Customer Segments and the Journey: The Average Buyer Does Not Exist

A customer segment is a group of buyers that behave the same way on one named measure. Setu Bazaar's 50,000 buyers split into 5,000 heavy, 15,000 regular and 30,000 light, earning Rs 16,000/-, Rs 4,800/- and Rs 1,600/- each across the year. The published Rs 4,000/- a buyer is none of those three. A customer journey follows one buyer through time instead of averaging every buyer at one moment.

Setu Bazaar, an invented marketplace, has a trading year already set out in full, and the year is quoted rather than rebuilt: Rs 500 crore of gross merchandise valueThe total worth of the goods that changed hands on a marketplace. The marketplace keeps only a slice of that money, never the whole of it. passing across the marketplace, a take rateThe slice of the goods value a marketplace keeps for itself, stated as a percentage. The rest goes to the merchant who sold the item. of 4.00 per cent, Rs 20,00,00,000/- of revenue, 50,000 buyers and 5,00,000 orders. The orders divided by the buyers give ten orders a buyer. The revenue divided by the buyers gives Rs 4,000/-.

Two findings already settled on Setu Bazaar's published year carry everything below. The first is that an average destroys the spread underneath it. The published Rs 4,000/- was left as an average because the spread had never been disclosed. The split below puts the spread on the table. The second is that how often a buyer comes back and how large an order is are counted against different totals, so one can move while the other stands perfectly still. The three groups that follow are that second finding turned into a population of fifty thousand heads.

What is a customer segment, and why does the measure have to be named first?

A customer segment is a group of buyers that behave the same way on one stated measure, and the measure is a choice somebody made rather than a property sitting inside the buyers waiting to be found. Two people who picked different measures can look at the same 50,000 buyers, produce entirely different groups and both be correct. So the first question to ask of any set of segments is never how many there are. The first question is what measure produced them.

A split is only a segment if the groups behave differently on the measure that made them, and any other split is a label. The test takes one line: name the measure, cut the buyers on it, then check whether the resulting groups actually differ. If they do not, nothing has been learned and three new names have been created.

Take a chemist's shop on a busy street. The shopkeeper can sort every customer by which lane they live in, and will end up with three tidy groups that buy the same things at the same rate on the same days. Sorting by lane is a label. Or the shopkeeper can sort them by whether they come back every month for the same repeat prescription, and will end up with a small group that turns up on a schedule and a large group that turns up when something goes wrong. Sorting by the repeat prescription is a segment, and the shop is exactly the same shop under both cuts.

How do Setu Bazaar's 50,000 buyers actually split?

The measure chosen here is how often a buyer orders in the year, and on that measure Setu Bazaar's 50,000 buyers fall into three groups. 5,000 heavy buyers place 40 orders each. 15,000 regular buyers place 12 each. 30,000 light buyers place 4 each. Heavy, regular and light carry through everything that follows, and the three names carry no compliment and no complaint: heavy means many orders and light means few, and nothing else is being said about anybody.

The addition is the only reason to accept the three groups. 5,000 and 15,000 and 30,000 is 50,000 buyers, the published head count. Each group multiplied by its frequency gives 2,00,000 orders from the heavy buyers, 1,80,000 from the regular buyers and 1,20,000 from the light buyers. The three add to 5,00,000 orders, the published order count. The split is not an extra assumption laid on top of the published year, it is the same year with the buyers sorted.

The three counts are worth being plain about. Setu Bazaar's own disclosure never showed a spread, so these three group sizes and three frequencies illustrate a distribution that was never published. The six figures were not observed. The six figures were chosen, and chosen under a hard constraint: they had to rebuild the head count, the order count and the revenue exactly, with nothing rounded and nothing left over. The constraint is what makes an illustration worth reading instead of merely worth looking at.

Why does only the frequency differ between the three segments?

Look at what is held perfectly still. Every order placed in this marketplace carries the same Rs 10,000/- of goods, whoever placed it, and Setu Bazaar keeps the same 4.00 per cent of it. The 4.00 per cent of Rs 10,000/- is Rs 400/- of revenue on every single order in all three groups. A heavy buyer's order and a light buyer's order are the same order. Nothing about the basket changes across the split, and nothing about the slice changes either.

Holding both still leaves exactly one thing free to move, and it is the count of returns.

So each segment's year is a single multiplication. A heavy buyer places 40 orders at Rs 400/- of revenue each, for Rs 16,000/-. A regular buyer places 12 at Rs 400/-, for Rs 4,800/-. A light buyer places 4 at Rs 400/-, for Rs 1,600/-. Not one rupee of order size separates a heavy buyer from a light one, and the entire distance between Rs 16,000/- and Rs 1,600/- is thirty-six returns. This is the reason a per buyer figure and a per order figure sit on different denominatorsThe number sitting underneath a division. Change the number being divided by and the answer changes, even when the amount on top has not moved at all. and can drift apart without either being wrong. The drift is set out in its own place.

Why does the published Rs 4,000/- a buyer describe nobody?

The figure is not wrong, and dismissing it teaches the wrong lesson, so it deserves fairness first. Rs 20,00,00,000/- of revenue divided across 50,000 buyers is Rs 4,000/-, and 50,000 buyers multiplied by Rs 4,000/- comes straight back to Rs 20,00,00,000/-. Nothing was fudged. Rs 4,000/- is the one per buyer figure that scales cleanly back to the whole year, and that is a genuine use rather than a consolation prize.

Now set the three real figures beside it. Rs 16,000/- is four times the average. Rs 1,600/- is less than half of it. Rs 4,800/- is the nearest of the three and still sits Rs 800/- away, a fifth of the figure it is supposed to be near. No buyer in this business earns Rs 4,000/- and not one of the three is even close, so Rs 4,000/- is an arithmetically correct description of nobody.

Two things explain why the average lands where nobody stands. The Rs 4,000/- is a weighted averageAn average where each figure counts in proportion to how many items sit behind it, so a group of thirty thousand pulls thirty times harder than a group of one thousand. rather than the middle of the three segment figures. The 30,000 light buyers drag it far down towards Rs 1,600/-, and the 5,000 heavy buyers pull it back up with a much smaller hand. The Rs 4,000/- is also not the medianThe figure sitting exactly in the middle of a list once every item has been lined up in order, so half the list is below it and half above., and a median would have landed on a buyer who does exist. An average is under no obligation to describe anybody, and Setu Bazaar's does not.

Consider a street of ten households. One eats out four times a week and nine eat out once a month. The average number of restaurant meals per household on that street is a figure that is perfectly correct, easy to check and true of not one dinner table on the street. Nobody on that street is average. The street still has an average.

How much of the year do ten per cent of the buyers carry?

Multiply each group out and the year reassembles. The 5,000 heavy buyers at Rs 16,000/- each produce Rs 8,00,00,000/-. The 15,000 regular buyers at Rs 4,800/- each produce Rs 7,20,00,000/-. The 30,000 light buyers at Rs 1,600/- each produce Rs 4,80,00,000/-. The three segment revenues add to Rs 20,00,00,000/-, the published revenue with nothing rounded and nothing left over. The revenue total is the third and last pin holding the split in place.

Now take the two figures that matter and put them next to each other. 5,000 buyers out of 50,000 is 10.00 per cent of the heads. Rs 8,00,00,000/- out of Rs 20,00,00,000/- is 40.00 per cent of the revenue. Ten per cent of the heads carry forty per cent of the year. Those two percentages are worked out from this marketplace's own figures and from nothing else. Another business, with its own head counts and its own frequencies, would produce a different pair, and no rule of thumb says what the pair ought to be.

The bottom row is the whole argument in one line, and it repays careful reading. Every figure in it is published. 5,00,000 orders across 50,000 buyers is ten orders a buyer, and Rs 20,00,00,000/- across 50,000 buyers is Rs 4,000/-. Every figure above it is illustration. And the three real frequencies are 40, 12 and 4, so the column of ten orders a buyer describes exactly as many people as the Rs 4,000/-: none.

What is a Customer Journey, and how is it different from a segment?

A customer journey is the ordered set of stages one buyer passes through, from the moment the business first spends anything to reach that buyer to whatever the buyer is doing when the period closes. A journey is not a summary. A journey is a sequence, and its whole content is the order things happened in.

A segment is many buyers at one moment and a journey is one buyer across time, and the two are the same data read along different axes. A cut down through the population at one instant gives segments. A run along one row through the whole period gives a journey. Neither reading is more true than the other and neither can be recovered from the other.

The segments come before the journeys for a reason. A journey drawn for the average buyer would be a journey nobody takes, and the claim about the average has arrived again in a different shape. If Rs 4,000/- describes no buyer, then ten orders spread neatly across the year describes no buyer's diary either. So the journeys below are drawn for one buyer out of each of the three segments, using the same population that was just sorted, with no second cast and no second set of figures.

A bus route carries the same distinction. The timetable states what the whole service does in a day, how many buses run and how full they are. Riding one bus from the first stop to the last shows what one passenger actually experiences, in order, including the twenty minutes stuck near the market. Neither replaces the other, and a passenger who has only read the timetable has not been on the bus.

What does one published year look like for one buyer from each segment?

All three journeys start in the same place. The start is where invention is most tempting, so exactness matters most there. Setu Bazaar's cost of winning a buyer is published as one single figure covering every buyer it won. Nothing published splits that figure by segment. So all three of the buyers below are treated as having arrived the same way, on the same terms, at the same cost.

After that identical arrival, the year runs. Every figure in it is a frequency from the split divided by four, and nothing new enters. The heavy buyer places ten orders a quarter, forty across the year, carrying Rs 4,00,000/- of goods and leaving Rs 16,000/- with Setu Bazaar. The regular buyer places three orders a quarter, twelve across the year, Rs 1,20,000/- of goods and Rs 4,800/- kept. The light buyer places one order a quarter, four across the year, Rs 40,000/- of goods and Rs 1,600/- kept.

The three journeys are the same shape at three speeds, and not one of them is the average journey. The buyer implied by the published figures places ten orders in the year, between the regular buyer's twelve and the light buyer's four, and matches neither. Whether any of these three is still ordering when the next year opens is a genuinely different question, one that turns on what keeps a buyer, and it is covered under Customer Loyalty.

Why do three buyers who arrived the same way end up ten times apart?

Put the two ends of the range together. Rs 16,000/- against Rs 1,600/- is exactly ten to one. Forty orders against four orders is exactly ten to one. The two ratios are the same ratio because they are the same fact: the order never changed size and the take never changed, so nothing was left that could produce a different multiple. The whole of the difference is the count of returns and none of it is the size of an order.

Now place the average against the heads rather than against the figures. The heads are where the damage happens. 30,000 of the 50,000 buyers sit below Rs 4,000/-, at Rs 1,600/- each. 15,000 sit just above it at Rs 4,800/-, missing it by Rs 800/-. 5,000 sit far above it at Rs 16,000/-. So three fifths of the buyers are below the average and two fifths are above it, and the two fifths above it are split between a group that barely clears it and a group four times past it. There is no percentileA position in a list that has been sorted and then cut into a hundred equal parts, used to say how far along the list something sits. of this population at which a buyer earning Rs 4,000/- can be found.

What does a segment split leave unanswered?

A segment split does not say why a heavy buyer is heavy. Nor does it say whether any of the three buyers will order again. The published arrival figure has never been split three ways, so a split does not say what any segment cost to win either. And a split does not say which buyers a business should chase. Choosing is not a shortage of information but a different activity altogether.

A segment split describes a year that has already closed and forecasts nothing. Deciding what to do about a group of buyers is a separate job from measuring one, done by different people with different evidence. A report that starts recommending which buyers to pursue has stopped describing the business and started running it, and its reader can no longer tell which of the two is being read.

When a per buyer figure moves, what is the third question to ask?

Two questions about any per buyer figure are already settled elsewhere and take a clause each. The first is what sits in the denominator, meaning which heads were counted and which were left out. The second is what the spread underneath the average looks like, and the spread is what everything above has been supplying. The third question is the one a segment split adds, and it is the one a reader can use tomorrow: when a per buyer figure moves, did the size of a purchase move or did the number of purchases move?

The two look identical in the figure and are not the same event. On Setu Bazaar's published year, revenue per buyer is Rs 4,000/-, built from Rs 400/- of revenue on an order and ten orders a buyer, the ten being 5,00,000 orders across 50,000 buyers. Lifting the order to Rs 500/- of revenue while the count stands still gives Rs 5,000/-. Lifting the count to twelve and a half orders while the order stands still also gives Rs 5,000/-. The figure cannot say which happened. Which of the two per buyer measures to reach for in the first place is covered separately.

A lender reading a marketplace's file, an analyst building a model and a household running a small shop all reach for the same move here, and it costs nothing. Before accepting any per buyer figure, the two components underneath it are worth asking for: how much is kept on one purchase, and how many purchases were made. Where a business supplies both, the average becomes a check on the reader's own arithmetic rather than the whole of the evidence. Where it supplies only the average, what is missing is known exactly. The shape of the population is missing, and that can be said out loud instead of quietly assumed to be flat.