Diminishing returns vs. saturation in MMM: Why many models cap budgets too early
Diminishing returns and saturation aren't the same. Learn why many MMMs force saturation curves that cap budgets early, and how to spot real limits.
Linnea Zielinski · 12 min read
Old hiking maps sometimes show a trail stopping at a ridge. It looks like the trail ends there, so most hikers turn around. Often, though, the trail keeps going for miles. The map ended where the surveyor stopped walking, maybe because daylight ran out or the budget for the survey did. Hikers who trust the map miss the best views on the route because they've mistaken the edge of someone's survey for the edge of the trail.
Many marketing mix models make the same mistake with your budget. They learn how a campaign performs from the spend levels you've already tried, and then they draw the curve flattening out right around where your spending stopped. That point usually reflects a quarterly budget cap or a finance guardrail—a decision someone made in a planning meeting—not a limit in what the market will respond to. When brands treat that flattening as real saturation, they pull money out of campaigns that still have room to grow and move it into weaker ones.
Key takeaways
- Diminishing returns means each additional dollar earns less than the one before, while saturation means additional spend earns almost nothing, so a campaign can show diminishing returns long before it's saturated.
- Many traditional and open-source MMMs use response curves that are built to flatten out eventually, which means they show saturation whether or not your data supports it.
- These models infer where saturation happens from your historical spend range, so they often mistake your own budget limits for the point where a campaign stops working.
- Prescient's saturation research found that a simple linear model matched real campaign behavior more closely than the most common saturation curves, which suggests many brands are underspending on campaigns that could scale.
- Saturation isn't permanent or universal, since upper-funnel campaigns can raise the ceiling for lower-funnel ones, and seasonality and promotions change how efficiently every campaign performs.
- Budget decisions should rest on how returns change as spend increases, judged at the campaign level and against your profitability goals, instead of on historical average ROAS.
- Before trusting a model's saturation curves, ask what shapes it allows, how it handles spend beyond your historical range, and how it shows uncertainty.
Diminishing returns and saturation aren't the same thing
These two terms get used interchangeably all the time, but they describe different conditions, and they lead to different decisions. Here's the simplest way to separate them.
| Diminishing returns | Saturation | |
| What it means | Each additional dollar earns less than the previous one | Additional spend earns little or nothing more |
| What happens to revenue | Keeps rising, just more slowly | Mostly stops rising |
| What it means for budget | Check whether the added revenue is still worth the spend | Move additional spend somewhere else |
| How common it is | Common, and often fine | Less common than most models suggest |
A campaign with diminishing returns can still be one of the best places to put more money. If it's earning less on each extra dollar but still more than your other options, scaling it makes sense.
There's also a third possibility that most models ignore. Some campaigns keep returning about the same amount on each extra dollar across the spend levels you actually use, which is called a linear response. Others perform better once they reach a certain scale. And some behave unpredictably, with efficiency rising and falling at different spend levels rather than following a smooth curve.
Why campaigns hit diminishing returns in the first place
Diminishing returns do happen, and it's worth knowing why. When you understand the cause, you can tell whether it's a real limit or something you can fix.
The most common reasons include:
- Running out of the most receptive audience, so additional spend reaches people less likely to buy.
- Rising frequency, where more impressions go to people who've already seen your ads.
- Higher costs at scale, since winning more auctions or buying more inventory can get pricier.
- Creative fatigue, when the same ads stop landing after repeated exposure.
- Limited demand, like a branded search campaign that can only capture as many searches as people make.
Many of these limits come from an older media world. The idea of saturation traces back to a 1957 operations research study of ad response in an era of radio, print, and magazines, when audiences were fixed and targeting was blunt. Modern digital platforms can reach new audiences, refresh creative, and target far more precisely, so those assumptions don't automatically hold today.
How most MMMs model response curves
Every marketing mix model needs some way to describe how a campaign's results change as you spend more. That description is called a response curve, and the shape a model is allowed to use has a big effect on the budget advice it gives.
Curves that are built to flatten
Most traditional MMMs and widely used open-source frameworks use a small set of mathematical curves, like the Hill, Weibull, or Michaelis–Menten functions. They differ in shape, but they share one key trait: every one of them levels off eventually by design. No matter what your data says, the model's curve will flatten at some point, because that's the only kind of shape it's allowed to draw.
That means saturation isn't something these models find in your data. It's something they assume and then place somewhere along your spend range. (All MMMs make assumptions, by the way, you just need to make sure yours is making assumptions that make sense to you and reflect the realities of marketing.)
Mistaking your budget limits for market limits
So where does the model decide the flattening happens? It works from the range of spend it has seen. If you've never spent more than $20,000 a day on a campaign, the model has no evidence about what happens at $40,000, and its curve fills that gap with the flattening it was built to include.
The problem is that your historical spend range was set by budget decisions, like quarterly plans, finance guardrails, or a channel manager's comfort level. None of those tell you where the market stops responding. The model ends up treating where your spending happened to stop as the point where the campaign stops working, which is the hiking-map problem in a nutshell.
What happens when a linear campaign gets forced into a curve
When a campaign is actually performing in a straight line but the model can only draw curves, something has to give. The model bends the curve anyway, then makes up the difference by handing some of that campaign's revenue to baseline demand or to other channels.
That creates two problems at once:
- The campaign looks like it's running out of room when it isn't, so the model recommends capping it.
- Credit ends up in the wrong places, which distorts the results for every other channel too.
What Prescient's saturation research found
Prescient tested this directly. In our research on saturation assumptions, we compared common saturation curves against a simple straight-line model, measuring how closely each matched a flexible, data-driven estimate of how real campaigns behaved.
| Response shape | Average error against observed campaign behavior |
|---|---|
| Linear (no assumed saturation) | 33.2% |
| Hill curve | 52.7% |
| Weibull curve | 331.2% |
The straight line was the closest match, and the most common saturation curves introduced much more error. The study also found campaigns where returns grew with spend, and others where spending more actually earned less, which neither smooth curve can represent. The research used campaign data from before Apple's iOS 14.5 update and relied on platform-reported revenue, so it's a strong signal rather than the final word. Still, it points to the same conclusion as the rest of the evidence here: many brands are likely underspending on campaigns that could keep scaling.
Why average ROAS hides what's happening at the edge
Even with a good response curve, a lot of budget mistakes come from looking at the wrong number. Average ROAS tells you how a campaign has performed across all of its spend. What matters for a budget change is how much more revenue you'd get from scaling it up, which can be very different.
Here's a hypothetical example of two campaigns with the same spend:
| Campaign A | Campaign B | |
| Current monthly spend | $100,000 | $100,000 |
| Average ROAS | 5.0 | 3.5 |
| Added revenue from a 20% spend increase | $26,000 | $58,000 |
| Return on that added spend | 1.3 | 2.9 |
Campaign A looks like the stronger performer, but most of its return comes from the early part of its spend. Campaign B is the better place to scale. A team using average ROAS alone would send the budget increase to the wrong campaign.
The profitable point usually comes before saturation
There's one more distinction worth keeping in mind. The point where a campaign stops being profitable usually comes well before it's saturated. If your margins mean you need $2.50 in revenue for every extra dollar of ad spend, any increase that returns less than that costs you money, even though it's still driving sales. Saturation tells you when spend stops working. Your margins tell you when it stops paying off.
How Prescient separates real saturation from forced curvature
Prescient's marketing mix model doesn't start from the assumption that every campaign eventually flattens. Our model lets each campaign's response take whatever shape the data supports. If a campaign is performing in a straight line across the spend levels you use, it stays a straight line, and curvature only shows up when there's real evidence for it.
That flexibility matters most in three places.
Different campaigns saturate for different reasons
Lower-funnel campaigns, like branded search, often do hit a natural limit, because they can only capture as many searches as people make. Upper-funnel campaigns, like CTV and YouTube, work differently. They create new interest in your brand, which increases the number of people searching for you later.
That means upper-funnel spend can raise the ceiling on your lower-funnel campaigns instead of hitting a fixed wall of its own. A model that treats each campaign's curve as independent will miss that relationship, and it may tell you to cut the very campaigns that are making your branded search more scalable.
Efficiency changes with the season, not just with spend
The same spend can produce very different results in July and December. Promotions, holidays, and competitor activity all change how responsive shoppers are. Our model treats a campaign's efficiency as something that changes over time, which helps it separate a seasonal slowdown from real saturation. A campaign that looks tapped out in a slow month might have plenty of room when demand picks up.
Saturation is measured campaign by campaign
A channel-level curve can hide big differences between campaigns. Paid social might look like it's flattening overall while a broad prospecting campaign inside it still has plenty of room, and a retargeting campaign next to it is nearly maxed out.
Prescient measures each campaign individually and updates daily. In the platform, you can see each campaign's saturation curve along with a confidence band showing how certain the model is at different spend levels. Our Media Forecasting feature then uses those curves, along with what the model has learned about your seasonality and halo effects, to forecast what happens when you scale a campaign or shift budget into it from other channels.
What forced curves can cost during peak season
The stakes get highest during peak periods like Black Friday and Cyber Monday, when budgets are biggest and spend rises alongside natural demand. That's exactly when assumptions about saturation and baseline demand do the most damage.
In a simulation benchmark Prescient ran at the channel level, with known correct answers built in, two widely used open-source MMMs gave very different peak-season advice. This benchmark hasn't been peer-reviewed, but the pattern is worth knowing:
| Model | Peak-season budget recommendation vs. the correct amount | Main reason |
|---|---|---|
| Open-source model A | Overspend by about 81% | Gave ads credit for holiday demand that would've happened anyway |
| Open-source model B | Underspend by about 11.5% | Forced curves underestimated how far campaigns could scale |
| Prescient's model | Within about 1% | Let response shapes and baseline demand be estimated together |
The 81% overspending finding has been reported publicly, and the underspending result shows the saturation problem directly. Errors that seem modest in a model's averages can turn into big budget mistakes once a budget planning tool acts on them.
Common mistakes when reading saturation curves
Even marketers who understand the concepts can fall into a few traps when it's time to act on a model's curves. These are the ones that come up most often.
| Mistake | Why it's a problem | What to do instead |
|---|---|---|
| Treating diminishing returns as a reason to cut | A campaign earning less per dollar can still beat your alternatives | Compare the added return against other campaigns and your margins |
| Trusting curves far beyond your spend history | The model has little evidence there, so the shape comes from its assumptions | Check how much uncertainty the model shows outside your usual range |
| Reading channel curves as campaign curves | Strong and weak campaigns get averaged together | Look at saturation campaign by campaign |
| Assuming saturation is permanent | New creative, new audiences, and seasonality all shift the curve | Re-check curves as conditions change |
| Judging a model by how well it predicts revenue | A model can predict totals well while bending individual curves incorrectly | Ask how the model's credit assignments are tested |
| Ignoring cross-channel effects | Cutting upper-funnel spend can shrink lower-funnel capacity | Use a model that measures halo effects |
How to tell real saturation from other problems
When a model says a campaign is running out of room, it's worth a quick check before cutting it. Performance can flatten for reasons that more spend won't fix, but a budget cut won't fix them either.
| What might be happening | What to look at |
|---|---|
| The audience really is tapped out | Reach, frequency, and audience overlap |
| Creative fatigue | Performance by creative and how long each asset has run |
| Rising media costs | CPMs, CPCs, and auction competition |
| Lower overall demand | Seasonality, category search volume, and promotions |
| Conversion problems | Site conversion rate, checkout issues, and stockouts |
| The model is forcing a curve | What shapes the model allows and how far you've historically spent |
If the issue is creative fatigue, fresh creative may restore performance. If it's a stockout, more ad spend won't help until inventory catches up. And if it's the model's assumptions, the campaign may not be saturated at all.
Some brands also test a few different spend levels to see how a campaign responds, which can be a useful data point. Prescient's view is that incrementality tests are locally accurate and globally inaccurate, since they describe one window and one set of conditions. That's why Prescient shows brands their model's accuracy with and without a test included, and the brand decides whether to build it in.
Traditional MMM vs. open-source MMM vs. Prescient
Response curves are one of the clearest places where MMMs differ, and those differences flow straight into budget advice. Here's how the main approaches compare.
| Traditional MMM | Open-source MMM | Prescient's MMM | |
| Response curve shapes | Fixed curves that flatten by design | Fixed curves that flatten by design | Flexible shapes that follow the data |
| Where saturation comes from | Built into the curve and placed near your historical spend limit | Built into the curve and placed near your historical spend limit | Only appears when the data shows it |
| Campaigns performing in a straight line | Forced into a curve | Forced into a curve | Stay linear when the data supports it |
| Seasonal efficiency changes | Usually treated as baseline | Usually treated as baseline | Measured as changes in campaign efficiency |
| Cross-channel effects on saturation | Rarely modeled | Rarely modeled | Upper-funnel impact on lower-funnel capacity measured through halo effects |
| Level of detail | Usually channel level | Usually channel level | Campaign level, updated daily |
A plain-English guide to the technical terms
Saturation discussions tend to come wrapped in heavy jargon. This quick reference can help when evaluating models or explaining results to leadership.
| Technical term | What it means | Why it matters |
|---|---|---|
| Bounded saturation curve | A formula that always makes a campaign look maxed out eventually | It can tell you to stop scaling a campaign that still has room |
| Budget policy confounding | Mistaking your own spending caps for market limits | It can make a campaign look exhausted just because you never spent more on it |
| Flexible response shapes | Letting campaigns scale in a straight line when the data supports it | It keeps strong campaigns from getting capped too early |
| Optimization regret | Revenue lost by following a model's flawed budget advice | It puts a dollar figure on the cost of bad saturation assumptions |
Questions to ask about any MMM's saturation curves
Not every model treats saturation the same way, and the differences aren't always obvious from a dashboard. These questions can help you find out what's behind the MMM saturation curves.
- What response shapes can the model use, and can campaigns perform in a straight line?
- Can different campaigns have different curve shapes?
- How does the model decide where saturation happens?
- How does it handle spend levels beyond what you've tried before?
- Does it show uncertainty, especially outside your usual spend range?
- Does it measure saturation at the campaign level or only by channel?
- How does it account for seasonality and promotions changing efficiency?
- Does it capture how upper-funnel campaigns affect lower-funnel capacity?
Where Prescient comes in
Prescient's marketing mix model was built so saturation shows up only when your data supports it, not because a formula requires it. Our model lets campaigns scale in a straight line when that's how they're performing, separates seasonal swings from real limits, and measures how upper-funnel campaigns raise the ceiling for lower-funnel ones. Every campaign gets its own saturation curve and confidence band, updated daily.
That gives omnichannel brands a clearer view of which campaigns are truly tapped out and which still have room to grow, plus our Media Forecasting feature to plan budget shifts across channels. If you want to see whether your campaigns are being capped too early, book a demo.