Paul Koshlap

Claude is amazing for Meta Ads analysis…with a few exceptions that could crash your account. For example, point Claude at your ad account and ask for your three best performing ads. You’ll get a stunningly detailed breakdown of a wide range of platform metrics with an emphasis on efficient Cost per Result. Then take a look at spend levels. There’s a good chance Claude picked 1-2 ads that barely spent anything. Why? It privileged CPR and picked ads that looked efficient simply because they hadn’t scaled. It didn’t understand that spend level is the single most important KPI in your account. You see, the Meta algorithm isn’t sitting on its hands letting spend flow randomly to ads. It’s using a mountain of historical inputs, combined with real-time data from your campaigns, to identify which ads have the highest probability of maximizing your goal at scale. And the more aggressively it spends on these ads, the greater the impact of diminishing returns, causing CPRs to rise (google “The Breakdown Effect” for more on this). But the CPR isn’t rising because Meta doesn’t know what it's doing. It's rising because it does. Turn off your highest-CPR ad and you might be turning off the ad scaling your whole account. And this is exactly the kind of nuance Claude can miss. The good news? You can fix this with a simple system prompt update. Here’s a quick hack: copy and paste this LinkedIn post into your project’s chat, have Claude pressure test with online research, and then ask for an updated system prompt that includes a weighted KPI hierarchy that gives spend the proper credit.

@ecomchasedimond Bonuses and bribes is a great band name.

Most ecommerce brands say they optimize their flows. They do. They test subject lines. And that's usually it. The real gains are structural. For example, take your welcome flow and move the social proof email ahead of the discount email. When a subscriber sees proof before they see a price cut, the discount feels like a bonus instead of a bribe. Or split your abandoned cart flow by cart value. A $30 impulse buy and a $200 considered purchase shouldn't get the same urgency sequence. The $30 cart needs a fast nudge. The $200 cart needs reassurance, reviews, and a reason to trust you. Post-purchase has the same problem. Most brands run the same sequence for every buyer. But a first-time customer needs product education and trust before you ask for another purchase. A repeat buyer already trusts you. They just want to see what's new. Same flow, but two completely different jobs. And the best part is that these changes compound on every send.

@garyvee Also, people (and companies) chase near-term money over chasing becoming more valuable. The latter is not only more satisfying, but a better long-term financial strategy.

People choose money over happiness and that is why so many people are losing.

Want timeless, future-proof skills? Don’t acquire new knowledge. Deepen your understanding of what you think you already know.

Want to learn marketing? Learn persuasion and stats. That gets you 80% of what matters.

@TravisHyer @karpathy 😂

@karpathy Bro discovered talk therapy from first principles

One pattern I find useful for working with LLMs is a nice long ramble session. Sometimes the LLM needs more bits to understand what you're trying to achieve, but you're too lazy to type them. In these cases I like to lean back, switch to /voice and just ramble for like 10 minutes, total mess, anything goes, full stream of consciousness. Sometimes I declare it up top, something like "switching to speech recognition sorry for any typos...". Sometimes I turn it into a small interview of a few turns. But I find that the LLMs are somehow very good at reconstructing long incoherent rambles and often their echo of your own tangle of thoughts comes out quite a bit cleaner than what you started with. The result is that you improve the mind meld and have to correct things less from that point on.

Ramble my thoughts with voice chat and scan the LLM‘s text response is the most efficient and effective back-and-forth communication style I’ve found. I also like to finish by clarifying that I want a short response and then I can just continue prompting it as needed if I need more. Love working this way.

The 5 questions that can manage any marketing team: 1) What does success look like? Everything else is downstream of this. If the team can't answer it in one sentence, stop here. 2) How are we measuring it? Two parts: the number that tells us we won, and the signals that tell us if we're winning. Won is the final score. Winning is knowing you're up or down mid-game — which requires metrics that actually PREDICT the outcome. Most weekly reporting never asks that question. It reports what the platforms surface and hopes. 3) What's the plan? What are we actually doing to get there. Not the deck version — what's happening this week. 4) What's the argument for this plan over the alternatives? Demanding an argument forces a logic gut-check. If the reasoning can't be articulated to someone else, you don't have a plan. You have a habit. 5) What new information do we have? Questions 1-4 gave you a strategy. This one keeps it honest. New data either confirms the plan or updates it — and a strategy that never updates isn't discipline, it's neglect. Ask the first four and you'll build the strategy. Keep asking the fifth and you'll always have one worth following.

@nicktheriot_ Love it! It’s more down funnel, but authentic scarcity can be very effective.

My go-to primary text angles: 1. Offer-focused. 2. Product USP. 3. Testimonial. 4. Authority. 5. Solution-aware. 6. Problem-aware. 7. Unaware (native story). Run with it.

@benradack Fantastic! Given these ad types tend to get delivered higher in the funnel, what KPIs do you look at to pick a winner here? Or do you have the budget/time to get sufficient conversion volume to determine winner?

This week I'm going to test whether "Whitelisting" is better than "Partnership" ads - Here's how: → ABO campaign → 1 ad set per test → Each ad set has the same exact ad The only moving variable is how the ad is served: → Ad set 1: Whitelisted through the creator's page (only their name at the top) → Ad set 2: Partnership ad (creator's name + brand name at the top) Then using the A/B test tool & audience/purchase exclusions, I'll prevent any audience overlap This should give me the answer we've been looking for

Best marketing tool of all time? The scientific method. Why? It’s the only way to tell what's actually true from what just looks true. And in marketing, that's the difference between real growth and lighting money on fire. Marketing’s most expensive mistakes aren't people believing obvious nonsense. They're people believing real platform data that’s reasonable, well-dressed, and dead wrong. The campaign that "drove" sales when it was really just standing next to them. The channel that looks cheap because it's quietly taking credit for demand you already paid to create. The “winning” ad with the minuscule sample size. None of those look like errors. They look like results. That's exactly what makes them dangerous. The scientist's toolkit is a set of defenses against being convinced by things that aren't real: A control group, so you can see causation, not just correlation. Statistical significance, so you don't bet the budget on noise. Causal inference when running an experiment isn’t an option. Don’t get me wrong, that platform reporting — that's data. Real, useful data. But it's missing the one thing that turns data into proof: isolating the variable. Hold everything else still, ask "what did this specific thing actually cause," and you're measuring impact. Skip that, and you're just measuring what happened nearby and hoping it's the same thing. That's the gap the scientific method closes; it’s the gap incrementality testing exists to fill. The marketers who solve this problem aren't smarter. They've just trained themselves to ask one question before they spend, scale, or celebrate: How do I know this is real, and not just convincing? Learn to answer that, and you stop making the most expensive mistakes in marketing.

@weaverobotics @grok how much does this cost?

Today, we’re launching our home robot Isaac 1. Isaac 1 deliveries will begin this fall. Order yours below.

@herrmanndigital And such a scalable format! Get a solid group of talented creators who understand the key beats and you’ve got a heck of a content machine.

The Meta ad accounts that have more than 20-30% of the entire ad spend dedicated to partnerships and or whitelisted pages are looking really good overall. Business level revenue is up and roughly slightly reduced CPAs YoY. That's a big win. This continues to be the biggest outlier to success on Meta. These also aren't AI creators btw. These are creators with their own dedicated followings. I believe their social graph is helping unlock these new audiences keeping things fresh. Making sure you update them frequently by getting them on some type of retainer. Long real creators

AI is augmented reality for cognition.

@BillyBroas And hear you on obvious points of agreement. Do you know @fbinegotiator? Reminds me of what he calls “labeling”. In negotiation, it refers to reiterating—and establishing a label for—something your counterpart said. This “repetition” increases trust by demonstrating empathy.

@PaulKoshlap You nailed it, Paul. You want to identify—and establish—the points of an agreement and start the conversation from there. I added establish because we've found that it's a safe bet to cover even the "obvious" points of agreement in marketing copy. It can be done quickly.

Hey @BillyBroas! QQ: I have my list of required beliefs, and my list of current customer beliefs. To begin belief building, I want to identify existing points of agreement and start the conversation there - correct? i.e. by calling out a belief that is in both lists.

@BillyBroas Fantastic! Thanks so much, Billy!

Brands need to tell a story. But, they aren’t the hero of the story. The potential customer is the hero of the story. The customer is Luke Skywalker. The brand is Yoda.

4 mins crash course on branding: