Commission based growth marketing: what each way of paying an agency rewards

Every fee pays for something, so ask what

When you pay someone to grow your shop, the way you pay them is a quiet instruction. It tells them what to do on the days you are not looking. An hourly rate pays for time. A retainer pays for staying hired. A percentage of ad spend pays for spending. A commission pays for whatever the commission is measured on, and that last part is where most of the trouble lives.

I have argued before that the usual agency deal pays for the wrong thing. Employers often pay just enough to keep people from jumping ship, and agencies on a fixed fee often do just enough to keep the client from walking away. Commission is my answer to that. But commission is not magic. It is a contract like any other, and a badly measured commission rewards the wrong thing just as reliably as an hourly rate does.

Below I show what each common way of paying actually rewards, what goes wrong with commission and how to fix it, and why the share of risk an agency should carry depends on how well the result can be measured.

The theory behind this is in my article on principal agent theory and agency incentives. The business and every number below are invented so that I can show every step. Nobody real, no client.

Worked example: meet the shop

Amara runs Suirside Home, an online shop in Waterford at suirsidehome.example. She sells cushions, throws, lamps and table linen. Her shop turns over about €60,000 a month. Of every euro of revenue she keeps 40 cent after the products, packaging, delivery and payment fees. That 40% is her contribution margin, and it runs through every example below.

She spends €6,000 a month on Google Ads and Meta ads, and has three quotes, each with a different way of paying.

Four ways to pay for growth work, four different incentives

None of the four common deals is wicked. Each rewards a different behaviour, and that is the behaviour you get more of.

The hourly rate pays for time

An hourly rate is the simplest deal there is. For a bounded task with a clear scope it is honest and fair. I publish hourly rates for ad hoc work myself, and every such task still gets a binding quote with defined deliverables before work starts.

The trouble starts when the job is open ended, like growing a shop. Then the rate pays for input, not output, and the more efficient the person, the less they earn.

An example: Agency one quotes Amara an invented €90 an hour for campaign management. In March they spend 25 hours, so she pays $25 \times €90 = €2{,}250$. In April they find a smarter structure that does the same work in 15 hours. Their invoice drops to $15 \times €90 = €1{,}350$. They did better work and earned €900 less for it.

The retainer pays for staying hired

A retainer is a fixed monthly fee for an agreed scope. It is predictable, and for ongoing work whose hours can be averaged it is often better than an hourly rate.

But a fixed fee is paid in a good month and a bad month alike, so what it rewards is renewal. An excellent month earns the same as an adequate one.

An example: Agency two quotes Amara an invented €2,500 a month. If her sales grow by €10,000 in a month, they earn €2,500. If her sales do not move at all, they earn €2,500. If her sales fall, they still earn €2,500, at least until she notices and gives notice. The only thing that changes their income is whether she stays.

A percentage of ad spend pays for spending

Many paid media agencies charge a share of the ad budget they manage. It is easy to invoice. It also means the agency earns more every time the budget goes up, whether or not the extra spend makes the client any money.

What matters for the next euro of spend is the marginal return, what that one extra euro brings in, not the average return across the whole budget. I worked through the break even line for a fee on spend in the principal agent article, so here is only the short version.

An example: The specialist offers 15% of ad spend, an invented figure. On Amara's €6,000 that is $0.15 \times 6{,}000 = €900$ a month. They propose raising spend to €9,000. Their fee becomes $0.15 \times 9{,}000 = €1{,}350$, which is €450 more for them whatever the extra €3,000 does for Amara. If the extra €3,000 brings in €6,000 of revenue, she keeps $0.40 \times 6{,}000 = €2{,}400$ of contribution, minus the €3,000 of spend, minus the €450 of extra fee. That is $2{,}400 - 3{,}000 - 450 = -€1{,}050$ a month. She is worse off and the agency is better off.

Commission pays for whatever it is measured on

Commission ties pay to a result. When the agency's income moves with the client's result, the two finally want the same thing. But "a result" can mean two very different numbers. The first is revenue that an ad platform attributes to the ads. The second is the extra profit the work added above what the shop would have made anyway. They sound alike. They behave very differently.

An example: Agency one also offers a commission of 8% on revenue that Google Ads and Meta attribute to the ads. In March the platforms attribute €30,000 to the ads, so the commission is $0.08 \times 30{,}000 = €2{,}400$. The independent specialist offers instead 20% of the contribution added above an agreed baseline. If Amara's revenue is €6,000 above the baseline, the added contribution is $0.40 \times 6{,}000 = €2{,}400$, and the commission is $0.20 \times 2{,}400 = €480$.

Same month, same shop, €2,400 against €480. The gap is the difference between credit and cause.

The flat month test

The quickest way to see what a deal rewards is to imagine a month in which nothing improves. What does each deal pay?

Worked example: a month where nothing changes

Take a month in which Amara's revenue is exactly €60,000, the same as her baseline, and her spend stays at €6,000.

The hourly agency bills its 25 hours, €2,250. The retainer agency bills €2,500. The fee on spend is €900. The commission on attributed revenue is still about €2,400, because the platforms go on attributing roughly €30,000 to the ads whether or not those sales were new. The commission on added contribution is €0, because nothing was added.

The horizontal axis lists the five deals. The vertical axis is what Amara pays the agency for a month in which her shop did exactly what it would have done without them. The takeaway: four of the five deals pay well for nothing. Only the commission on added contribution pays nothing for nothing. That is its strength, and also its weakness, because a good agency can have a flat month through no fault of its own.

Why commission works: pay that follows the result

If effort makes a good outcome more likely, and pay rises with the outcome, then effort raises the agency's expected pay. Under a flat fee it does not.

Expected value is the average result you would get if you could replay the same decision many times, each outcome weighted by its probability.

$$E[t] = \sum_{i} p_i \, t_i$$

Here $E[t]$ is the agency's expected pay for the month, $p_i$ is the probability of outcome $i$, $t_i$ is what the agency is paid if outcome $i$ happens, and $\sum_{i}$ means adding up over every possible outcome. The idea is set out from scratch in Foundations: expected value.

Worked example: does effort pay?

Assume, for illustration, that each month Amara's shop has two possible outcomes. Either the work adds €10,000 of contribution above the baseline, a good month, or it adds nothing, a flat month.

With real effort, the agency makes a good month 60% likely. With a coasting effort, only 20% likely.

Under the commission of 20% on added contribution, a good month pays $0.20 \times 10{,}000 = €2{,}000$ and a flat month pays €0.

Expected pay with effort: $0.6 \times 2{,}000 + 0.4 \times 0 = €1{,}200$.

Expected pay when coasting: $0.2 \times 2{,}000 + 0.8 \times 0 = €400$.

Effort is worth $1{,}200 - 400 = €800$ a month to the agency. Under the €2,500 retainer, effort is worth $2{,}500 - 2{,}500 = €0$ to them. That €800 is the whole argument for commission in one number. The incentive exists because pay follows the result.

The catch: the commission only rewards effort if the measured result moves with effort. If it moves with something else, the commission rewards that instead.

What goes wrong with commission: credit is not cause

A commission is only as honest as the number it is calculated on. Four things push that number away from what the agency really caused.

Attribution decides who gets the credit, not who caused the sale

When a customer clicks an ad and later buys, the ad platform has to decide how much credit each ad gets. Google says it plainly: attribution models let you choose how much credit each ad interaction gets for your conversions. Its default for most conversion actions is data driven attribution, which distributes the credit based on your past data.¹

That is a rule for sharing credit, not a test of whether the sale would have happened without the ads. A customer who already meant to buy and clicked an ad on the way still hands it credit. And each platform counts its own touches, so Google and Meta can both claim the same order.

An example: In April Amara's shop takes €60,000 of orders. Google Ads attributes €24,000 to its ads and Meta attributes €14,000 to its ads. Added up, the platforms claim €38,000. But some of those orders were claimed by both, and some came from people who would have bought anyway. A commission of 8% on the claimed €38,000 is $0.08 \times 38{,}000 = €3{,}040$, whatever the true number was.

Brand search sells people what they were already buying

The easiest sales for an ad to claim are from people searching for your own name. They already know you. A good few of them would click the ordinary result just below the ad.

This is not a guess. In a large field experiment at eBay, Tom Blake, Chris Nosko and Steven Tadelis switched off paid search ads and measured what happened. They found that ads on brand keywords had no measurable short term benefit. They also found that the frequent users who make up much of the ad spend were barely moved by the ads, while new and infrequent users were.² eBay is a very well known brand, and a small shop with a less familiar name may see more effect. That is exactly why you test it rather than assume it.

Worked example: the brand search shuffle

Agency one, on its 8% commission on attributed revenue, moves €800 a month into ads on searches for "suirside home". The platform credits those ads with €8,000 of revenue. On paper that is a return of €10 for every euro spent, the best campaign in the account.

Amara runs a test. For four weeks she switches the brand ads off in half the country and compares the two halves. Assume, for the example, that orders from people searching her name barely change, and that only 15% of the €8,000 turns out to be caused by the ads. That is $0.15 \times 8{,}000 = €1{,}200$ of truly new revenue.

Step one, the contribution on the new revenue: $0.40 \times 1{,}200 = €480$.

Step two, the commission on the attributed revenue: $0.08 \times 8{,}000 = €640$.

Step three, the result for Amara: $480 - 800 - 640 = -€960$ a month.

The agency earned €640 for moving credit from the free column to the paid column. Amara paid for sales she already had.

Returns turn a sale into a refund after the commission is counted

An order is recorded when it is placed. A return happens weeks later. If the commission is calculated on orders as they come in, it is calculated on revenue some of which will go back to the customer.

Google Ads lets you restate the value of a recorded conversion or retract it completely, for example when items in an order are returned, and its help page says adjustments can be made within 54 days.³ Whether anyone actually uploads those adjustments is another matter, and a commission deal should say who does it.

An example: In November the platforms attribute €30,000 of orders to the ads. Assume, for this example only, that 15% of that value is later returned and refunded. The commission on gross orders is $0.08 \times 30{,}000 = €2{,}400$. The commission on the net figure, $30{,}000 \times 0.85 = €25{,}500$, would be $0.08 \times 25{,}500 = €2{,}040$. Amara overpays €360 that month for sales that came back.

Natural growth and the season pay the agency for the weather

Some growth would happen anyway. Your shop is older, more people know it, the season turns. A commission on all revenue, or on all growth, pays the agency for that too.

An example: Suirside Home's December revenue is €84,000, against the €60,000 average month. That looks like a 40% jump. But last December was €78,000, and over the past year the shop has grown about 5% on its own. The fair baseline is $78{,}000 \times 1.05 = €81{,}900$. The growth the work may have added is $84{,}000 - 81{,}900 = €2{,}100$, not €24,000.

Safeguards that keep a commission honest

Each problem has a fix.

Pay on margin, not on revenue. A sale at a thin margin is worth less to the shop, so the commission should be too.

Pay on revenue net of returns. Agree who uploads the return adjustments and when the month is closed.

Pay only above a baseline. The baseline is agreed in writing before the work starts, from the shop's own history, adjusted for the season and for the growth the shop was already making.

Test the claims that matter most. Brand search and retargeting are the usual suspects. A holdout, switching something off for part of the audience or part of the country, measures what the ads caused rather than what they were credited with. The methods behind that are in causal inference for marketing and in media mix modelling.

Keep the accounts, the tracking and the data in the client's name. Then the client can always check the numbers, and nobody can make leaving expensive.

Put together, the commission is calculated on the extra, not the total.

$$C = k \times m \times \left( R_{\text{net}} - B \right)$$

Here $C$ is the commission for the month, $k$ is the commission rate as a share of the added contribution, $m$ is the contribution margin as a share of revenue, $R_{\text{net}}$ is the month's revenue after returns, and $B$ is the agreed baseline, the revenue the shop would have made without the work. If $R_{\text{net}}$ is below $B$, the commission is zero. The point of paying on $R_{\text{net}} - B$ rather than on all revenue is the difference between the marginal and the average, explained in Foundations: marginal versus average.

Worked example: Amara's May commission

In May Suirside Home takes €72,000 of orders. Returns from those orders come to €3,600, so $R_{\text{net}} = 72{,}000 - 3{,}600 = €68{,}400$.

The baseline was agreed before the work started, from last year's May and the shop's own growth: $B = €62{,}000$.

Step one, the revenue above the baseline: $68{,}400 - 62{,}000 = €6{,}400$.

Step two, the contribution on it: $0.40 \times 6{,}400 = €2{,}560$.

Step three, the commission at $k = 0.20$: $0.20 \times 2{,}560 = €512$.

Had the platforms attributed €36,000 to the ads, an 8% commission on attributed revenue would pay $0.08 \times 36{,}000 = €2{,}880$. The honest commission is smaller because it pays only for the profit in the €6,400 that is genuinely extra.

How much risk the agency should carry depends on the measurement

The old version of this article said that under commission the agency carries the bulk of the risk. That sounds generous. It is also not what good contract theory recommends, and it is not how my own model works.

The reason is noise. Even the honest commission above is calculated on a measured result, and the measured result moves for reasons the agency does not control: a rainy week, a competitor's sale, a courier strike, an algorithm change. The Royal Swedish Academy of Sciences, explaining the work behind the 2016 Nobel memorial prize in economics, puts the rule simply. The harder it is to observe effort, because many factors blur the link between effort and performance, the less pay should be based on performance. In high risk settings pay should lean towards a fixed salary, and in stable ones towards a performance measure.⁴

There is a second reason, from the same work. When a job has several tasks and some are hard to measure, strong pay on the easy ones pulls effort away from the hard ones. The Academy's example is the teacher paid on test scores who neglects creativity and independent thinking.⁴ In growth work the easy task is harvesting this month's sales. The hard task is building tracking, creative and new audiences that pay off later. An agency paid only on this month's number will harvest.

The size of the swing in pay is easy to see.

$$\sigma_C = k \times \sigma_{\varepsilon}$$

Here $\sigma_C$ is the standard deviation of the commission from month to month, $k$ is the commission rate, and $\sigma_{\varepsilon}$ is the standard deviation of the noise in the measured result, the part of the swing the agency does not cause. Standard deviation is the usual measure of how much a number jumps around its average, explained in Foundations: spread.

Double the rate and the agency's income swings twice as hard, and a noisier measure makes it swing harder too. An agency carrying a big swing will want paying for it somewhere. So the fix is twofold: measure better, which shrinks $\sigma_{\varepsilon}$, and keep a fixed part of the pay, which makes the swing bearable.

Worked example: three noisy months

Assume Amara's monthly added contribution swings with a standard deviation of €4,000 for reasons nobody controls. At $k = 0.20$ the commission swings by $0.20 \times 4{,}000 = €800$ a month around its average, on top of whatever the work achieves.

Here are three months in which the agency works equally hard. Measured added contribution: minus €3,000 in January, a wet and quiet month, plus €2,000 in February, plus €10,000 in March, when a feature in a magazine lands.

Under a pure commission the agency earns €0, then $0.20 \times 2{,}000 = €400$, then $0.20 \times 10{,}000 = €2{,}000$. Same effort, pay from nothing to €2,000. And the €2,000 month was mostly the magazine.

Now add an invented base fee of €2,000 with the rule "base fee or commission, whichever is higher". The agency earns €2,000 in each of the three months. The commission would only take over once it beat €2,000, which needs an added contribution above $2{,}000 / 0.20 = €10{,}000$, or added revenue above $10{,}000 / 0.40 = €25{,}000$ a month.

That is the trade, stated honestly. The base fee protects the agency from noise, and it means Amara pays something in a flat month. The commission still points both sides at the same number, and it takes over when the growth is large and sustained enough to stand out from the noise. Better measurement, with a seasonal baseline and holdout tests, lets more of the pay sit on the result.

Deciding whether a commission deal fits

Not every shop is ready for a commission deal, and not every piece of work should be paid that way. A short sequence of questions sorts it out.

Worked example: a shop that is not ready yet

Cian runs Thomond Coffee Supply, an online shop in Limerick at thomondcoffee.example, launched three months ago. He has €15,000 of sales in total and no clean tracking. He asks for a commission deal.

Step one, the baseline. Three months of a brand new shop cannot say what month four would have been without help, so any baseline is a guess.

Step two, the measurement. Without reliable tracking the commission would rest on the platforms' own attribution, the credit problem from earlier.

Step three, the answer. A fixed price to build the tracking, or a small retainer while the history builds, is fairer to Cian. In a year the question can come back with numbers behind it.

How my own commission model works

Here is what my pricing and my terms actually say, and nothing more.

The commission model is offered for online shops. It has a monthly base fee, which is the floor. The commission is calculated on the profit my work adds beyond your natural growth, adjusted for repeat customers. You pay whichever is higher, the base fee or the commission, never both. If the commission beats the base fee, the base fee disappears.

Schedule A of my terms sets out the mechanics. Before the work starts, a revenue baseline and a natural growth rate are agreed in writing, from the shop's history, typically the preceding six to twelve months, adjusted for the season where the data supports it. Commission is paid only on measured performance above that baseline, never on the baseline itself. The base fee is agreed in writing and cannot be waived. If sales fall or growth stalls, only the base fee is payable, and commission already earned is never clawed back. Commission engagements are offered at my discretion, typically only where the shop's existing revenue justifies a results based model. And your ad accounts, tracking and data are yours from day one.

My FAQ puts the condition in one line: when tracking is clean enough to measure my contribution honestly, I am happy to tie part of my income to results. Part, not all. That is the theory above in practice. Commission aligns the incentives. The base fee is there because no measurement is perfect, and the baseline, the margin and the repeat customer adjustment are there so that the commission pays for what the work added, not for the weather or for sales you already had.

The commission rate and the base fee are agreed per shop. The commission playground lets you model a deal with your own numbers, and its figures are illustrative estimates, not an offer. If you want to see a whole year of this model, invoice by invoice, including a slow case, I walk through it in commission based pricing month by month.

When an hourly rate or a retainer is the better deal

Commission does not suit everything. It suits growth work where the result can be measured and the history exists to set a baseline.

A bounded task with a clear scope, a migration, a tracking audit, a new checkout step, is better on a fixed quote or an hourly rate. There is no result to share, just a job to do well. Ongoing support with uneven hours is often best on a retainer, where the hours are averaged over a period. A brand new shop, like Cian's, needs history before anyone can measure what was added.

An example: Amara also wants her product feed rebuilt, a job with a clear end. On an invented €90 an hour and an estimate of 12 hours, the quote is $12 \times €90 = €1{,}080$. Trying to pay for that on commission would mean arguing for months about how much of next year's revenue the feed caused. A fixed quote is simpler, and nobody is tempted to game it.

The question to ask of every deal is the same one this article started with. What does this fee pay for? What will a sensible person do to earn it? And is that what you actually want? Choosing the person is its own question, covered in my guide to choosing a growth agency or specialist.

Footnotes

1. Google, About attribution models, Google Ads Help, https://support.google.com/google-ads/answer/6259715?hl=en-GB

2. Thomas Blake, Chris Nosko and Steven Tadelis, Consumer Heterogeneity and Paid Search Effectiveness: A Large Scale Field Experiment, NBER Working Paper 20171, National Bureau of Economic Research, 2014, later published in Econometrica, 2015, https://www.nber.org/papers/w20171

3. Google, About conversion adjustments, Google Ads Help, https://support.google.com/google-ads/answer/7686280?hl=en-GB

4. The Royal Swedish Academy of Sciences, Contract Theory, popular science background to the Prize in Economic Sciences in Memory of Alfred Nobel 2016, Nobel Prize Outreach, 2016, https://www.nobelprize.org/uploads/2018/06/popular-economicsciences2016.pdf