Which Customers Actually Make You Money?

There is one number you should be able to say out loud right now, and most operating executives cannot: the share of your gross profit that comes from your ten largest customers. Not revenue. Profit.

If you can only produce the revenue version, that is not a reporting gap. That is the finding. It means every decision you have made about pricing, capacity, service levels and sales coverage has been made on a ranking that describes how busy each customer keeps you, not how much each one pays you for the privilege.

This article is the complete method for fixing that. It is long because the subject has four parts that people usually attempt separately and in the wrong order, and because the objections that kill these projects are predictable enough to answer in advance. I have run this analysis in businesses from a few million in revenue to divisions north of a billion, and the shape of the answer has never once been what the leadership team expected.

What is customer profitability analysis, actually?

It is the exercise of assigning your real cost to serve down to the individual customer and product family, so you can see profit rather than revenue by relationship. The output is not a report. It is a list of specific decisions about price, terms, product and channel that were previously invisible.

The distinguishing feature is the cost side. Revenue by customer is a query anyone can run. Profit by customer requires you to decide how the costs of serving that customer get attached to them, and that decision is where the entire exercise is won or lost. Every business already has a method for this. It is usually an allocation on revenue or direct labor, it was chosen for financial reporting rather than for management, and it is systematically wrong in the same direction.

It is wrong because it spreads cost in proportion to size. A customer who places forty small orders a year with three engineering change requests and a weekly expedite gets charged overhead in proportion to their revenue, which is small. The actual consumption is enormous. Meanwhile your largest, simplest account subsidizes them and shows up on the report looking merely adequate.

Three beliefs keep these customers on the books. See 3 customer profitability myths costing you millions.

To grow the customers worth keeping, see your CS team is a retention team and the NRR fix.

Once you know which customers make money, measure how much of their spend you capture with a share of wallet audit.

Watch the companion video on 80/20 Customers: Who Are Your Top 5%?.

Watch on YouTube

The P&L is built to hide this

Nothing in your financial statements is designed to answer this question, and that is not an oversight. External reporting exists to produce a comparable, auditable company-level result. It has no obligation to tell you which relationship is carrying which, and it does not.

Walk down the income statement and watch where the information goes. Revenue is aggregated. Cost of goods sold is aggregated, and to the extent it is broken out, it is broken out by product or plant rather than by customer. Everything below the gross margin line, which is where a very large share of cost-to-serve lives, is organized by function: sales, engineering, customer service, logistics, finance. Functions are not customers. By the time the statement is assembled, the connection between a specific account and the cost it generated has been destroyed three separate times.

This is why the exercise always feels like it is fighting the systems. It is. Your ERP, your chart of accounts and your management reporting were all built to produce the company view. Producing the customer view means reassembling the pieces in an order nobody designed for, which is unglamorous and is the reason most organizations talk about doing this for years before anyone does it.

What are the four populations you will find?

Four, and they need completely different responses. Large customers on core products, large customers on tail products, small customers on core products, and small customers on tail products. Ranking customers on one axis collapses these four into a line and destroys the information you are trying to surface.

The method I use for this is the 80/20 Matrix of Profitability: customers on one axis, product families on the other, gross profit after allocated cost-to-serve in the cells. The reason it has to be two axes is that the interesting behavior is never at the customer level or the product level alone. It is at the intersection, where a perfectly good customer buys a product that costs you three times what you charge for it.

Large customers, core products. This is the business. It is usually thinner on margin percentage than anyone expects, because these accounts negotiate, and much fatter on margin dollars than anyone expects, because volume through a standard process is cheap.

Large customers, tail products. The trap. Volume makes the complexity feel earned. Somebody agreed to a custom variant for an important account in 2019 and nobody has priced it since.

Small customers, core products. Frequently your best margin percentage in the building. Standard product, standard terms, no leverage, no engineering support. Crude rationalization kills these first.

Small customers, tail products. Tiny revenue, disproportionate cost, and the population everyone means when they say the word tail.

The four populations produced by a customer and product profitability matrix Customers by product families Customer size Product family Core products Tail products Large Small THE BUSINESS Thin percentage, fat dollars. Protect the process, not the price. Action: defend and grow. THE TRAP Volume makes complexity feel justified. It is not. Action: standardize or reprice. THE QUIET WIN Best margin percentage you have. Crude cuts kill these first. Action: find more like them. THE TAIL Small revenue, large cost, almost never allocated. Action: reprice, re-channel, exit.

Isn’t this just Pareto with extra steps?

No, and the difference is the cost side. Pareto says a minority of inputs drive a majority of outputs, which is an observation. This is an allocation exercise that tells you which specific accounts are subsidizing which other specific accounts, by how much, and through which cost driver. One is a shape. The other is a list with names on it.

The practical gap shows up the first time someone runs the revenue version. Sort customers by revenue, take the top twenty percent, and you have a list that is roughly eighty percent right and wrong in exactly the places that matter. Your most expensive customer to serve is very often near the top of that list, because size buys leverage and leverage buys discounts, extended terms, expedites and free engineering.

There is a second difference that matters more over time. A Pareto observation is static. A profitability matrix is a management instrument you re-run annually, and the interesting information is in the movement: which accounts crossed from profitable to unprofitable, and what changed. That is a question about your own behavior, and it usually has an uncomfortable answer involving a concession nobody documented.

Rough allocation beats precise allocation

The single most common way this project dies is that somebody decides to do it properly. Properly means a full activity-based costing implementation, which means a project charter, a software evaluation, a data cleanup, and eighteen months. Businesses that go that route usually end up with neither the system nor the answer.

Three weeks of approximate allocation will get you within a range that supports every decision you actually need to make. You are not trying to price a government contract. You are trying to find out whether an account is meaningfully profitable, roughly break-even, or clearly destroying value, and the gap between those three states is large enough that a competent estimate resolves it.

Use drivers you can already count. Order lines, shipments, setups, expedite events, engineering hours, returns, days sales outstanding. Assign a rate to each, apply it, and accept that you will be wrong at the margin. Then look at the result and ask a different question than the one people usually ask, which is not “is this number right” but “is any decision I would make different if this number moved twenty percent.” Almost always, no.

Keep the model in a spreadsheet the first year. Software solves a problem you do not have yet, which is repeatability at scale. You have a different problem, which is finding out whether the answer is interesting enough to justify anything.

Which costs actually drive the difference?

Six of them, in roughly descending order of impact: price realization against list, order pattern, customization, expedite frequency, freight and terms, and returns. Almost every account that looks wrong on a profitability matrix is wrong because of one or two of these rather than all six.

Price realization. The gap between your list price and your invoiced price, by account. This is the largest single driver in most industrial businesses and it is almost never looked at as a distribution. Pull it by account and you will find a spread far wider than any policy you have ever written.

Order pattern. Forty orders of one unit and one order of forty units produce identical revenue and wildly different cost. Order lines is the most useful single driver you can add to any model.

Customization. Any deviation from standard, priced once at introduction and never again, while the cost of carrying it compounds annually.

Expedites. Each one consumes a setup, a schedule disruption, premium freight, and somebody’s afternoon. Count them by account and the distribution will be brutally concentrated.

Freight and terms. Absorbed freight and extended payment terms are both price concessions wearing accounting costumes. Convert them to a percentage of revenue and put them in the model.

Returns and rework. Usually small, occasionally decisive, and always concentrated in the accounts you would guess.

What data do you actually need to start?

Five fields and one driver count, all of which already exist somewhere in your systems. Invoice line revenue, invoice line cost of goods, customer identifier, product family, and invoice date, plus a count of order lines by customer. Everything else is refinement.

People stall here for months and the reason is almost always the same: somebody looked at the customer master, discovered it is a mess, and concluded that the analysis cannot proceed until it is cleaned. It can. Your customer master probably has the same company under three spellings, a handful of ship-to addresses masquerading as separate accounts, and at least one entry that has not bought anything since 2019. None of that changes the shape of the answer.

Deal with it by working top-down rather than cleaning bottom-up. Take the accounts representing the first eighty percent of revenue, which is usually a manageable list, and clean only those by hand. Roll everything below that threshold into a single line called tail and treat it as one population for the first pass. You have just removed the largest obstacle to starting, and the tail behaves as one population anyway.

The one field people genuinely do not have is cost of goods at invoice line level, because many systems carry a standard cost that was set at some point and has drifted. Use it anyway for the first pass, and note which product families have the largest gap between standard and actual. That gap is itself a finding, and it belongs in a separate conversation with whoever owns the standards.

What changes in year two?

The exercise stops being an analysis and becomes a control, and the interesting information moves from the levels to the movement. In year one you are finding out who subsidizes whom. In year two you are finding out what you did about it, which is a harder question and a more useful one.

Run the same model on the new year and put the two side by side. Three populations matter. Accounts that crossed from negative to positive, which tell you your interventions worked and are worth understanding in detail. Accounts that crossed the other way, which are the early warning system, because something changed and nobody noticed. And accounts that never moved despite being on the action list, which tell you something about your own follow-through.

That third group is the one worth sitting with, because it is usually larger than anyone expects. A decision was made in a meeting, it was recorded, everyone agreed, and twelve months later the account is priced exactly as it was. Nobody defied anything. The conversation simply never happened, because it was nobody’s specific job on any specific day.

This is why the annual cadence matters more than the precision of the model. A rough analysis run every year with a follow-up review will beat an excellent analysis run once, by a wide margin, because the failure mode in this work has never been analytical. It is that the list of twenty decisions produced in March is a list of nine decisions by December and nobody has counted.

What the finished distribution looks like

Steeper than anyone in the room expects, and with a negative tail nobody has ever seen before. The standard shape is a curve that rises to a peak above one hundred percent of total profit and then declines, because the loss-making accounts at the bottom are subtracting from what the profitable ones generated.

That shape is the reason this exercise changes behavior when a hundred slide decks did not. A leadership team can argue about whether an account is strategic. It is much harder to argue with a chart showing that the company generated more profit than it reported, and then gave a portion of it back at the bottom of its own customer list.

Two things to prepare for when you present it. The first is that the peak is usually reached faster than anyone predicted, with fewer accounts than anyone predicted. The second is that at least one name in the negative tail will be a customer the organization is proud of, and the room will spend forty minutes on that single name rather than on the other two hundred. Decide in advance how you will handle that, because it will happen.

What do you do with the profitable core?

Protect the process, not the price, and resist the instinct to go extract more margin from them. The accounts carrying your business are usually carrying it because they are cheap to serve, and the fastest way to ruin that is to start treating them as a source of price increases.

The real work here is defensive and it is about concentration. Name the accounts that produce the majority of your gross profit. Count them. If the number is small enough to be alarming, that is a strategic finding worth more than the entire rest of the exercise, because it tells you your risk is not in your tail at all.

Then map what makes them cheap: standard products, predictable order patterns, stable specifications, clean handoffs. Those are the characteristics you want to reproduce, and they should become the profile your commercial organization hunts for. Most sales teams are targeting on size. Size is the wrong screen, and this analysis gives you a better one.

The small profitable accounts are the quiet win

This quadrant is where I have found the most unclaimed value, and it is the one that gets destroyed by rationalization programs run on revenue rankings. Small accounts buying standard products at list, with no engineering support and no expedites, are frequently the highest-margin-percentage relationships in the company.

They are invisible for a structural reason: nobody is assigned to them. They are below the threshold for direct coverage, they never escalate, and they generate no meetings. In a revenue-ranked world, silence and smallness look identical to irrelevance.

The move is not to add coverage, which would destroy the economics that make them attractive. It is to find more of them and serve them through a channel built for the purpose: distribution, a configured web order path, a standard catalog with published pricing. This is also where you send accounts from the tail quadrant that will not accept a repriced custom variant but would happily buy the standard one.

What about large customers who lose money?

These are fixable and they are where the money is, because the loss is nearly always caused by a handful of documented concessions rather than by anything structural. Find the concession, price it, and a large unprofitable account becomes a large profitable one without any change in volume.

Reconstruct the history before you open a conversation. What did this account look like four years ago, what changed, and when. You are looking for the specific event: a discount granted to hit a quarter, a freight term absorbed during a supply disruption, a custom item added during a launch, an index clause that stopped being invoked. These things do not reverse themselves and nobody ever put an expiry date on them.

Then build the conversation around the driver rather than the account. “We need three points from you” is a negotiation you will lose. “This variant costs us more than the standard version and here is the difference, which would you prefer” is a product decision, and product decisions get made on facts. If the sequencing question of whether to move on price or cost first is live for you, the four-year diagnostic is here.

What about small customers who lose money?

Four outcomes exist and only one of them is exit: standardize, reprice, re-channel, or leave. The popular version of this topic assumes mass firing, which is both the rarest correct answer and the one most likely to cost you something you did not see coming.

Work the outcomes in order. Can the custom item be replaced with the standard one, which frequently it can because the reason for the variant retired years ago. If not, will they pay a price that covers the real cost, which a low-volume account with few alternatives frequently will. If not, does a distributor serve this profile better and more cheaply than you can, which is usually a yes. Only if all three fail does exit come into play.

When exit is genuinely the answer, the execution matters more than the decision, because the relationships attached to a small account are rarely mapped. The full sequence for doing that without damaging the accounts you want is here, and the short version is that you should know who owns the entity before you write to it.

A note on who runs it. The best outcomes I have seen came from pairing one finance analyst with one commercial person for the three weeks, reporting jointly to whoever owns the P&L. Finance alone produces a defensible model nobody acts on. Sales alone produces a model that exonerates every account. The pair produces a model both organizations already believe, which is worth more than any amount of additional precision, because the argument about whether the numbers are real is the argument that consumes the quarter after.

How long does this actually take?

Three weeks to a first usable answer, one quarter to decisions in market, and a year before the results are fully visible in the P&L. Anyone promising faster is selling you a report, and anyone quoting eighteen months is selling you a system.

Week one is data assembly: revenue by customer by product family, and whatever driver counts you can extract without a project. Week two is allocation and modelling, which is one person with a spreadsheet and access to people who know how the business actually runs. Week three is reading the result, testing the accounts that look wrong, and finding the two or three places where the model is misleading you.

The quarter that follows is where it becomes real: conversations, repricing, product substitutions, channel moves, and a small number of exits. Then twelve months before the full effect lands, because notice periods, contract renewals and annual price windows all have their own clocks, and because the freed capacity needs somewhere to go before margin dollars actually rise.

One practical note on the quarter of decisions. Batch them rather than running them one at a time. Twenty account conversations executed over six weeks by a prepared commercial team is a campaign with momentum, and it produces a body of evidence about what customers actually accept. The same twenty conversations spread across nine months are twenty separate negotiations, each one reopening the argument internally, and the last few never happen at all.

What if sales refuses to accept the numbers?

Expect it, and treat the objection as a data request rather than resistance, because in my experience a meaningful share of the specific challenges are correct. Sales knows things about accounts that no allocation model captures, and the fastest way to lose the whole exercise is to defend the model instead of improving it.

The standard objection has three versions. The allocation is unfair, this account is strategic, and this account is about to grow. Version one is often right and you should fix it. Versions two and three are testable: strategic means someone can name the specific benefit and when it arrives, and growth means a forecast somebody will put in writing.

What does not work is running this from finance and delivering conclusions. What does work is running it with a commercial leader in the room from week one, letting them challenge the drivers before the answers exist, and agreeing on the method while it is still abstract. You are not trying to win an argument about a spreadsheet. You are trying to get twenty decisions made, and the person who has to make most of them works in sales.

One more objection worth pre-empting, because it arrives late and it is the most dangerous one: that the whole exercise is an artifact of your allocation choices and a different method would produce a different list. Partly true and mostly irrelevant. Run the model twice with deliberately different driver rates, high and low, and compare the two lists. The accounts that appear in the negative tail under both are the real ones, and there will be plenty of them. That test takes an hour and it ends the argument permanently.

Three things this analysis will not tell you

It is historical. Every number in it describes a period that has ended, and it has no opinion about what a relationship will be worth in three years. An account that lost money last year while qualifying you into a platform that ships in 2028 is not an unprofitable customer, it is an investment, and no allocation model can tell the difference. Somebody has to.

It does not measure strategic value it cannot see: reference accounts that close other deals, customers whose technical demands make your product better, relationships that keep you present in a market you intend to grow into. These are real and they are also the most abused category in the exercise, because every account is strategic to whoever owns it. Require a named, dated benefit and most of the claims dissolve.

And it says nothing about whether the freed capacity gets used. This is the failure mode that turns a technically successful exercise into a worse business: margin percentage improves, margin dollars fall, because contribution was removed and the fixed cost stayed. Decide where the capacity is going before you start removing the things that occupy it.

There is a fourth thing worth naming, which is that the analysis has no opinion about capacity constraints. If a plant is full, the account consuming the most capacity per dollar of profit is costing you every alternative order you could not take, and none of that appears in the model. In a constrained business, rank by profit per unit of the constraint rather than profit per account, and the answer changes considerably. In an unconstrained business, ignore this paragraph entirely.

Where do you start on Monday?

One query and one column. Pull twelve months of gross profit by customer by product family, then add a single driver column: order lines. That combination, which most businesses can produce in a day, will show you roughly eighty percent of what the full exercise will eventually tell you.

Sort it by gross profit descending and add a running total as a percentage of the company’s total. Find the row where the running total crosses one hundred percent. Every account below that row is being paid for by the accounts above it. You have now done the analysis, crudely, and you know whether the real version is worth three weeks.

If you want the historical version of this argument, applied at scale for decades by a company that built an operating system around it, the mechanics of how Illinois Tool Works actually ran it are here. The uncomfortable part of that story is the same as the uncomfortable part of this one. The analysis is the easy half. The structural change that makes the analysis stick is the half that fails.

One limitation worth stating plainly. This method is a cost-to-serve instrument, and it is blind to everything that has not happened yet. It will tell you precisely who paid you last year and it will tell you nothing about who will pay you in 2030. Run it annually, act on it decisively, and keep a separate conversation, with different people in the room, about the accounts you are deliberately choosing to lose money on.

About the Stagnation Assassin

Todd Hagopian is a Fortune 500 transformation executive who has generated $3B+ shareholder value across five turnarounds at Berkshire Hathaway, Illinois Tool Works, and Whirlpool. He is VP of Product Strategy and Marketing. Known as The Stagnation Assassin, he is the author of two published books: The Unfair Advantage: Weaponizing the Hypomanic Toolbox and Stagnation Assassin: The Anti-Consultant Manifesto. His blog is published in 15+ languages and read by operators worldwide. Bring him to your stage via his speaking page or connect with him on LinkedIn. His newest book is 10-Minute Transformation (February 23, 2027).

The analysis is worthless if the decisions do not follow it. A Decision Velocity Audit measures how long it takes your organization to convert a finding like this one into a price in market, and where the delay sits. Most businesses that run the profitability work and see nothing change have a velocity problem, not an analysis problem. Request the audit here.

Want the short version you can use this week? Pick your problem and get the matching free minibook.

Find my free minibook

The full diagnostic and 90-day playbook: 10-Minute Transformation, February 23, 2027.

Preorder 10-Minute Transformation

Not what you needed? Find Your Solution →