ARPU vs Average Order Value: Two Different Denominators

Both measures divide revenue by a count, and that is all they share. Average revenue per user counts people. Average order value counts orders. One quantity joins them, orders per person, and once that is in hand the rest follows: when people order more often, average revenue per user rises and average order value does not move at all.

Two averages that sit next to each other on the same slide, carry the same currency, and are built from the same revenue line will be read as two views of one thing. The two measures are not two views of one thing. A division is defined by what sits underneath the line, and these two put different populations there. Average revenue per user divides by a head count. Average order value divides by a transaction count. Setu Bazaar, an invented marketplace, carried 50,000 buyers and 5,00,000 orders in the year covered here. Two counts ten times apart cannot divide one revenue figure into the same answer. One quantity links the two, and each of the two hides something the other shows. Both are defined under average revenue per user.

What does each of the two measures actually count?

Almost everything about the two is identical. Start there. Setu Bazaar reported revenue of Rs 20,00,00,000/- across the year, or Rs 20 crore in the shorter unit, and that one figure is the numerator of both measures. Nothing about it is in dispute. The revenue line is audited, earned over one trailing twelve monthsThe twelve months ending at the most recent reporting date, rather than a calendar year or a financial year. Companies use the window so that a measure can be refreshed every quarter without waiting for the year to end., and both measures reach for it without changing a rupee of it.

Now look underneath. Average revenue per user puts 50,000 buyers there and returns Rs 4,000/- for each buyer across the year. Average order value puts 5,00,000 orders there and returns Rs 400/- for each order, when it is measured on the revenue Setu Bazaar keeps, or Rs 10,000/- for each order when it is measured on the Rs 500 crore of goods that passed across the marketplace. The only difference between the two measures is the denominator, and the only difference is the entire difference. There is nothing else to learn about how the two differ. Everything else about the two measures is a consequence of that one line.

The same split is easier to hear outside a marketplace. A tea stall takes Rs 6,000/- across a morning. Working out what one regular customer is worth means dividing by the forty regulars who came. Working out what one cup fetches means dividing by the two hundred and twenty cups poured. Both answers are true, both are in rupees, and anyone who quotes one when asked for the other has answered a question nobody put. The person and the cup stand in front of the stall owner as visibly different objects. He would never confuse them. On a slide, the two denominators are invisible and the confusion becomes easy.

What single quantity connects average revenue per user and average order value?

Orders per person. The bridge between the two measures is that one quantity and nothing more, and it is worth writing out slowly. Setu Bazaar's 50,000 buyers placed 5,00,000 orders between them, so each buyer placed ten orders across the year. The average order value measured on revenue, Rs 400/-, multiplied by those ten orders gives Rs 4,000/-. Average revenue per user is Rs 4,000/- exactly.

The other side works the same way. The relationship is not an accident of the base chosen. Average order value on gross flow is Rs 10,000/-. Ten orders a person gives Rs 1,00,000/- of goods bought per buyer across the year, and Rs 500 crore of flow spread across 50,000 buyers is indeed Rs 1,00,000/- each. The bridge is not a fact about Setu Bazaar at all, and it works on either base.

The bridge is an identity, not a cross check. The distinction matters more than it sounds. Revenue divided by orders, multiplied by orders divided by people, is revenue divided by people: the orders cancel, and what is left is the definition of the other measure. The two sides are the same quantity written two ways, so they can never disagree, and getting the same answer twice confirms nothing at all about whether Setu Bazaar counted its buyers properly. The identity earns something structural instead. The bridge shows that average revenue per user is built out of two independent ingredients, how large an order is and how many orders a person places, and that anything happening to either ingredient will show up in it. A fact about the architecture of the measure is worth far more than a false sense of verification.

One more consequence falls straight out of the identity and it surprises people. Because average revenue per user is a weighted averageAn average where each item is counted in proportion to its size rather than once each. Total revenue divided by total orders weights a large order more heavily than a small one. A simple average of the order amounts would not. of order sizes multiplied by a frequency, the head count itself vanishes from the arithmetic. Setu Bazaar could sign up another 50,000 buyers who behave exactly like the first 50,000, and neither measure would move by a rupee. Both are per-something figures, so the something cancels. Neither of the two says anything about scale, then, and a business reporting a rising average revenue per user has said nothing about whether it is getting bigger.

Why does one measure move while the other stands still?

Hold the size of an order exactly where it is and let people come back more often. One experiment settles the difference between the two measures, and a reader who watches it once will not confuse them again. Setu Bazaar's buyers placed ten orders each. Suppose they place twelve, and suppose every one of those orders is worth the same Rs 10,000/- of goods as before. Orders rise from 5,00,000 to 6,00,000. Gross flow goes to Rs 600 crore from Rs 500 crore. Revenue, at the same take of 4.00 per cent, rises from Rs 20 crore to Rs 24 crore.

Now divide. Average revenue per user is Rs 24,00,00,000/- over 50,000 buyers, or Rs 4,800/-, up from Rs 4,000/-. Average order value is Rs 6,00,00,00,000/- over 6,00,000 orders, or Rs 10,000/-, exactly what it was. Frequency lives in one denominator and not the other, so it lifts average revenue per user in strict proportion and leaves average order value untouched. Twenty per cent more orders a person, twenty per cent more revenue a person, and not one rupee of change in what a transaction is worth.

The everyday version is a neighbourhood grocer. A customer who used to shop on Sunday starts shopping on Sunday and Wednesday, buying the same basket each time. The till receipt looks identical every single visit, and the grocer's takings from that household have doubled. Nothing about the receipt could ever have shown that, and nothing about the household's repeat rateThe share of buyers in a period who had already bought in an earlier one. The rate describes how many people come back rather than how much they spend when they do. could ever be read off the receipt. Two measures, two different objects, and a receipt has no household printed on it.

What happens when the order gets bigger instead of more frequent?

Run the mirror experiment. Hold frequency at ten orders a person and let each order grow instead, from Rs 10,000/- of goods to Rs 12,000/-. Orders stay at 5,00,000 because nobody is shopping more often. Gross flow still rises to Rs 600 crore and revenue still rises to Rs 24 crore, exactly as in the frequency case, so the top line cannot tell the two stories apart. But underneath, the two measures now behave identically. Average revenue per user goes to Rs 4,800/-, up a fifth. Average order value goes to Rs 12,000/- on flow and Rs 480/- on revenue, also up a fifth. Both moved, and they moved by the same proportion.

The two measures agree only when orders per person is held fixed, so a reader who sees them agree has learned that frequency did not change. That is not a small consolation prize. The finding is genuine and obtained for free, and it runs the other way too: a reader who sees them disagree has learned that frequency did change, and by exactly the ratio between them. The two measures are independent instruments and the bridge is the only thing that ties their readings together. An extra line itemOne product row inside a single order. An order for three different things carries three line items. A fourth line item makes the order larger with no extra visit from the buyer. added to an existing order makes the order larger without making the buyer come back, which is why a business trying to grow can pull either lever and the two measures will report the choice differently.

Which question does each measure answer?

The comparison stops being a curiosity here and starts being useful. Average revenue per user answers one question and one question only: what is a single person worth to this business across a reporting period. The worth of one person across a period is a question about a relationship, and average revenue per user is the right figure whenever the decision at hand concerns people, including a weighing of what it costs to acquire a buyer against what that buyer brings back.

Average order value answers a completely different question: how large is a single transaction at the moment it happens. Transaction size is a question about an event, not a relationship. Average order value is the right figure whenever the decision concerns the transaction itself, and that covers a surprising amount of operational life: what a delivery costs to fulfil, whether a free shipping threshold is set above or below what people typically spend, how much packaging a typical order needs, whether a payment method makes sense at that ticket size.

A measure answers only the question its own denominator asks, and almost every misuse of these two is somebody answering the other question. Notice how cleanly the test works. Setu Bazaar asks whether it is worth spending more to bring in buyers. Bringing in buyers is a question about a person, so average revenue per user is the figure, and average order value has nothing to contribute. Setu Bazaar asks whether it can afford to deliver free above a certain order size. Free delivery above a size is a question about a transaction, so average order value is the figure, and average revenue per user has nothing to contribute. There is no third kind of question hiding here, and no measure that answers both.

When does each of the two measures mislead?

Average revenue per user misleads when the head count underneath it changes definition. Every buyer figure is somebody's counting rule, and the rules available are all defensible and all different: everyone who ever registered, everyone who bought at least once this year, everyone who bought at least once this month. Setu Bazaar's 50,000 is one such rule. Narrow the rule to an active accountAn account meeting a stated activity test in a period, such as one purchase in the last ninety days. Every company sets its own test, so the same word can cover very different populations. test and the count falls, so the same Rs 20,00,00,000/- lands on fewer heads and the average rises without a single buyer spending more. Widen it to include every dormantA registration nobody has used for however long the company decides to allow. Registrations like these swell a buyer count while adding nothing at all to the revenue being divided across it. registration and the average falls. Neither move says anything about buyers. DeduplicationCollapsing several accounts that belong to the same person into one before counting. Skipping the collapse inflates the head count. Doing it thoroughly can shrink a reported user base overnight. does the same thing again, quietly, because two accounts held by one household are two heads until somebody decides they are one.

Average order value misleads for a different reason, and it is the reason every average has. Any average is a summary of a distribution, and a summary throws the distribution away. Setu Bazaar's Rs 10,000/- an order is consistent with five lakh orders of Rs 10,000/- each, and it is equally consistent with a long tailThe thin far end of a distribution, where a small number of very large or very small observations sit. A long tail pulls an average away from what a typical observation looks like. of small orders sitting beside a handful of enormous ones. The medianThe value sitting exactly in the middle when every observation is lined up from smallest to largest. Half the observations fall below it and half above, so a few extreme values cannot drag it around. order would separate those two worlds instantly and the average never will.

Both measures inherit every weakness an average has, and neither of the two escapes it by having a different denominator. Changing what sits under the line changes which question is being answered; it does not buy any protection from the flattening that an average performs. Average revenue per user hides the spread across people. Average order value hides the spread across orders. The choice of denominator buys relevance, not safety, and any reader taught to distrust one of these two for being an average should distrust the other one for precisely the same reason.

How does a working analyst decide which of the two to reach for?

By asking one question before looking at either number: is the question about a person or about a transaction. There is no third answer, and once the answer is given the measure follows without any further judgement. One question sounds too simple to be a method. The mistakes it prevents are the ones that actually get made, so it survives contact with real work better than most methods do.

Watch it run in three places. A lender judging how much short term credit a marketplace can carry needs a view of the goods it will push through next year. Goods pushed through a marketplace is a question about transactions and how large they are, so average order value is the input and the buyer count is background. An equity analyst modelling what happens if acquisition spending doubles wants to know what an incremental buyer brings back. An incremental buyer is a person, so average revenue per user is the input. An operator setting a free delivery threshold is asking about a transaction; an operator setting a loyalty budget is asking about a person; and the same two numbers serve both, in opposite roles, in the same afternoon. The denominator is not a technical detail buried inside the formula, it is the statement of what the measure is about, and reading it off is the whole of the skill. That rules something out. Each of the two is the only correct answer to its own question and useless against the other. Neither is the better measure, and no ranking of them is available at all until somebody names the question. Any source that ranks them has quietly decided which question was being asked.

What do the two measures do on one published set of figures?

Everything above in one table, on Setu Bazaar's own numbers, with every component published so that either route can be recomputed rather than taken on trust. The middle column raises frequency and holds the order steady. The right column holds frequency and raises the order. Both columns land on the identical Rs 600 crore of flow and the identical Rs 24 crore of revenue. The top line cannot separate them and the two measures underneath it can.

Read the bottom three rows across. In the middle column, average revenue per user has moved and both readings of average order value have not moved at all. In the right column, all three have moved and they have moved by the same fifth. One set of accounts, two ways of getting to Rs 24 crore, and the pair of measures tells them apart while the revenue line cannot. Setu Bazaar's own bridge sits in the second row: ten orders a person in the outer columns, twelve in the middle one, and that row alone predicts which of the two readings will move before anything is divided.

What happens when the three inputs move?

The panel below carries the three quantities that decide both measures and moves them separately: how many buyers are counted, how many orders each of them places, and how large one order is. Everything else is arithmetic. The dashed lines drawn across the bars are Setu Bazaar's published readings, pinned there so that a glance shows which bar has left its own line and which one has not. A setting can be reached where one bar has moved and the other sits exactly on its dashed line, and so can the reverse; with the buyer slider all the way down, the panel refuses to divide.

Two settings are worth reaching deliberately. Set the order size to Rs 12,500/- and the frequency to eight orders. A larger order and fewer of them cancel exactly, so average revenue per user lands back on its dashed line at Rs 4,000/- and average order value sits well above its own. Then set frequency to twelve and leave the order at Rs 10,000/-, and the opposite happens. Both states are ordinary, both would be reported honestly, and in each of them one measure would tell a reader that nothing had changed while the other said something had.

Here is the same collision as it actually arrives in working life, in the form of a monthly trading review nobody would question.

Setu Bazaar and Anjani Stationers Private Limited are invented.
Educational material. Not advice on any investment, tax, budget or market position.