Structuring an estimate

Estimation questions in product manager interviews usually open with something vague, like "Estimate EV charging demand in the US." The person asking may not know the true figure either, so there may be no answer key to match. The job is like that too: before a launch, a support plan only needs to know whether to expect 200 extra tickets a week or 20,000. Cracking the PM Interview is plain about what gets judged: how you reason through the problem and how you handle numbers.

Pin down what is being sized

Start by saying the prompt back in your own words, which exposes any gap between what was asked and what you took in. Then look for terms that could mean two things. "Charging demand" could mean several, and the answers would not even share a unit:

Had the prompt been about dollars, one more call would matter: what drivers pay at the charger, or what the operator keeps after paying for power.

Ask about anything that would change your method, and attach your default so the interviewer can settle it in a word. A string of open questions hands the work back. Out loud:

"So, charging demand in the US. I'll take that to mean sessions that actually happen at public fast chargers on an ordinary day this year. I'll count battery-only cars, since most plug-in hybrids can't use fast chargers. Tell me if you had something else in mind."

For practice on this step alone, write a two-sentence scope for Estimation Case: Counting Contractors Who Work on U.S. Homes before doing anything else.

Sketch two routes before you commit

My first instinct on this question is to start from the chargers: sessions at one fast-charging port per day × number of ports. I can picture a busy port's day well enough. Then I reach "number of ports" and have nothing to build it from, and all that thinking about one port has led nowhere. So before any arithmetic, I put a second route beside the first and ask of every term in both: where would this number come from?

The two routes have names. Top-down starts from a big total you can anchor on and cuts it down. Bottom-up starts from one unit, such as one charger or one customer, and multiplies up. Our longer article on market sizing compares the two in general. Here is how they look for this question:

Write those last two rows for any question: facts you hold and facts you're missing. They often decide the route.

Pick the route whose hardest number you could still argue for. I have a feel for how common EVs are and how a driver spends a week, and none for the national port count, so I go top-down. Someone who has installed chargers might know that count and go the other way. The route you skip can come back later as a check.

Write the structure in words

The equation carries the most weight in the book's method. Numbers spoken too early pull you into arithmetic before anyone has agreed on the logic, so write it in words:

Daily sessions = EVs on the road × fast-charging sessions per car per day

Then read the units across. Cars times sessions per car per day leaves sessions per day, the unit we agreed on. If the units don't reduce to your target, a term is missing or wrong.

Pause and ask whether the structure looks reasonable. A structure takes one sentence to correct.

Find the driver

The driver is the quantity that makes one more unit of your answer happen. Ask that literally: what makes one more fast-charging session happen? A car running low, away from any plug it can use overnight. Charger count looks like a driver, but chargers mostly serve demand that cars create. So the question becomes how often a car ends up in that spot, and the biggest factor I can think of is whether it has a plug at home. That is the split worth making.

Split only when the split changes the answer

Splitting a population into groups, called segmenting, costs time and adds assumptions. Use one test: do these groups differ enough on the rate in my equation to move the total?

Use one split at a time. Crossing two multiplies your guesses for little gain.

Neither term of the equation can be guessed whole, so each gets a small equation of its own. Every figure below is my own assumption unless I say otherwise.

• EVs on the road: 6 million. Take all US cars × share that are electric: say 280 million cars with 1 in 50 electric, so 280 million ÷ 50 = 5.6 million. I chose 1 in 50 knowing the published count below, so its closeness proves little. US homes × share with an EV, 130 million × 4% = 5.2 million, is the same guess over a different denominator, so the two figures aren't agreeing, but neighbors may be easier to picture than traffic. In the room, that is as far as you can get. Here I can look up a published count: the US Department of Energy's Alternative Fuels Data Center lists about 5.7 million all-electric vehicles for 2025, all light-duty registrations, so company cars are in it. I'm rounding up for sales since then.

• Can plug in at home: 80% of 6 million, which is 4.8 million cars. They fast-charge mainly on long trips: 3 trips a year × 4 stops a trip = 12 sessions a year. 12 ÷ 365 is about one session every 30 days, so 4,800,000 ÷ 30 = 160,000 sessions a day.

• No home plug: 20% of 6 million, which is 1.2 million cars. Say 12,000 miles a year, with half charged at fast chargers: 12,000 × 50% = 6,000 miles. At 120 miles a session, 6,000 ÷ 120 = 50 sessions a year, about one a week. 1,200,000 ÷ 7 ≈ 171,000 sessions a day.

The total is 160,000 + 171,000 = 331,000, or about 330,000 sessions a day.

Look at what the split bought. A fifth of the cars produce about half the sessions (171,000 ÷ 331,000 ≈ 52%). Keeping car counts and per-car rates apart until the last step of each line shows why: the no-plug group has a quarter as many cars (1.2 million ÷ 4.8 million) at about four times the rate (50 sessions a year ÷ 12). Had I applied the home-plug habit to every car, I'd have 6,000,000 ÷ 30 = 200,000, about 40% lower, with nothing on the page to warn me.

Hunt for what the equation leaves out

Now attack your own model. Ask who behaves unlike the typical car in each group, and what else produces the thing you're counting that the equation never mentions. Here, rideshare drivers charge far more often than my model says, rental cars add road-trip sessions, and second cars that barely move pull the other way.

Size the biggest one roughly. Assume 1% of EVs do rideshare work and none of those cars has a home plug: 6,000,000 × 1% = 60,000 cars. If they fast-charge five times a week against the no-plug group's one a week, that adds 60,000 × 4 ÷ 7 ≈ 34,000 sessions a day, about 10% of 331,000. That is worth naming, and too small for a third group.

My judgment is that rental cars are too few to model, so I leave them out. Say that out loud, so the interviewer knows it was a choice. Raising a hole yourself shows you looked, and one you pass over in silence can look like one you never saw.

When you get stuck

When a term stalls you, try these, roughly in order:

1. Shrink it to one. If a whole fleet's habits feel unknowable, picture one driver's week: how far they go and when the battery runs low.

2. Borrow a place you know. Take one apartment building you can picture, work the term out for it, then scale it up.

3. Break the term in two. "Share with no home plug" is hard to guess. "Share of EV owners who rent" times "share of renters with no plug at their building" gives you two smaller things to reason about.

4. Ask for a moment. "Give me thirty seconds to sketch this" is a normal thing to say.

If a term is still a guess after that, say so, pick a number, mark it as low confidence and keep going.

To practice this opening stretch, try Market-Sizing Estimate: Rain-Driven Umbrella Purchases in One City or Market sizing: how many people in India hold a credit card, and stop as soon as your equation is written.