What is the fair way to split gas in a carpool?
Split it by miles ridden, not by heads, and measure those miles once. A commute carpool drives the same route with the same pickup points week after week, so each rider’s distance is a fixed number. Put those numbers into one standing rule, agree on a per-mile rate, and the split for every trip after that is already done. Nobody logs anything. Nobody argues.
That is the whole argument of this guide, and it runs against the advice a carpool expense app gives its own users. The usual recommendation is to split evenly, because splitting by distance means bookkeeping and bookkeeping is what kills carpools. The premise is right. The conclusion is wrong, because it treats the bookkeeping as a cost you pay on every trip. For a route that repeats, it is a cost you pay one time. That assumes what a commute carpool usually has, a stable route and a stable roster; when either changes, you measure again.
The framework comes from an unlikely place: cost accounting. The question of when precision is worth what it costs to measure is one that Harvard Business School’s Robert Kaplan and his co-authors returned to over sixteen years of work, and their answer maps onto four people and a Honda with almost no translation.
Why does every carpool end up splitting evenly?
Because the alternative looks like homework. Take a carpool cost-splitting guide published in July 2026 by GoodShare, an expense-sharing app. It compares three models in a table and scores them honestly: the mileage-based split gets high fairness and medium effort, the even split gets medium fairness and low effort. Then it recommends the even split for most carpools anyway, “with an app doing the bookkeeping,” reserving mileage “only when distances differ a lot.”
The diagnosis behind that recommendation is worth quoting in full, because it is the premise this guide accepts: “The biggest problem in carpools is not the splitting model. It is the bookkeeping. Who filled up when? Who owes whom how much? Most carpools do not fail over money; they fail because the accounting becomes annoying.”
Ratings as published by GoodShare. Source: How to Split Gas Money Fairly: 3 Carpool Cost Models, GoodShare, July 2026
Read that table again with one question in mind: effort per what? The table does not say where the mileage split’s “medium” effort comes from, but the only work in a mileage split is measuring miles, and the rating only makes sense if you measure them every trip, the way you would on a road trip to a place you have never been. A commute is not that. The route is the same on Tuesday as it was on Monday. The riders board at the same driveways. The distance from each front door to the office does not change when the calendar does. Measured once, the mileage split’s effort drops to the same “low” as the even split’s, and its fairness stays “high.” The table’s own logic then points the other way.
What does cost accounting say about paying for precision?
That precision is a purchase, and you should only buy as much as the errors it prevents are worth. In 1988, Robin Cooper and Robert Kaplan opened a Harvard Business Review article with a warning to companies selling many products: their managers were “making important decisions about pricing, product mix, and process technology based on distorted cost information.” Averaging costs across things that consume resources differently produces a number that is easy to compute and wrong for any item that differs from the average. An even split is the same kind of number, produced the same way.
The other half of the trade-off arrived when companies tried to fix the distortion. Activity-based costing promised accuracy, and in a 2004 Harvard Business School working paper Kaplan and Steven Anderson catalogued what it cost to get it. One bank’s brokerage operation “required 70,000 employees at more than 100 facilities to submit monthly surveys of their time,” and “had to provide 14 full-time people just to manage the ABC data collection, processing and reporting.” Their verdict on the whole approach: collecting the data “has proven to be time-consuming and costly,” and because updating the model cost so much, “many ABC systems are updated only infrequently, leading to out-of-date” numbers.
The design principle they draw from this is the one a carpool needs. A cost system can measure an activity by counting occurrences (a transaction driver) or by timing them (a duration driver). “While duration drivers are generally more accurate than transaction drivers, they are also more expensive to measure, so cost system designers have typically used transaction drivers whenever they reasonably approximate resource demands.” And on the measurements themselves: “Precision is not critical; rough accuracy is sufficient.”
Translate it to the car. Counting heads is the carpool’s transaction driver. Counting miles is its duration driver: not time on a clock, but how much of the trip each person actually consumed. The rule, carried over, says: use heads whenever they reasonably approximate who consumed the trip, and switch to miles when they don’t, provided miles are cheap enough to measure. A carpool with four riders boarding at four distances is the case where heads stop approximating. A repeating route is the case where miles become cheap.
One qualification belongs in this frame before the rule touches real money, because the accounting literature does not let a duration driver off the hook that easily. In 1991, Eric Noreen derived the conditions under which a cost-driver allocation gives relevant answers and found them “quite stringent”: the cost in each pool must be “strictly proportional to the level of activity,” which “rules out, at the level of the cost pool, nonlinear cost functions and linear functions in which there are nonzero intercepts.” His test case was a factory: “Is factory rent, for example, strictly proportional to machine hours?” A carpool has the same fault line. Fuel comes closest to passing it; burn twice the distance, burn roughly twice the gas. The costs of simply owning the car do not; they accrue whether or not anyone rides.
Honest boundary. A single per-mile rate blends the strictly proportional cost (fuel) with costs that are not. The IRS builds its standard rate from “an annual study” of “the fixed and variable costs of operating an automobile”: for 2026 that study produced 72.5 cents, 35 of them treated as depreciation. The mid-year bump to 76 cents on July 1 was a fuel adjustment; the IRS says the change “results from recent increases in the price of fuel.” This guide borrows the blend for the same reason a tax form does: one rate, not six. A group that wants Noreen’s stricter standard can split fuel by the mile and split the fixed costs of the car some other way, evenly or by whoever’s name is on the title. Either is defensible. What is not defensible is treating the blended rate as more precise than the trade-off it is.
Sources: Measure Costs Right: Make the Right Decisions, Harvard Business Review, 1988; Time-Driven Activity-Based Costing, Harvard Business School working paper 04-045, 2004; Conditions Under Which Activity-Based Cost Systems Provide Relevant Costs, Journal of Management Accounting Research, 1991; Notice 2026-10, IRS (2026 rate of 72.5 cents set from the annual cost study; 35 cents per mile treated as depreciation); Announcement 2026-11, IRS (76 cents per mile from July 1, 2026, “results from recent increases in the price of fuel”)
Why is a commute carpool different from a road trip?
Because a road trip is measured while it happens and a commute is measured before it starts. On a one-off drive to a wedding three states away, nobody knows the mileage until the odometer reports it, every fuel stop is a new receipt, and the fairest thing you can do is agree on a per-mile rate and count the driver as one of the riders. Our guide to splitting gas on a road trip covers that case, including why the number you split should be larger than the gas.
A commute has none of that uncertainty. The Census Bureau’s American Community Survey puts the mean one-way travel time to work at 27.2 minutes in 2024, and the Federal Highway Administration’s 2022 National Household Travel Survey puts the average to-or-from-work trip at 13.4 miles. Those are averages across every worker, but the point is structural, not numerical: a work trip is the same trip, twice a day, on a schedule, for as long as the job and the home stay put. The one thing a mileage split needs, each rider’s distance, is a constant.
Sources: United States Commuting At A Glance, U.S. Census Bureau, ACS 1-year estimates, revised September 2025; Commuting in the United States: 2022, U.S. Census Bureau, ACSBR-018; Summary of Travel Trends: 2022 National Household Travel Survey, Federal Highway Administration, Table 3-3
This is what the cost-accounting frame makes visible. Kaplan and Anderson’s bank paid its measurement cost every month because the thing being measured, how employees spent their time, kept changing. A carpool’s measurement, the distance from each door to the destination, does not change. The cost of measuring it is a fixed cost, paid one time, and a fixed cost spread over a year of trips rounds to nothing per trip. That is the difference between “medium effort” and “no effort,” and it is the entire reason the even split is the wrong default for a commute.
How do you measure a carpool route once?
In one sitting, with a maps app, before the first trip. The output is a short table: one line per rider, one number per line. Everything after that is arithmetic the group never has to redo.
Draw the route with its pickup order
Start at the driver's door, add each pickup in the order the car actually collects people, end at the destination. Use the route you really drive, detours included.
Record each rider's miles in the car
For every person, the distance from the point they board to the destination. The driver's number is the full route. Write these down; they are the standing weights.
Agree on the per-mile rate
Gas only, or all-in. The IRS business rate is 76 cents per mile from July 1, 2026, up from the 72.5 cents its annual study of the fixed and variable costs of operating an automobile produced for the year, a revision the IRS attributes to fuel prices; AAA's 2025 study put fuel alone at 13 cents per mile. Pick one, write it down, and do not relitigate it per trip.
Choose passenger-miles or segments
Passenger-miles charges each rider for their own distance out of the total distance everyone rode. Segments charge each stretch of road to whoever was in the car for it. Both are defensible; they differ on who carries the empty first miles. Decide once.
Turn the weights into shares
Each rider's share of a trip is their weight divided by the sum of the weights. That percentage is the rule. It applies to every trip, every fill-up, every month, until the route or the roster changes; then you measure again.
Set the settlement cadence and the rotation
If one person always drives, riders pay their share of the rate. If driving rotates, keep a running score and let the score choose the driver (the algorithm for that is below). Settle on a fixed rhythm (GoodShare's advice is monthly) so nobody has to remember a day.
Step 4 deserves a sentence more, because it is the only judgment call on the list. Under passenger-miles, the driver who lives farthest out pays for the most miles but shares the cost of the empty stretch before the first pickup with everyone. Under segments, that empty stretch is the driver’s alone, the way the hybrid bill split pulls a line with a single owner off the top before it divides anything. Neither is wrong. What matters is that the choice gets made once, on paper, while nobody is annoyed.
What does a measured split look like?
Here is an illustrative four-person carpool. The distances are invented to make the arithmetic visible; the method is the point. Ana drives from her house. Ben boards at mile 3, Cara at mile 8, Dev at mile 11, and the office is at mile 15. The group agreed on the IRS rate of 76 cents per mile as its all-in number, so the group values one direction at 15 × $0.76 = $11.40.
Passenger-miles total 15 + 12 + 7 + 4 = 38, so each passenger-mile costs $11.40 ÷ 38 = 30 cents, and each rider pays 30 cents times their own miles. The segment method gives a different answer from the same measurements: Ana alone owns the first 3 miles ($2.28), Ana and Ben share the next 5 ($1.90 each), three people share miles 8 to 11 (76 cents each), and all four share the last 4 (76 cents each). That lands Ana at $5.70, Ben at $3.42, Cara at $1.52 and Dev at 76 cents. Same $11.40, same fairness principle, different view of the empty miles.
Now look at what the even split does to Dev. He rides 4 of the 15 miles and pays the same $2.85 as Ana, who rides all 15. The gap is $1.65 a trip under passenger-miles. A commute is two trips a day, so at twenty working days that is roughly $66 a month flowing from the rider with the shortest ride to the rider with the longest. Ana, who does all the driving and all the miles, is the one being subsidised. That is the hidden tax of the even split, and in a carpool it is not hidden at all: everyone can see who gets in last.
The measurement that fixes it took one evening with a maps app, and it never has to happen again.
When is the even split good enough?
When the error it makes is smaller than anyone cares about, which depends on three things the group controls: how far apart the pickups are, how many trips a month the rule covers, and what rate the group agreed on. The spread and the trip count multiply the error. The rate scales it.
Rerun the illustrative carpool at fuel only. One direction is 15 × $0.13 = $1.95, and Dev’s overpayment under the even split shrinks to about 28 cents a trip, around $11 a month. Under the all-in rate it was $66. The measurement cost is identical in both cases. Only the value of being precise changed. A group can also price its actual car: AAA’s all-in figure for a small sedan is 55.87 cents a mile, and its six cost categories (fuel, maintenance and tires, insurance, license and registration, depreciation, finance charges) show what the gap between 13 and 76 is made of.
Sources: AAA: New Vehicle Costs Drop to $11,577, AAA Your Driving Costs, 2025 (average fuel cost of 13.00 cents per mile across 45 new-vehicle models; small sedan all-in 55.87 cents per mile); Announcement 2026-11 (Internal Revenue Bulletin 2026-29), IRS (76 cents per mile for business use on or after July 1, 2026)
So the even split earns its place in two kinds of carpool. The first is the one where everybody boards within a few blocks of each other, so the spread is small and the miles rule reproduces the even split anyway. The second is the one that splits only gas, at today’s fuel prices, for a short route, where the monthly error is coffee money. Notice that neither case is “the bookkeeping is too much.” In both, the miles rule is just as cheap to set up; it simply has little to fix.
For most other carpools, including any that price the car at an all-in rate, the question Kaplan’s framework asks has an easy answer. The error cost of counting heads recurs every trip. The measurement cost of counting miles is paid once. With a stable route and a stable roster, a one-time measurement only has to be cheaper than the error it prevents, and the illustration above shows how quickly forty trips a month add up.
What if nobody pays and you just take turns driving?
Then you still need a rule, and the one the literature settled on was published by two IBM researchers in 1983. Ronald Fagin and John Williams set out to schedule a carpool in which people skip days, so simple rotation breaks: the person whose “turn” it is may not be riding. The obvious fix is to keep a separate rotation for every combination of riders who ever share the car. It is fair, and it is unusable. In their words, “the bookkeeping for this algorithm becomes a nightmare (if the number N of people is, say, four or more) because the size of the book grows exponentially with the size of the carpool.”
Their replacement is a single score per person. Pick a unit cost for a trip, call it U (they use 12 for a four-person pool so every entry stays a whole number). On a day when k people ride, the driver’s score goes up by U(k−1)/k and each passenger’s goes down by U/k, so the day sums to zero. Tomorrow’s driver is whoever has the lowest score among the people riding. Someone who stays home is untouched: “no penalty is assessed to a carpool member who does not ride on any given day.” And the ledger stays small. “The amount of bookkeeping grows only linearly with the number of carpool members.”
John +8. Phyllis −4. Ron −4. Don, at home, stays at 0.
Phyllis and Ron are tied at −4; Ron drives. Ron +9, the other three −3 each.
Phyllis (−7) is below Don (−3), so Phyllis drives: Phyllis +6, Don −6.
The scores are a running statement of who owes the pool a drive. Whoever is lowest is next.
Two decades later the computer scientist Moni Naor showed this was not just a tidy heuristic. He set out four basic requirements any “fair share” of driving should satisfy and proved that the Fagin–Williams share “is the unique one satisfying these requirements,” and that it coincides with the Shapley value of a coalitional game built from the carpool. A fair answer, under four stated requirements, was already known; what the 1983 paper contributed was a way to compute it with a ledger a person would actually keep. That is the measurement-cost problem again, solved in the literature forty years before an app told you to split evenly because the accounting was annoying.
Sources: A Fair Carpool Scheduling Algorithm, IBM Journal of Research and Development, 1983; On fairness in the carpool problem, Journal of Algorithms, 2005
The rotation and the miles rule can be combined, though the combination is ours, not theirs: Fagin and Williams and Naor analyse equal shares among the riders present. If the weights are uneven, price the day at U and charge each rider their weighted share of U instead of a flat U/k; the row still sums to zero, and the book is still one row per day and one column per person. The uniqueness proof does not carry over automatically, so treat the weighted version as a bookkeeping convention the group agrees to, not a theorem.
Where does splitty fit in a carpool?
At the receipts. splitty does not measure miles or compute a per-mile rate; the standing rule above is yours to set, and once it is set there is nothing to compute. What a carpool still produces every week is paper: the fill-up one person fronted, the toll statement, the parking garage, the coffee run that became a habit. Those are itemized receipts, and itemized receipts are what splitty splits.
Scan the receipt, and every line starts shared among everyone in the car. Tap to remove the people who didn’t have that item, and tax and any tip land in proportion to each person’s share. Then splitty sends each rider a pre-filled payment request in the app they already use. Save the four of you as a Group and the next receipt takes seconds. Only one person needs the app.
That figure says only that the itemizing habit already reaches past the dinner table. The snapshot behind it has no gas-station, toll or parking category, so this guide claims no carpool-specific number. What it does support is the mechanism: an app built to itemize whatever paper a group hands it does not care whether the line is a shared appetizer or a shared tank of gas.
Honest boundary. splitty settles a bill now. It is not a running ledger, so the Fagin–Williams score lives in your notes app or a shared sheet, and if you want a long-term balance across months, Splitwise tracks and splitty settles. Use the standing miles rule for the car. Use splitty for everything the car stops to buy.
FAQ
Carpool cost splitting FAQ
Common questions about sharing the cost of a commute carpool.
01 How do you split gas money fairly in a carpool?
By miles, measured once. Record each rider's distance from their pickup point to the destination, agree on a per-mile rate, and give each rider a standing share equal to their miles divided by everyone's miles. Because a commute route repeats, those shares apply to every trip without any further tracking. Split evenly only when pickups are close together or the group is pricing gas alone on a short route, where the even split's error is small.
02 Should the driver pay a share of the gas?
Yes, if the driver is also commuting. Count the driver as a rider for the full route; they consume the trip like everyone else. What the driver should not absorb is the car itself. Fuel is a fraction of what a mile costs: AAA's 2025 Your Driving Costs study put fuel at 13 cents per mile, while the IRS all-in business rate is 76 cents per mile from July 1, 2026. If the group only reimburses gas, the driver carries everything else the car costs alone; AAA's all-in figure also counts maintenance and tires, insurance, license and registration, depreciation and finance charges. Whether riders should share those fixed costs is the group's call, as the honest-boundary note above explains. What is not defensible is pretending they don't exist.
03 What per-mile rate should a carpool use?
Whatever the group agrees on in advance, as long as it is written down. The IRS business standard mileage rate, 76 cents per mile from July 1, 2026, is a defensible all-in reference because it is published and dated, and the standard rate it revises is built from an annual study of the fixed and variable costs of operating an automobile, not fuel alone. A fuel-only rate is legitimate too, if everyone understands that it leaves the driver carrying the rest. The choice changes how much the split is worth getting right, not how much work it is.
04 How does a carpool handle someone who skips a day?
With a rule that does not penalise absence. In the Fagin and Williams carpool algorithm, each person keeps a score: on a day with k riders the driver gains U(k−1)/k and each passenger loses U/k, where U is a unit trip cost. Someone who stays home is unchanged, and the next driver is whoever has the lowest score among the people riding. If your carpool pays in money rather than turns, use the standing miles of the riders who were actually in the car as that day's weights and renormalise among them; if the absence also shortened the route (a skipped detour), the driver's miles that day are the shorter route's.
05 Do tolls and parking split the same way as gas?
Split them at the exact amount on the receipt, among the people in the car when the charge happened. Tolls and parking are pass-throughs, not per-mile costs, so they do not belong inside the per-mile rate. The toll statement and the garage receipt are itemized, which makes them the part of a carpool that a receipt-splitting app handles directly.
