Every winning ad starts dying the day it starts winning, and the more money you put behind it, the faster it goes. That's creative fatigue. Most teams answer it by testing everything and shipping more variants, then wonder why they're losing clarity. The technician sees the decay as a frequency problem and reaches for a fresh batch of assets. I came up through copywriting before I worked as a digital futurist, so I see it as a words problem: the idea ran out of people to talk to. At $10M a month, that decay sets your production plan whether you've written one or not.
The industry is counting assets
One industry survey of video buyers, released in July 2026 [1], shows how fast production has changed. Nearly two thirds of them now use generative AI to make video creative, up from about half the year before. They expect about a third of their video ad assets to involve generative AI this year, up from a quarter, with a projection of around 43% by 2027. The same survey puts US digital video ad spend at double what it was five years ago.
So production capacity is exploding. More assets, cheaper, faster. Hold that number loosely, because it's counting the wrong thing.
Twenty variants, one well
My read: the industry is measuring output the way a technician would. Assets are going up. Ideas may not be. Nobody's survey asks how many new things those assets actually say.
Most generative production makes variations: a new hook on the same angle, a new face on the same script, a new aspect ratio, a new color grade. Variations matter. They extend a winning idea's life and they fit it to new placements. But a variation pulls from the same pool of people who respond to that idea. Ten variants of one concept are mostly ten ways to reach the same buyers.
When a team sees a winning ad fatigue and responds with twenty AI variants, they often buy a week or two. Then all twenty fatigue together, because they were drinking from one well.
Same dying ad, two different fixes
Put the same dying ad in front of two people.
The technician pulls up frequency, CPM and CTR by day. He sees the line bend, calls it fatigue, and orders twenty fresh cuts. New faces, new hooks, new aspect ratios. He's right about the line. What's underneath it is the part he misses.
The strategist asks one question: who was this message for, and how many of them are left? If the ad said “stop losing your weekends to inventory counts,” it was talking to a specific person in a specific situation. When those people have heard it, twenty new edits of the same sentence won't find anyone new. A different sentence will. “Your accountant found the shrinkage before you did” opens a different room full of different buyers.
The number bendswhen the messageruns out of people.Only new words opennew people.
That's the whole thesis of the Lab. None of these numbers matter until the words are right. The dashboards tell you when an idea is dying. Only the strategist can tell you what to say next.
Interest Half-Life: the clock behind creative fatigue
I've used Interest Half-Life to describe how buyer intent decays when nobody responds to it. The same law applies to a creative concept.
A concept has a pool: the people who'll respond to that specific idea at a price you can afford. Each impression spends some of the pool. The concept's efficiency doesn't fall off a cliff. It decays, steadily, as the pool shrinks and the system reaches further out for people who care less.
Define the concept half-life as the time it takes a concept's marginal cost per acquisition to climb halfway from its launch level to your break-even level. Three things set it:
- Pool size. How many people the idea speaks to. Broad ideas have bigger pools and weaker pull. Specific ideas have smaller pools and stronger pull.
- Burn rate. Daily spend behind the concept. Spend twice as fast, burn the pool twice as fast.
- Refresh lift. How much a variation extends the pool. Usually a modest amount, because a variant reaches mostly the same people.
The math: how many new concepts $10M a month eats (illustrative)
These are modeled numbers, a way to think and no forecast for your account. Use your own.
Half-life and spend. Suppose a concept's half-life is 30 days at $20,000 a day. Push it to $100,000 a day and, if the pool burns roughly in proportion to spend, the half-life drops to about 6 days. Same idea, same quality. It just dies five times faster.
Concepts needed per month. Say a winning concept, before it becomes inefficient, delivers about 40 million good impressions. At a $15 CPM, that's about $600,000 of efficient spend per winner.
- At $1M a month: about 1.7 winners needed per month.
- At $5M a month: about 8.
- At $10M a month: about 17.
Concepts tested per month. If one in six tested concepts becomes a winner, that's:
- At $1M: about 10 tests a month.
- At $5M: about 50.
- At $10M: about 100.
Testing cost. If it takes about $7,500 to read a concept with confidence, 100 tests a month is $750,000, about 7.5% of a $10M budget. Think of it as the price of staying efficient. Teams that refuse to spend it pay more than that in fatigue, they just pay it inside the CPA where nobody labels it.
Where variants fit. Say the first good variant extends a concept's efficient life by about 15%, and each one after that adds less, because it reaches mostly the same people. That's an assumption, but it's the pattern I'd expect most accounts to show. Twenty variants might then add 40% or 50% to the concept's life in total. Useful, and nowhere near the twenty fresh lives the asset count implies.
Variants are amultiplier onconcepts. They can'tstand in for them.
Reading decay in your own account. Your own account already shows where your concepts start to bend. When several of them bend at about the same spend, you've found the size of your pool, and with it the number of new ideas you owe it every month. Reading that bend concept by concept is a Lab tool.
At $10M a month, about 100 tests cost around $750,000: the price of staying efficient.
Four things fatigue does when nobody's counting
- Your best ad becomes your biggest risk. The more spend concentrates on one concept, the shorter its half-life and the steeper the cliff when it goes. Concentration feels efficient until the week it isn't.
- Hit rate is set upstream, by the words. One in six versus one in twelve doubles the tests you need. Hit rate rises when concepts start from real buyer evidence: sales calls, reviews, objections, the exact words customers use. It falls when concepts start from a brainstorm about formats.
- AI moves the bottleneck upstream. When production is nearly free, the scarce resource is the insight that becomes a concept. Spend the savings on finding new buyer situations.
- Fatigue hides inside blended numbers. A dying concept still converts, just more expensively, and blended reporting averages it with the new winners, so the dashboard says you're doing great. By the time the average moves, you've overspent for weeks.
Why refreshing every two to four weeks breaks between $1M and $10M a month
The standard rule of thumb says refresh creative every two to four weeks. At $1M a month that's a reasonable rule.
Between $3M and $10M, a calendar rule stops working, because half-life depends on spend. A concept carrying $20,000 a day might be fine for a month. The same concept at $150,000 a day might need replacing in a week. Refresh on measured decay.
Then there's “test lots of hooks.” Hooks are variations. At scale, test angles and buyer situations, the things that open new pools, and use hooks to stretch the winners.
Past $10M a month, concept velocity is an org chart
- Concept velocity becomes an org chart question. At 100 concept tests a month, you need a pipeline: people mining buyer evidence, people turning it into words that land, a production layer, and a testing system with clear kill and scale rules. AI makes the production layer cheap. The words still take people with judgment, and most teams at this size are staffed heavy on technicians and light on people who can write them.
- The test budget needs its own line. If tests compete with proven ads for money inside the same campaign, proven ads win the auction and the tests never get read. Ring-fence the test spend.
- Agencies get paid on assets, but you need concepts. If an agency's scope is “40 assets a month,” you'll get 40 assets, many of them variants. Write concept counts into the scope, defined as new buyer situations or angles, with variants counted separately.
- Fatigue becomes a portfolio problem. At this spend you're managing dozens of concepts at different points in their decay. Somebody needs a view of the whole portfolio: how much spend sits on concepts past half-life, and how much new concept volume is coming in to replace them.
- Finance needs to see the replacement math. Testing at 7.5% of spend looks like waste until a CFO sees the decay curve.
The monthly check for creative fatigue
Whether you run growth in-house, run the agency doing the work, or sit in the CFO's chair asking why CPA keeps creeping, ask two things every month. How much of your spend sits on concepts already past their half-life? And how many genuinely new concepts did you ship against it? When the first number climbs and the second one doesn't, the fatigue bill is already coming due, whatever the dashboard says.
The full monthly checks, with the formulas that turn your spend into a replacement target, are what members get inside the iFix Lab.
Questions senior buyers ask
How would I know a new batch of ads isn't just the same old thing in a new outfit?
A new concept would make sense to a different buyer, in a different situation, for a different reason. If the same person would respond to both, it's a variant.
Does generative AI help with concept velocity at all?
Yes, in two places: it can mine and sort buyer evidence faster, and it makes testing a rough version of a concept cheap. It doesn't replace the judgment that decides which buyer situation is worth an idea, or which sentence will make that buyer stop.
What's a good hit rate?
It varies by category and by how close your concepts start to real buyer evidence. Measure your own, by concept source, and you'll see which sources earn their keep.
Should we cap spend per concept?
A soft cap helps. Pushing one concept harder shortens its life faster than it adds results. Spreading spend across several winners usually holds marginal cost lower.
How long should a concept test run?
Long enough to reach the number of conversions you set before launch as your read threshold. Decide that number first, then the budget per test follows from your cost per conversion.
What I'd do this month
Count your genuinely new concepts from the last 90 days, leaving every variant and resize out of the count. Then divide your monthly spend by what a winner carries before it fades.
If the second number is bigger than the first, you don't have an asset shortage. You have a words shortage, and no amount of production fixes it. That's the shift the Lab exists to make: technicians who can run every dashboard learning to pull up and decide what the business should say. Inside the iFix Lab, the Hype vs Working Ledger keeps dated verdicts on trends like this one, plus my own experiment log, flops included, so you can see what held up before you spend on it. Every week a concept runs past half-life, the cost hides in your CPA. Join the iFix Lab for $97 a month. Cancel any time, with a 14-day refund on your first month. Next note: why connected TV makes the limits of your message impossible to hide.