I recently sat through a webinar with Maxim and Seva Ustinov. Three points stuck with me hard enough that I want to unpack them properly. Not because the conversation was weak. Because a few of the “default” answers still cost teams real money at scale.
Here is the short version. Most performance teams waste budget three ways at once. They overpay to learn if a creative works. They ask production for a thousand unique videos a month when the team is already overloaded. And they let simple campaign management mistakes burn money for a day or two before anyone notices.
This piece walks through three takes I would actually run:
Creative testing math: why ~$100 per US creative becomes a $100,000 problem, and how I would cut that without leaving your main markets.
Remix ideation: how to turn 10 burned-out winners into ~100 relaunch-ready videos.
Campaign hygiene + team leverage: the mistakes I keep seeing, and how automation (plus encoded buyer rules) fixes the “rare media buyer” problem.
Take 1. The $100,000 creative testing bill (and the middle ground platforms do not sell you)
The math Maxim put on the table
Maxim said they allocate about $100 per creative for testing in the US. On a single ad, that can sound fine.
Now scale it the way big performance businesses actually operate. Especially in categories that already ship creatives at industrial volume, teams upload thousands of creatives every month into the US market.
Do the multiplication:
1,000 creatives × $100 = $100,000 spent only on creative testing
Typical win rate is roughly 1% to 3%
That means about $97,000 to $99,000 goes to creatives that do not win, every month
That is not a soft inefficiency. That is a tax on finding winners.
Why 7 to 14 day test windows make this worse at volume
Meta, TikTok, and Google generally recommend a minimum of 7 or 14 days for creative testing. The logic is straightforward: one or two windows usually need to fit inside the test so creatives can settle and mature.
That advice is not wrong for a small set of ads. It becomes expensive when you are launching at industrial volume. You are not validating five creatives. You are renting long windows for hundreds or thousands of them, knowing most will never pay you back.
So the real question is not “should we test.” The question is: how do we find a middle ground that still gives us signal, without paying the full 7 to 14 day tax on every creative.
Maxim’s other-market idea, and why I would stay on target markets
Maxim suggested testing on other markets. I understand the appeal. Cheaper markets can look like a shortcut to cheaper learning.
I would still focus on your main target markets.
Customer behavior can differ a lot by market. On my own experience, avatar creatives can perform quite well in one place and much worse in another. Cultural differences matter. If you learn “what works” in a market that does not match your real customers, you can optimize yourself into the wrong creative language.
So I would keep testing where the customers actually are.
What I would do instead: target markets, cheaper signal, auction metrics first
Keep the campaigns in your target markets. Use them for optimization in places where:
conversions are cheaper, and/or
you can collect communication metrics and auction metrics faster
My main focus on the early read is auction metrics:
CTR
CPC
CPM
These tell you how the creative is performing in the auction, and whether the algorithms are picking it up. That is the first filter. If the auction does not like the creative, you usually do not need a full conversion window to know you have a problem.
The second layer: correlation analysis for a target budget per creative
Auction metrics are not the finish line. Ideal next step: run a correlation analysis on historical data to answer one practical question:
What average budget per creative do we need before we can see metrics that later correlate with final conversions?
When you have enough historical data, you can get roughly 80%+ correlation on the way out, plus a concrete target budget per creative.
Ideally, that target budget is lower than what the platform’s 7 to 14 day recommendation implies. That is the middle ground.
You still need enough conversion events for the final conversion metric. In practice, I usually want a minimum of about 10 conversions before I treat that final conversion read as stable.
What you get when this works
You end up with campaigns that run on test metrics, not on a calendar default. You know what budget each creative needs. You stop paying full freight on every loser. And you save a lot of money without pretending other markets magically behave like your core audience.
Take 2. Do not just relaunch burned-out winners. Use them for remix ideation
The production bottleneck behind “we need more video”
On Meta and TikTok, volume matters. If you need something like a thousand videos, and your creative team is asked to produce only unique new work every month, the team gets crushed.
UGC helps. But if you use UGC in a limited way, or you do not get strong performance from it across markets, you still end up with the same problem: too much unique production demand, not enough capacity.
That is the context for the second take from the webinar.
Creative revival is useful. Relaunch-only is too narrow
The webinar discussed analyzing creatives that had strong metrics but burned out, or only ran on limited markets, then relaunching them. That analysis is useful.
My recommendation: do not use it only to relaunch the same creatives.
Use it for ideation of remixes and video rebuilds.
The method: winners + top tags/hooks/CTAs -> remix brief
Combine two analyses:
Creatives that performed well historically, but are burned out or underused
Historically strongest tags, hooks, and CTAs for your audience
Then give both lists to the creative team and ask them to rebuild.
What “rebuild” means in practice:
different hooks
different settings
different models or actors
AI adjustments to models or actors where that helps
short cut reuses where a strong piece still works inside a new assembly
You are not asking the team to invent 100 net-new concepts from a blank page. You are asking them to recombine proven assets with proven language.
The production math that makes this easy to sell
Say you have 10 videos worth relaunching.
If the team makes 10 small variations of each, you get 100 videos.
That is dramatically easier than shooting ~90 videos from scratch. And for a production team under pressure, that difference is the whole game.
What counts as a “new” video for Meta and TikTok
People ask me this all the time: which elements do the algorithms treat as meaningful when you remix?
Because platforms like Meta and TikTok have duplicate mechanics. If two videos are too similar, the system can treat them as the same video for analytics and credit. Your “new” creative is not actually new in the machine’s eyes.
So when you remix, change something significant for the algorithm, but still cheap enough for production.
What usually works:
a different hook
a rebuilt hook with different wording or a different idea
a different actor or person on camera
What usually does not work on its own:
changing subtitle font
A font swap is not a meaningful remix. The algorithm will likely still treat it as the same video.
Outcome
You take creative revival analysis, combine it with top tag/hook/CTA analysis, send a remix brief to production, and change significant elements. On the way out, you can get roughly 10x more relaunch volume than if you only tried to re-run the original videos.
Take 3. Campaign management mistakes, rare buyers, and the analyst gap
The “obvious” mistakes that still burn money
Maxim made a point I like: work hard to avoid campaign management mistakes.
I have seen a lot of these in the wild. They are not clever failures. They are basic ones that stay invisible for a day or two and then show up as a scary spend spike:
campaigns left running too long, or simply forgotten
wrong budget, or no correct budget set, so the campaign spends without a proper limit
unlimited spend that burns hard for a day or two
bid set to zero
incomplete settings (fields never finished)
These should be controlled. Ideally before the money leaves, not in the post-mortem.
Great media buyers are rare. Encode the one you have.
Finding a strong media buyer is hard. In practice, they are a rare kind.
So the useful idea is not “keep searching the market for another rare hire.” The useful idea is to use automation (and AI) to catch errors and help managers avoid simple, obvious mistakes.
If your team already has a strong manager or media-buying consultant, go further: encode that person’s rules.
Then the rest of the team levels up toward that standard. Not to the exact level of a buyer with ~10 years of experience, but much closer, because they are working with the same decision rules that person uses.
Instead of hunting the market for one rare hire who will work for you and do the job well, you build the operating system in-house.
The marketing analyst shortage, and why gut is not enough
There is a second resource problem I see in almost every company I work with.
Product analysts exist. Sometimes marketing analysts exist too. But marketing analysts are often hired as a second priority. They are scarce. They are hard to find. A lot of companies simply do not prioritize hiring them.
So marketers still have to move. And very often, they move on gut feeling. I hear this constantly: “I just use my gut and optimize from there.”
Gut can be fine. Statistics is better.
When you have a tool that can run analyses with proper methodology and proper confidence, you get something marketers usually lack: analysis you can actually decide on.
That is an ideal use case for teams that do not have enough marketing analyst capacity. You are not replacing judgment. You are giving judgment better inputs.
Final tip: pick one lever this week
If you only ship one change, start with Take 1.
Pull last month’s creative testing spend. Divide by creatives launched. If you are anywhere near the $100 per creative / $100,000 per 1,000 creatives line, cutting test budgets via auction metrics + correlation analysis is the highest-leverage move on the page.
If production is the bottleneck, layer Take 2 next: 10 winners × 10 significant remixes = 100 videos, without asking the team to invent 90 net-new shoots.
Run Take 3 in parallel. Automate the stupid mistakes. Encode your best buyer’s rules. Give marketers analysis when analyst seats are empty.
Three moves. Same theme: stop paying full price for learning you could get cheaper, volume you could remix, and errors you could catch before they spend.
That is the playbook.
Quick checklist
Creative testing
Calculate current spend per creative on US tests (target reference: ~$100)
Multiply by monthly creative volume (example: 1,000 × $100 = $100,000)
Sanity-check against a 1% to 3% win rate ($97k to $99k on non-winners)
Keep tests on main target markets (do not assume other markets behave the same)
Prioritize early reads on CTR, CPC, CPM
Collect communication metrics and auction metrics where conversions are cheaper / signal arrives faster
Run correlation analysis on historical data for a target budget per creative
Aim for ~80%+ correlation before locking the budget rule
Prefer a target budget below the platform 7 to 14 day default
Require ~10 conversions before trusting the final conversion metric
Remix production
Pull burned-out or limited-market winners (not only for same-creative relaunch)
Pull historically best tags, hooks, and CTAs
Brief creative for remixes: hooks, settings, models/actors, AI tweaks, short cut reuses
Change algorithm-significant elements (hook / wording / actor), not just subtitle font
Target 10 source videos × 10 variations = 100 outputs
Campaign hygiene and team leverage
Watch for forgotten campaigns, wrong/missing budgets, unlimited spend, zero bids, incomplete settings
Automate checks for those mistakes
Encode rules from your strongest buyer or consultant (~10 years experience standard)
Give marketers analysis with proper methodology when marketing analyst seats are scarce
Prefer statistics over gut when both are available
