Two products with the same margin percentage can deserve opposite budget decisions this month.
That sentence sounds wrong to anyone who built their ad account around margin tiers. I understand why. Margin segmentation was a real improvement over blended ROAS targets, and if you run an ecommerce P&L, you did the right thing by adopting it.
The improvement stops short, though. A margin percentage describes efficiency at a moment in time. The budget decision depends on something else: the commercial job that product needs to carry this period, and the evidence that would change that job.
I want to give you the full decision rule, because the version worth taking to your buying and finance teams has two parts. The first part assigns each SKU a job. The second part names, in advance, what would retire that job and trigger the next one. The second part is the one that gets skipped, and it is where the money sits.
Why Margin Alone Cannot Make the Decision
Margin tells you how much room a product has per unit sold. It says nothing about three things that decide where your next pound of ad spend belongs:
Stock position. A 60% margin product with 40 weeks of cover sitting in a warehouse is a cash problem wearing a profit costume. A 60% margin product you can reorder in two weeks is a growth engine.
Customer economics. Some products bring in first-time buyers who come back and build basket value over time. Others sell almost entirely to people who already know you. Same margin, completely different value per sale.
Lifecycle and season. A product entering its peak selling window and a product six weeks from going out of season sit at different points on the clock, even when their unit economics match to the decimal.
Here is an illustrative example, clearly labelled as such. Imagine two SKUs, both at 55% margin. Product A is a replenishable core line that acquires new customers who reorder within 90 days. Product B is a seasonal line you over-ordered, with stock ageing toward the end of its window.
Product A deserves more budget this month, pushed toward new-customer traffic, judged on acquisition cost. Product B deserves budget too, pushed hard toward anyone who will buy, judged on cash recovered against the cost of holding that stock. Same margin. Opposite decisions. Both correct.
The Mechanism: Every SKU Has a Job®
At JudeLuxe we work from a simple premise: every SKU has a job®. The job sits at the intersection of four inputs:
- Margin, as the efficiency ceiling
- Stock position, as the constraint or the pressure
- Customer-acquisition role, as the strategic value of each sale
- Lifecycle stage, as the clock
Once you read those four together, a small set of distinct jobs appears. A product can be tasked with growing volume, with acquiring new customers, with protecting profitable demand, or with turning stock back into cash.
Each job carries its own success measure. This is the part that is commonly overlooked. A profit-role SKU judged on revenue leaves money on the table, because the spend level that maximises revenue sits above the spend level that maximises contribution. Judging a clearance SKU on ROAS makes the same mistake in the other direction: the ROAS looks terrible while the decision is correct, because the alternative is writing the stock off entirely.
Automated bidding makes this worse when left unsupervised. The algorithm is good at finding sales. It has no knowledge of which sales are affordable, which stock is overstocked, or which customers return. You supply that intent, or nobody does.
The Decision Table
Here is the table I would put in front of a marketing, buying, and finance meeting. It is illustrative in its thresholds; the structure is the point.
| Job | Assign when | Judged on | Evidence that changes the job |
|---|---|---|---|
| Grow volume | Replenishable line, healthy margin, short reorder lead time | Margin floor | Contribution per unit drops below the floor, or cover falls under lead time |
| Acquire customers | High share of first-time buyers who reorder | Allowable acquisition cost | New cohorts repeat below the rate the cost was set on |
| Protect profit | Steady demand, mostly existing customers | Contribution per order | Selling window closing while stock cover stays high |
| Recover cash | Overstocked, ageing, near end of season | Cash recovered against holding cost | Stock cleared, or cover back to normal |
Read the fourth column carefully. That column is the decision rule other writing on margin segmentation leaves out. Every job assignment comes with a named trigger. You decide, before the month starts, what evidence would change the job.
Matching the Metric to the Job
Once the job is assigned, measurement follows it. Specific numbers for specific jobs:
Growth jobs get judged against a margin floor. You accept thinner contribution per unit in exchange for volume, and the floor defines how thin is acceptable. Below the floor, growth is a vanity exercise.
Acquisition jobs get judged on allowable acquisition cost. Calculate what a new customer is worth over a defined period, decide what share of that you will spend to get them, and hold the line. First-order ROAS on these products will look poor by design, and that is fine, because ROAS was never the target.
Protection jobs get judged on contribution per order: revenue minus product cost, delivery, and ad cost. When you report contribution, state what you included, because a contribution number with hidden exclusions misleads everyone downstream.
Cash-recovery jobs get judged on cash recovered against holding cost. Sometimes the correct call is to cut losses and free the cash tied up in stock, even when the P&L takes a visible hit. Profit and cash cannot always both be won in the same period, and pretending otherwise just delays the decision.
One caution worth stating directly: a product only becomes a loss leader when you explicitly assign it that role. Spend drifting toward whatever converts, under a blended target, assigns loss-leader status by accident. The business pays for that accident quietly, every month.
The Job Is Expected to Change
The clearest example of a job change is seasonal. A product spends four months in a protect-profit role, running at steady contribution per order. Then its selling window starts closing, stock cover remains high, and the correct job becomes cash recovery. Nothing about the product changed. The calendar and the warehouse changed, and the job followed.
Reassignment is a human decision made deliberately. The algorithm will keep optimising the old objective long after the commercial reality has moved on, because it was told to. You review the triggers, you check the evidence, and you move the job when the evidence says so.
Cadence matters here. A monthly review makes waste visible too late. A product can burn three weeks of budget in the wrong job before anyone reads the report. A frequent review beat, built around the trigger column of the table above, catches the change while it is still cheap to act on.
Stating the whole thing in business language is what makes it work across teams. Finance can sign off an allowable acquisition cost. Buying can confirm stock cover and reorder lead times. Marketing can set the campaign structure to match. Everyone acts on the same job definition, because the job is written in margin, stock, cash, and customers, in terms every function already uses.
Where This Framework Has Limits
I want to be honest about the boundaries, because a decision rule you trust too far becomes its own problem.
First, the framework assumes your measurement is sound. Reported ROAS alone does not establish profitability, and attributed revenue differs from incremental value. Check measurement quality before you move budgets based on it, because a precise-looking number built on a broken attribution model produces confident, wrong decisions.
Second, job assignment at full SKU granularity is unrealistic for a catalogue of thousands of lines. In practice you assign jobs at the level of product groups that share margin, stock behaviour, and customer role, and you break out individual SKUs only where the money justifies the attention.
Third, downstream value estimates for acquisition products carry uncertainty. Repeat rates shift, and an allowable acquisition cost set on last year's cohort behaviour needs rechecking as new cohorts mature. Treat those numbers as working hypotheses with review dates attached, as distinct from settled facts.
The examples in this article are illustrative. They show the structure of the decision, and your own thresholds come from your own P&L.
The Question to Take Into Your Next Planning Meeting
Pick your five biggest ad-spend products. For each one, write down the job it is doing this month in one sentence, the single metric that job is judged on, and the specific evidence that would change the job.
If any of those three answers is missing, that product's budget is running on autopilot, and the algorithm is setting your commercial strategy by default.
The margin percentage got you to the table. The job, and the evidence that would change it, is the decision. Which of your products is carrying the wrong job right now, and what would prove it to your finance team this week?