Short answer
You find a channel’s saturation point by estimating its response at many spend levels, then confirming it with a match-market test. That takes three things: enough spend variation in your history, a model that separates ad effect from baseline, and a real-world test that says whether the model is right.
The knee of the curve tells you what to fund, what to cap, and what to move. Getting there takes three things.
Pick the one that sounds like you.
Use your platform reporting.
One or two channels does not need a model. There is not enough going on for one to untangle.
Watch out · We would turn this work down. It would be a waste of your money.
What data do you need?
- Weekly spend by channel, two to three years, with real ups and downs. Flat spend teaches a model nothing.
- One outcome, weekly: revenue, orders, store visits, or leads.
- Everything that moves sales without ads: seasons, holidays, promotions, price, distribution, the economy.
- Geography, if you control spend by market. It is the most valuable thing in a dataset.
How do you separate ads from everything else?
- Baseline
- The sales you would get with zero advertising. Not a flat line — it moves with season, holidays, price, and the economy. This is where we spend most of our time. Get it wrong and every channel looks better than it is.
How do you know the curve is right?
- 01Hide the last three to six months. Build the model without them. Make it predict them. We will not ship above 10% error.
- 02Run a match-market test — the channel up or off in some markets, comparable markets left alone.
- 03When the test and the model disagree, the test wins and the model gets rebuilt.
Result
+18% lift · 95% interval 11% to 25%
The bar is the range the true effect probably sits in. If it touches zero, you did not measure a result — you measured your own uncertainty.
The whole interval is above zero. This channel is doing real work — fund it further and re-test at the higher level.
“The model has to match the real world before you move a dollar on it.”
What do you do with the answer?
- 01Rank channels by what the next dollar returns. Not average return. Not total volume.
- 02Cap the channels past their knee. Do not zero them — that is a worse mistake.
- 03Move the freed budget to channels with room, in steps you can measure.
- 04Test again after the move. Curves shift when spend shifts.
How much history do I need?
Two years of weekly data, mostly so the model sees your seasonal cycle twice. With good regional data you can do useful work with less.
Can I find the knee without a model?
Partly. A test gives you one point on the curve. Repeat at several spend levels and you sketch it by hand. Accurate, slow, expensive.
What if my spend has been flat?
Create the variation on purpose. Move spend up in some markets and down in others. It is the only way to learn anything.
How often should curves be refreshed?
Quarterly. More often and you are fitting noise. Less often and you are planning against a stale picture.
Find your wasted third
Want this run on your own numbers?
The two-week audit shows you where each of your channels sits on its own curve — what to cut, what to grow, and how much budget is sitting in the wrong place.
Book the two-week auditKeep reading
- What ad saturation actually looks likeSaturation is the point where the next dollar in a channel stops paying for itself. What the curve looks like, how to read it, and what it costs to be past it.
- What a 3.87% error on unseen data actually provesOne of our models predicted unseen months with 3.87% error. What that proves, what it does not, and why almost nobody publishes the number.