume one shop per 30,000 people, about 11,000 shops, each open 70 hours a week with one technician on duty: 1 × 70 ÷ 40 rounds up to 2 a shop, so 22,000 technicians. The floor beats the work estimate nearly seven times over, so it's the answer, and it says each technician fixes about one screen a day. Either shops live on other work, such as batteries and cases, or my shop count is high.
A repair-booking line that never closes, with three agents answering at once, needs 3 × 168 ÷ 40 ≈ 12.6 people, about 15 once holidays and sick days are covered.
Then add support staff, such as a manager or a parts buyer: at one per three technicians, 22,000 becomes about 29,000.
To practise this step, try Ride-Hailing Supply Estimate: Active Uber Drivers Serving a Mid-Sized American City and work out both a number from the work and a floor before you choose.
Pack by the carton grid, not the phone's volume
The same habits work on physical questions: "How many boxed phones fit in a 40-foot shipping container, and what does it weigh loaded?"
The quick answer divides the space by one phone box. Inside, a 40-foot container is about 12 m long, 2.35 m wide and 2.4 m high. I'll assume 11.8 by 2.35 by 2.3 m is usable, leaving a gap under the roof and at the doors: about 64 cubic meters, or 64,000 liters. My last phone box was about 16 × 8 × 4 cm, about half a liter. 64,000 ÷ 0.5 gives 128,000 phones.
That's too many. Phones ship in cartons, and cartons don't fill the space exactly. Say a carton holds 40 phones with dividers and measures 34 × 34 × 22 cm outside. Now fit cartons along each inside dimension. Across 2.35 m, 6 fit, leaving a 31 cm strip down the whole length that no carton can use. Along 11.8 m, 34 fit. Up 2.3 m, 10 layers.
6 × 34 × 10 = 2,040 cartons, and × 40 = 81,600 phones. So each phone really takes 64,000 ÷ 81,600 ≈ 0.8 liters, half as much again as its box: the dividers, the carton walls and the strip at the side wall all count against it. Divide by the packed space per item, never the item's own volume.
Weight or space: which runs out first
The grid says 81,600 phones fit. Now check the container can carry them, starting from the cargo. Boxed phones at about 0.3 kg each make 81,600 × 0.3 ≈ 24.5 tonnes, and 2,040 cartons at about 1 kg each with dividers add 2 tonnes.
Don't build the container's own weight from scratch: it's painted on the doors, a little under 4 tonnes for a 40-footer. Loaded, that's about 30 tonnes. With no plate to read, you'd size the steel from area, thickness and density (about 8 times water), but the cargo is most of the total, so a rough figure will do.
The same door gives the maximum gross weight, around 30 to 32 tonnes for most 40-footers. So weight and space run out at about the same point, and the road decides: road limits can be lower than the plate. Where they are, weight is the limit, and the full grid of 81,600 can't travel. You'd ship fewer cartons and leave space empty, so give the answer as the lower of the two counts.
Loading time turns on how the crew splits the work
Time to finish is the number of cycles times the time per cycle. Here a cycle moves one carton, so there are 2,040. For one person carrying alone, I'll assume 15 seconds a trip, mostly walking: early trips go to the far end, late ones stop near the door, so the walk averages half the container each way. 2,040 × 15 = 30,600 seconds, about 8.5 hours before breaks.
Add a helper who hands cartons in at the door, and the number of cycles doesn't change: every carton still makes the walk. You save only the seconds spent picking each one up, so the answer barely moves. The walk is the limit, and only one person is walking.
Put two people on carrying instead, and each makes 1,020 trips of the same 15 seconds: about 4¼ hours. That holds while both can stack at the face at once, and with 6 cartons across there's room for two. Add more carriers and they queue at the face, so room at the stack, not headcount, sets the answer.
Keep constants simple: a cubic meter is 1,000 liters, and a liter of water weighs a kilogram. Test any finished estimate as in Checking and presenting the answer.
To practise packing, units and weight, try Fermi estimate: total kernel count in a 10 kg sack of rice and Estimation Drill: Weighing the Empire State Building Without a Scale.
Fermi estimate: total kernel count in a 10 kg sack of rice
A classic estimation question for product manager interviews: estimate the number of rice grains in a 10 kg bag.
Estimation Drill: Weighing the Empire State Building Without a Scale
A classic estimation question asking you to work out how much the Empire State Building weighs.