A kitchen equipment dealer sells machines and the replacement parts that keep those machines running. The machines fill the showroom. They dominate every sales conversation and most of the revenue line. The parts sit in a back room and get ordered by phone, usually by someone who is mildly annoyed that they need them. Ask the owner which side of the business pays the rent and he will point at the showroom.
He may be wrong. McKinsey has reported that industrial spare parts and consumables typically account for about 20% to 40% of revenue while contributing a much larger share of profitability. The small, boring, low-visibility stream is often the engine.
This is why the effort to find the most profitable part of your business usually breaks down before any arithmetic happens. Owners reach for the sales report because it is the report that exists. Sales reports rank by volume and visibility, which are exactly the two things that have no reliable relationship to profit.
Decide what you are actually comparing
“Most profitable part” is not one question. It is at least four, and they produce different answers.
- Products or services. Which offer earns the most per sale after the costs that sale causes?
- Customers or segments. Two customers can buy the same thing at the same price and leave you with very different amounts of money, because one of them calls four times a week.
- Channels. The same product sold through a marketplace, your own site, a reseller, or in person carries different fees, different discount pressure, and different acquisition cost.
- Streams attached to the core sale. Parts, consumables, servicing, installation, training, renewals, delivery.
Pick one to start. If you sell a handful of things to many customers, start with products. If you sell one thing to a wide range of buyers, start with customers. If you sell the same catalogue through several routes to market, start with channels. Trying to answer all four in one spreadsheet is how this exercise dies in week two.
A useful prompt: which of these four could you plausibly change within a quarter? You can change your pricing and your marketing focus quickly. You can change which customers you pursue fairly quickly. Rebuilding a channel takes longer. Analyse what you can act on.
Step one: rank by contribution dollars, not revenue and not margin percentage
The first honest cut is contribution. The U.S. Small Business Administration defines contribution margin as the difference between the sale price and what it costs to make the product, expressed as a ratio: (sale price per unit minus variable cost per unit) divided by sale price per unit. The same framework gives the break-even point in units as fixed costs divided by sale price per unit minus variable cost per unit, and the break-even point in sales dollars as fixed costs divided by contribution margin.
That is the whole idea in one line. Every sale contributes a certain number of dollars toward your fixed costs. Once those fixed costs are covered, the same contribution becomes profit. So the question is not which item has the prettiest margin percentage. It is which item throws off the most dollars toward keeping the lights on.
This matters because the two rankings diverge constantly. A service with a 70% margin that sells twice a month contributes less than a 35% margin product that moves every day. Margin percentage tells you how good the deal is. Contribution dollars tell you whether you survive the month. When people ask me which number to lead with, I say contribution dollars, every time, and keep margin percentage as a second column for pricing decisions.
The variable costs almost everyone leaves out
Here is where older advice fails. Most guidance still says selling price minus cost of goods sold, which was reasonable when a sale was a sale. It no longer is.
Payment processing is a real, per-transaction, variable cost and it is not small. Stripe’s standard online card pricing is 2.9% plus 30 cents per successful card charge. PayPal’s U.S. merchant fee schedule, updated in September 2026, lists PayPal Checkout domestic commercial transactions at 3.49% plus a fixed fee of 49 cents, standard domestic card payments at 2.99% plus 49 cents, and Pay Later options at 4.99% plus a fixed fee. Those are not the same number. If a meaningful share of your customers choose Pay Later, your payment mix is quietly reshaping your channel profitability while your P&L shows one blended line called “bank charges.”
Notice what the fixed portion does to small orders. Thirty or forty-nine cents on a $200 order is noise. On a $12 order it is a visible slice of the contribution. Low-ticket items get punished by fee structures in a way that gross margin analysis never reveals.
Returns are the second omission. Shopify’s return calculation tooling models refund amounts, taxes, restocking fees, return shipping fees, and discount adjustments. That list is a good checklist even if you have never touched an API. A product with a 30% return rate and outbound plus inbound shipping is not the product your gross margin report thinks it is. Dispute costs belong in the model too, though with one nuance in your favour: Stripe returns the dispute fee if you win the dispute.
So the variable cost line for a product should include the product cost, inbound freight, payment fees at the actual rate for that channel, packaging, outbound shipping you absorb, expected returns and their shipping and restocking cost, and any discount you routinely give. For a service, it includes the delivery labour at real loaded cost, subcontractors, software licences bought per client, and travel.
Do this for your top ten or twenty offers and something usually falls over. The bestseller slides down the list. Something quiet climbs.
Step two: test the ranking against costs the offer actually causes
Contribution is the right starting point and the wrong stopping point. It ignores every cost that is fixed in the short run but caused entirely by one part of the business.
McKinsey’s work on portfolio complexity makes this vivid. Traditional accounting systems, the firm notes, often fail to capture and correctly allocate the actual costs of product complexity. In one portfolio example, a small engine was believed to cost about 10% less to manufacture than its larger sibling. Fair cost allocation showed it actually cost 20% more. The reported number and the real number pointed in opposite directions.
You do not need an automotive portfolio to reproduce that mistake. A custom size that requires its own tooling, a certification that only one product line needs, a client who consumes a third of your project manager’s week, a location whose lease exists for a single service line: all of these are fixed costs with a name attached.
The disappearance test
Rather than allocating overhead by a formula, ask a causal question about each offer, customer, or channel:
If this disappeared tomorrow, which costs would actually leave the business within ninety days?
Those costs belong to it. The rest do not. Your accountant’s rent is not caused by your third product line, and spreading rent across products by share of revenue produces confident numbers that mean nothing. Worse, it flatters high-revenue lines and penalises small high-margin ones, which is precisely backwards.
This is where I part company with the standard “allocate all overhead” advice. Allocate by resource consumption where you can trace a real link. Leave genuinely shared costs unallocated and cover them from total contribution. A fully loaded profit number built on arbitrary percentages is not more accurate than contribution. It is just more decorated.
Watch for complexity that earns nothing
Some complexity pays. A wide range that pulls in customers who then buy the profitable core is doing a job. Other complexity is pure drag: extra SKUs that slow production changeovers, split your stock, confuse your own sales team, and generate support tickets.
The revealing pattern is concentration. In one McKinsey manufacturing example, 30% of the portfolio generated more than 80% of cumulative gross margin. Build that curve for your own business. Sort every offer by contribution dollars, plot the cumulative total, and look at where the line flattens. Everything past the flattening point is under review. Not condemned, under review. Related symptoms often show up on the operations side too, which is worth reading alongside the warning signs that a business strategy has become too complicated.
Profit per unit of the thing you run out of
This is the step that almost no article on this topic covers, and it is the one that changes decisions.
Your business does not run on revenue. It runs on whatever you run out of first. A design studio runs out of senior designer hours. A bakery runs out of oven time between 4am and 7am. A shop runs out of shelf facings. A service business with a strong pipeline runs out of the founder’s attention. An ecommerce brand with plenty of stock runs out of advertising budget before it runs out of anything else.
Once you know your binding constraint, the correct ranking is not contribution per sale. It is contribution per unit of that constraint.
Suppose a studio offers a brand identity project contributing $6,000 and a monthly retainer contributing $3,000. The project looks twice as good. Then you count hours. The project consumes 90 senior hours. The retainer consumes 25. That is roughly $67 per senior hour against $120. The retainer is the better use of the constrained resource by a wide margin, and the ranking flips.
Run the same test with whatever you actually run short of: contribution per machine hour, per square metre of shelf, per delivery slot, per dollar of ad spend, per founder hour. If you have never worked out which resource is binding, that diagnosis comes first, and it deserves its own session. It is worth reading how to identify the real constraint limiting your growth before you rebuild your pricing around the wrong one.
A second consequence: constraints move. Solve the oven bottleneck and the constraint becomes delivery vans. When the constraint moves, the profitability ranking moves with it. This is a rolling calculation, not a one-time verdict.
What to do once you know
Analysis without a decision is just an expensive way to feel informed. Each part of the business should land in one of five buckets.
Scale it
High contribution per constrained unit, demand available, no hidden support burden. Push marketing, sales attention, and capacity here. One caution before you spend: confirm the demand actually exists where you plan to sell. Google’s own documentation is explicit that average monthly search volumes depend on location, language, date range, and Search Network settings, and that keyword demand should be checked with those filters applied. A profitable line in one market can have almost no search demand in the next country over, or be searched for using completely different words. The SBA’s planning guidance makes the same point in broader terms, asking for demand, market size, customer location, market saturation, and pricing before you commit.
Reprice it
Strong demand, thin contribution. This is the most under-used response. A 5% price increase on a line with a 30% contribution margin does far more for profit than a 5% volume increase, and it costs nothing to implement. Test it on new customers first.
Fix its cost structure
Good demand, decent price, leaking money. Look at the variable costs you just discovered. Change the payment mix you promote. Reduce the return rate with better sizing information or clearer photographs. Renegotiate freight. Stop absorbing shipping on low-ticket orders.
Bundle or attach it
Weak on its own but it pulls profitable work behind it. The equipment dealer’s machines may be exactly this. A low-contribution product that reliably brings in the parts business is not a problem to be solved. It is an acquisition channel with a misleading name.
Cut it
Weak contribution, no attachment effect, real complexity cost. McKinsey reports that some companies achieve SKU reductions of 30% to 50% while stabilising production at lower cost and minimising stock-outs, and that a portfolio-healing process can reduce costs by up to 7%. That is a real prize. But cut by testing, not by reflex. Pull the item from promotion for a quarter and watch whether the customers who bought it disappear entirely or simply buy something else. Cutting is a skill of its own, and it fails when it is done emotionally rather than as a deliberate discipline.
Cadence, tools, and one habit worth changing
Quarterly is the right rhythm for a full pass. Monthly is enough for a quick contribution check on your top ten offers. Anything more frequent and you are reacting to noise.
You do not need enterprise software. Accounting tools have moved down-market on this: QuickBooks Enterprise markets a Profitability by Product report and invoice-level profitability reporting, and the same logic can be rebuilt in a spreadsheet in an afternoon with an export of your sales lines, your cost prices, and your processor statement. The constraint is rarely tooling. It is that nobody has decided to look.
One habit is worth changing permanently. If your team reviews a sales dashboard every week, they are being trained to chase revenue. Put contribution dollars next to the revenue column, for every product and every salesperson. Behaviour follows whatever is on the screen. The same applies to lead generation: once you know which customers carry the contribution, you can define what a good lead looks like instead of guessing, which is a different exercise from simply sorting good leads from bad ones by whether they reply quickly.
Finally, ignore published margin benchmarks. There is no shared standard for a “good” margin across industries, channels, and capital structures, and the numbers that circulate online are usually quoted without the context that made them meaningful. The only benchmark that matters is the internal one: which part of your own business produces the most contribution per unit of the resource you run short of. That is a question you can answer this week, and the answer is almost never the part you talk about most. Once you have it, turning it into actual change is a separate job, and it benefits from being handled as a structured ninety-day plan rather than a resolution.
