Alper Taner

Don't assume creatives that get less algorithmic spend are bad: they may just be unlucky in the initial 5-10K impressions. When one creative dominates 90% of spend, pause it and give other creatives equal opportunity. The algorithm's early decisions aren't always optimal.

When algorithms overspend your daily budget while maintaining or improving performance, lean into it aggressively. This signals untapped demand and you can often scale 5x in weeks by ignoring conservative 20% daily increase rules and letting performance guide spend.

Maintain logical funnel consistency in event mapping: don't send mixed signals like mapping 'subscription cancellation' to 'add to cart.' Algorithms optimize based on typical user behavior patterns, and illogical event mapping confuses the machine learning, reducing delivery confidence and auction competitiveness.

Event mapping can dramatically impact performance, both ways. Example: in one case, switching from mapping start trial from 'purchase' to 'trial' reduced cost per trial by 35% in one case. When your trial value ($10-15) can't compete in auctions with e-commerce brands bidding $200 for purchases, optimize for trial instead to access different inventory and lower CPMs.

Test extreme budget strategies, but with proper guardrails: set campaign-level budgets to $1M/day (to eliminate pacing constraints) while using campaign spend limits and bid/cost caps for control. This can unlock higher-quality traffic by removing artificial pacing limitations. Only do this if you have a large account, a lot of learnings, and the proper monitoring in place.

You need to act based on the campaign performance but don’t double budgets overnight when you see good campaign efficiency. Increasing budgets too quickly will most likely result in a big loss in efficiency and fluctuations in KPIs. Prefer small increases in budgets, each time keeping an eye on KPIs_ (20% increments per day will still increase your campaign budget very quickly)._

Before scaling ad spend, look at MarTech to ensure you track everything correctly and that your main metrics are measured accurately.

Start scaling “vertically” (more spend on existing channels) until the marginal unit cost is no longer ideal. Then, scale “horizontally” (spending on other channels).

There are scenarios where increasing spend can improve efficiency. The reasons can be:

• It gives more signal for the ad network algorithms to optimize

• Your K factor (virality factor) generates organic uplift

• Your brand awareness increases

SKAN 4.0 will help us shift to longer-term optimization again.

You’re expected to lose some of the efficiency while scaling. It’s about finding the strategic balance. Example: losing 20% of efficiency while gaining 200% of users might be acceptable based on the company’s strategy.

When evaluating new channels, consider the balance between efficiency and scalability. Example: a 10% loss in efficiency might still be better if you can scale much more. Considerations for scaling also include reliability and how efficiency evolves as you scale.

If you scale UA spend too fast, you lose the learning phase. It also makes it harder to understand how the algorithm and creative fatigue work.

Determine your creative testing spend threshold by analyzing when cumulative spend and CPX metrics stabilize. This prevents both overspending on bad creatives (wasting 10K+ when $500 would suffice) and underspending on potentially good ones. Find your account-specific decision point.

The real KPI of a creative testing system is cost per successful creative, not the number of creatives tested.

Use Meta Ad Library strategically: if a competitor has 15+ variations of one concept versus 2-3 of others, it's likely a winner they're iterating on. But treat this as inspiration only because their success doesn't guarantee yours due to different audiences, funnels, and positioning.

When a company is testing a high volume of creatives, the level of detail and quality to shape the next input is sometimes jeopardized.

In creative testing, quality of learning beats quantity of test. Creative strategy is about forming hypotheses and learning what works or not, and why. You still need to test a concept with a few variations!

Merge multiple data sources for creative hypotheses: Meta performance data, in-app analytics showing what high-LTV users engage with most, competitor concepts, trending formats, and category-level inspiration. Don't rely solely on ad platform data: your best creative ideas may come from understanding what retains users.

You must ensure you have a wide variety of creative concepts, from animation to UGC, with different approaches to the problem or solution you’re focusing on. These days, what typically works is broad targeting with creatives aiming at the right personas.

To improve your creative strategy, reverse-engineer the creative’s brief based on performance. Find the similarities amongst the winners and similarities amongst the losers. Slice and dice the opening, the thumbnail, the main character, the USP, the CTA, the duration, the format, etc.

Look into the IPM of your creatives and how efficiency changes over time, including on shorter time frames (e.g., last 7 days instead of just the last 30 days). Don’t let your creatives die; be ready to replace them with your next best creatives.

The learning phase is very important, for all platforms. Don’t upload a new creative into the ad set every week. This will lead to fluctuations. Instead, make changes every two weeks and bulk upload the test winners.

Match your measurement tool to your question: use MMP for creative comparison, but use incrementality testing (geo-lift, holdouts) for channel comparison. Example: Google UAC's search intent makes it appear cheaper in last-click attribution, but that doesn't reflect true incrementality.

Take the time to test your hypotheses properly (with adequate spend, duration, and variations) because premature conclusions waste more resources than thorough testing.

Don't label metrics as 'problematic' until you've systematically tested and iterated on solutions to improve them. Focus on understanding 'why' before declaring something broken. _Example: a low conversion rate is just a fact until you've diagnosed the root cause, created an action plan, and measured results. _

Always look at both short-term and long-term KPIs in decision making and budget allocation. Something that looks very good in the first place might be hurting the business in the long term. Example: giving too many vouchers, with users ordering only with vouchers.

The first step to understanding customers is to establish a robust tracking setup: right event, right event parameters, user properties that give you a comprehensive picture of users’ behavior. Then, analyze the latest actions and common behavior patterns of churners vs. those of high-LTV users. What are the key differences?