Media buying nuggets from six operators on X

$18M on Meta, wrong CPA maths, and a shipping-fee test

ScaleSurfer pinned his last-30-day spend screenshot ahead of Q4 and spent the day making fun of advice from platform reps. Herrmann caught Meta reporting the wrong CPA. Nate shared a shipping-fee result, and Cody asked for better product photos.

• Channel mix
  @scalesurfer_At about $21.6M a month across channels, Meta is still 83% of his paid spend. Google is a distant second, and AppLovin is a small test line. post ↗
• Bidding
  @scalesurfer_A Meta rep said a campaign that spent $3K at 0.5 ROAS was fine "because of target ROAS" and suggested moving it to Highest Volume at $2K/day "to give it room to learn". He says the more he spends, the less he trusts best practices or the account score. post ↗
• Meta issues
  @herrmanndigitalAds Manager's ad set "cost per result" has been wrong almost every day. It showed $103.01 when the real CPA that day was $86.97, and $103.01 didn't match the last 7 days either. Check Meta's CPA against your own spend ÷ purchases before acting on it. post ↗
• Offer
  @natelagosFor products priced above $150, he's always seen a shipping fee under $10 have no effect on conversion rate. Worth split testing before you give shipping away. reply ↗
• PDP images
  @codyplofHe quoted a take that you'll lift conversion rate more by spending 3x the time on product images than on product page copy, because people barely read product pages, and asked for a new Winks shoot. Nate argues the opposite later in the month, so test both. post ↗
• Rebrand
  @ecomfrrRenaming his store from Alevia to Auniva over a trademark issue didn't raise CPMs. reply ↗

An agent uploads the ads, and a new account pays its "tax"

• AI ops
  @codyplofHe wasted about 2 hours uploading ads in a slow Ads Manager, so now he gives instructions to an agent (runneth). It analyses every ad with the Gemini API, picks the landing page, headline and copy, and sets the ad up paused for him to review. It's connected to Trybe for creator content and can add captions. (Runneth sponsors Marketing Operators, the show he co-hosts; he says this post wasn't paid.) post ↗
• New account
  @codyplofOn Winks' new ad account, CPMs are finally falling as they "pay down the tax". Next step: let lots of UGC go live this week, then partnership ads, to keep pushing CPMs down. Asked later how much the new-account tax cost, he said $18K. post ↗ reply ↗

Bottoms-up creative testing, and Meta's new reporting beta

• Creative
  @natelagosNate's "Bottoms Up" method: the reasons people buy have little to do with how you first got their attention. So test concepts as close to add-to-cart as possible. Start with product page headlines and images, move the winners to high-traffic landing pages, pre-sell pages and the homepage, then turn them into ad briefs. He thinks this beats dropping 100 new ads a week and hoping one scales. He learns from a winning product page headline rather than pasting it into top-of-funnel ads. post ↗ reply ↗
• Meta issues
  @herrmanndigitalMeta's new Ads Manager reporting beta contradicts itself. It says "you increased spend, so CPA climbed", then recommends "increase spend to get more volume". His view: Meta won't tell you to pull back or that you're wasting money, so experienced buyers are still needed to judge results. post ↗
• Ad research
  @codyplofHe used the Astra agent to build himself a social feed made only of ads, filtered by customer persona and pulled from several data sources, with boards to save the good ones. He uses it as a swipe file. post ↗

Herrmann's whole BFCM plan, and how Nate sets up a new account

• BFCM
  @herrmanndigitalMeta: keep the ads you've run all year switched on, and add a second campaign on lifetime budgets just for the sale ads. AppLovin: take your top creatives, add a discount badge, update the end cards with the offer and the URL, and leave everything else on. Pinterest: launch pre-seeding campaigns now to prime gift buyers in your niche, aiming to break even, then retarget heavily all through BFCM. "There, saved you another webinar." post ↗
• New account
  @natelagosHow he launched his newest ad account: one CBO campaign, ad sets split by product or audience, and a minimum spend on each ad set so every one gets some budget. It "went pretty good". reply ↗
• LP testing
  @natelagosIf the landing pages target the same persona, split test them on-site with Intelligems. Only test pages through the ad account when each page targets a different person (he counts a different awareness stage as a different persona). The end goal is variety: different ad sets sending traffic to different landing pages. reply ↗ reply ↗
• Winks CPA
  @codyplofSean Frank teased that the Winks offer would never "crack a $300 CPA". Cody replied with an Ads Manager screenshot showing $30.31 cost per purchase at 2.23 purchase ROAS, and later called it "just a good day". The offer in the thread: a 28-night supply for $48 (was $65) on subscribe-and-save, renewing every 4 weeks, with free shipping and 4 free gifts. post ↗

AI models put a buyer off an apparel brand

• Creative
  @herrmanndigitalAs a shopper, AI-generated models put him off an apparel brand completely, and he sees many clothing brands doing it. If you sell clothes, show real people wearing them. post ↗

Creator ads: 2 or 3 out of 10 work

• Creators
  @codyplof@JoshJDurham said the best creator ad programs he's run all test 10 new creators a month, double down on winners and cut losers. Asked about the hit rate, Cody said 2 or 3 out of 10. Budget for 10 creators to find 2 or 3 good ones. reply ↗

"$15K a day is a quick test run"

Manny Barbas posted that he'd taken an account from $2K to $15K+/day on a cost-cap CBO. Jordan Menard, Jem Bourouh and ScaleSurfer piled on that $15K/day isn't scale.

• Bidding
  @scalesurfer_Proof from someone else's account: a cost-cap CBO went from $2K to $15K+/day. The pushback was only about whether that counts as scale. To ScaleSurfer, $15K is test-run money. thread ↗
• Organic
  @codyplofHe offered to pay for anyone to do Cut 30 (an organic creator course) if they'd film Winks content. No one committed, so he enrolled himself. He's never been on camera, and expects becoming a creator to make him a better creative strategist. post ↗

ScaleSurfer's AMA: 15 answers on scaling, structure, team and product

ScaleSurfer took follower questions for about six hours. This was the most useful single thread of the month. The last answer came in at 00:55 UTC on Sep 16 and is included here. AMA post ↗

• Scaling & bidding
• Scaling
  @scalesurfer_If a new creative batch is profitable on day 1 and CPM and CPC also look good, double the budget straight away. If they don't, hold 1–2 days and then double. He called 20% every 2 days far too timid. reply ↗
• Bidding
  @scalesurfer_You can scale without caps, but only with killer creatives and a really, really good media buyer. tCPA and tROAS both work. reply ↗ reply ↗
• Campaign structure
• Structure
  @scalesurfer_MOF/BOF ads run in their own campaign, purely for visibility in reporting. Never add MOF/BOF statics to scaling campaigns: they eat the spend and make the campaign less incremental. If a bid cap scaling campaign keeps sending budget to BOF ads, take them out. reply ↗ reply ↗
• Structure
  @scalesurfer_He dismissed interest testing as outdated. That was his reply to a follower's setup, which used this bid rule: start at target CPA + $4, then test five steps above it (for a $50 target, try $54–$59). quote ↗
• Markets
  @scalesurfer_Use one pixel for every market when you expand so your learnings and pixel data carry over. For his campaign structure and kill rules, he pointed to his April post (see Reference below). reply ↗
• Accounts, creative & team
• Accounts
  @scalesurfer_Only in-house ad accounts, never agency ones. There's a ban wave going on, and if your agency account provider goes down, so does your business. reply ↗
• Team
  @scalesurfer_The creative team works in pods of 1 strategist + 2–3 editors. Skip UGC agencies and do creator outreach in-house. Real UGC, "yapper" UGC, animated and fully AI ads can all scale right now. reply ↗ reply ↗ reply ↗
• Product & business
• Plateaus
  @scalesurfer_A follower had dropped from $20K to $10K revenue days while a competitor's traffic surged. His answer: a product that has done $20K days is proven. It's a creative problem, so raise creative volume, find new angles or a better unique mechanism. reply ↗
• Channels
  @scalesurfer_Start other ad channels once Meta spend reaches $500K–$750K a month. Supply chain stays in-house. At scale, forecasting and cash flow are a full-time job, and his big jumps meant risky purchase orders for millions of units. reply ↗ reply ↗
• Elsewhere
• Freight
  @herrmanndigitalDiesel is $8 in California and the LA ports are among the biggest in the US. Freight costs will rise if you haven't locked them in before Q4. post ↗
• Kill rules
  @herrmanndigital"Sometimes you just need to turn bad ads off." @jforjacob replied that it's amazing this counts as controversial on DTC Twitter. post ↗

Don't send sale emails before the first order arrives

• Retention
  @herrmanndigitalA red flag that loses him as a customer: brands sending sale emails before his first package has even arrived. Pause promotional flows for new buyers until after delivery. post ↗

AI agent designs in Figma: good start, junior quality

• AI design
  @codyplofThe Astra agent working in Figma (via its MCP connector) produced product page images and an email from a 2-minute prompt in one attempt. He says it looks like a junior designer's work, and getting it good needs skills files and a design system. He was clear he wouldn't ship it. post ↗
• Angle
  @codyplofWinks' main sleep angle: no melatonin. He says melatonin is habit-forming and makes sleep worse over time because your body produces less of it. Winks uses 10 other ingredients instead (L-theanine, ashwagandha, saffron and more). reply ↗

Popups don't hurt conversion rate, and 200 ROAS on 28p means nothing

• Popup test
  @codyplofA split test he weighed in on: popup vs no popup across 38K visits. The popup collected 2,140 SMS and 1,321 email sign-ups, and site conversion rate was 2.49% with it vs 2.43% without. Cody has seen the same, and says it's worth running for the list even if conversion is flat. reply ↗
• Reading data
  @scalesurfer_Someone posted a 200 ROAS screenshot and asked how to raise ROAS. ScaleSurfer's answer: stop showing off 200 ROAS on £0.28 of spend. Results on tiny spend mean nothing. reply ↗

Headlines phrased as questions, and the autosub "black hat" argument starts

ecomfrr called the store tryresia.com heavily black hat (83K views). Kes (@kes11av) said the current dropshipping meta is "50 shades of Resilia".

• Headlines
  @natelagosSarah Levinger shared a "micro-yes" headline that turns the claim into a question: "Are you finally ready to stop [symptom] at the source?" Nate says he's tested the same idea across a dozen brands in the past 2 months with "huge wins everywhere". reply ↗
• Market intel
  @ecomfrrMost dropshipping stores right now copy the Resilia model, which is built on auto-subscription offers. quote ↗
• Compliance
  @ecomfrrDefinitions in use, from a reply in his thread. White hat: clear subscribe-and-save options on the product page. Grey hat: a subscription-only option the product page doesn't explain. Black hat: the subscription is hidden on both the product page and the checkout. post ↗

ecomfrr defends Auniva's autosub; ScaleSurfer's Q4 research tool

LB (@realmodageek) said ecomfrr runs the same autosub he'd criticised. ecomfrr posted Auniva's 4.9/5 Trustpilot score, and replies pointed out that reviews can be bought.

• Compliance
  @ecomfrrAuniva's disclosure, as he describes it: terms on the product page, clear wording at checkout, and a subscription confirmation email after purchase. The critics' answer was that customers can know about the subscription and still not want the second charge. reply ↗
• Q4 research
  @scalesurfer_He uses TrendTrack's "Q4 Mode" to see competitors' past BFCM ads and what they're testing this week. Pull last year's BFCM ads now, before November. He posted it as a recommendation, so treat it as a plug. post ↗

Meta outage, and a structure for launching hundreds of creator ads

• Meta issues
  @herrmanndigital"Meta ad performance has been pretty awful lately. If you're wondering whether it's just you, it's not." On the outage: the time Meta says an issue "started" is when it was detected, not when it began, and expect 2–3 more days of uncertain performance while things recover. post ↗ post ↗ reply ↗
• Creator ads
  @herrmanndigital@zachlduncan's setup for launching hundreds of Trybe creator ads a month: a cost-cap CBO with one ad set per batch of 8 ads, and any creator with 4+ ads gets their own ad set. He says it spends evenly. Herrmann: "literally my exact same set-up". reply ↗
• AI ops
  @codyplofMeta's Muse is "lightning fast", Ads Manager isn't, and he now uses agents to upload ads. reply ↗
• Hiring
  @natelagosHe's hiring someone to run CRO tests (Intelligems and Shopify experience) at over $100K a year while he keeps doing all the writing. Writing copy and running tests are being split into separate jobs. post ↗

After the outage: budgets gone in an hour, lift studies wiped

• Meta issues
  @herrmanndigitalSince the outage, new ad sets are spending 80% of their budget in the first hour after launch. Set conservative budgets on new launches and watch the first hour. post ↗
• Lift studies
  @herrmanndigitalThe outage seems to have cleared all of his Conversion Lift Studies. Check that yours are still running. post ↗
• AI buying
  @herrmanndigital"Muse is going to be your media buying agent. Probably faster than you think." post ↗
• Subscriptions
  @codyplofAsked to choose between Skio, Recharge and StayAI for subscriptions, he picked Skio. reply ↗

Lift studies on autopilot

• Measurement
  @herrmanndigitalHe had Muse go into Meta and create a Conversion Lift Study for every month through the end of 2027, then built a system that reminds it to keep creating them. It used to be one of his least favourite tasks. (77K views.) post ↗
• AI tools
  @scalesurfer_Someone promoted an "AI agent runs your Google Ads" product. Jem Bourouh pointed out the demo account spends $8.5K a month, not a day, and ScaleSurfer laughed along. thread ↗

A non-sale headline beat "70% off"

• Headlines
  @natelagosOn a landing page, an evergreen headline with no promo is outperforming one that shouted "70% off". You don't always need a bigger discount to win. post ↗
• AI avatars
  @codyplofHe's making photorealistic AI avatar videos with Seedance 2.5 through Higgsfield. A reply warned that California's SB 1050 takes effect Jan 1; Cody's answer was "just disclose it's AI". Plan to disclose AI people in ads before January. thread ↗
• Meta issues
  @herrmanndigital"Yesterday Meta had one of its best days all month, so naturally today is the complete opposite." Judge performance over several days, not one. post ↗

Which video edits AI can do now

• AI editing
  @codyplofTalking heads are easy. "Yapper" and basic paid-social edits are easy. VSLs with Higgsfield b-roll are doable if you train the skills for it. He uses DaVinci Resolve for basic edits. reply ↗
• Headcount
  @codyplofIt would cost him thousands to have people do what runneth does. They're about to roll it out at JRB and cut headcount. Pricing is custom and based on ad spend. (Runneth sponsors Marketing Operators.) reply ↗

Hiring posts, and why supplement "expertise" stopped working

• AI UGC
  @scalesurfer_He's hiring AI video editors for a $100M+/yr brand. They need proof that their AI-made ads drove $50K–$100K+ in spend, and the edits must be direct-response, not just polished. post ↗
• Comp model
  @scalesurfer_Creative strategists at a 9-figure brand get base salary plus a % of ad spend. They must be full-time and exclusive, with $1M+/month brand experience and ads that spent $100K+. post ↗
• Supplements
  @natelagosSupplement brands used to win on expertise: doctor founders and clinical studies. Now any marketer can claim the same ingredients and cite the same studies, so customers can't tell brands apart. What's working for his clients instead is showing customers you understand exactly how they've been let down, the way political campaigns do. post ↗
• Offer naming
  @codyplofAsked what to call "free wallet with purchase of any full-priced bag": "GWP is the only answer. It's not even a debate." reply ↗

There's no CPM lever

• Auction
  @codyplofAfter visiting Meta: "Went to Meta today to look for the CPM lever and couldn't find it. I guess it's expected action rate after all." You can't adjust CPM directly. It follows how likely Meta thinks people are to act on your ad, so better creative is how you lower it. post ↗
• Angles
  @natelagos"The political marketing strategy": name the customer's specific problem, then name a common enemy that caused it. Use one enemy per angle or persona. Sleep supplements can blame blue light or high-pressure jobs, nutrition brands can blame the processed modern diet, and luxury brands can blame fast fashion. Then present your product as the fix. post ↗
• Homepage
  @natelagosHe posted a big test win and said all it took was a new H1 headline on the homepage. reply ↗
• CRO audit
  @zachmstuckHis notes on 4patriots.com: the homepage hero is so big the CTA sits below the fold. On mobile, add "Best Sellers" for first-time visitors, a "Reviews" page for social proof, and customer care to the main nav. reply ↗

Gift-season copy rules, and fewer audience exclusions

• BFCM
  @natelagosHis gift-season rules for ads, site copy and creator briefs:
  • The buyer is now the gift giver, most often the spouse, so cast the opposite gender to usual.
  • Sell a token of love, not features, and reassure the giver they're choosing well (social proof helps).
  • The pricier the product, the earlier you start: run gifting ads in October, before the sale.
  • Gift buyers may be your lowest-LTV cohort, so push bundles and "ultimate gift sets" to raise AOV.
  • From Nov 1 to Dec 19 they'll be flooded with 60%-off ads, so your hook has to do more than the discount. post ↗
• Results
  @natelagosOn a client he's worked with since April, from stacking winning copy tests: sitewide conversion rate up 61% year to date, CAC a third of last year's, and profit up 3x. post ↗
• Exclusions
  @herrmanndigital"90% of you probably don't need all those audience exclusions you're adding to your Meta ads to reach new customers." Try opening targeting up. post ↗
• Personas
  @codyplofHis favourite software right now is Outersignal, the best persona-building tool he's tried, with "huge implications across the funnel". reply ↗

Start gift-buyer ads now and break things before Q4

• BFCM
  @natelagosThe brands that make the most over BFCM are already running gift-giving creative aimed at the gift buyer, sending traffic to a landing page that speaks to gift buyers. Over 70% of buyers are gift buyers for every brand he's worked on, though you still need a funnel for people buying for themselves. This is gift messaging, not sale messaging yet. post ↗ reply ↗ reply ↗
• Copy
  @natelagosNo "spooktacular sale" or "scary savings" this month unless you sell Halloween products for kids. Write something better. post ↗
• Testing
  @herrmanndigitalThe end of Q3 into early Q4 is the time to break things and test anything you assume has to stay the same. He showed an early Intelligems test up about 50% (roughly +70 orders a month, not yet significant). His point was to make changes fast this month. post ↗ reply ↗

Narrow the product page to one audience, and how many creatives a week

• PDP
  @natelagosDon't market to your whole target audience on the product page. Speak to the subset you can really move: sell testosterone supplements to dads, not men; watches to rugged guys who like watches; probiotics to women who've tried everything. He says this test "absolutely crushed". The post he quoted suggests duplicating the product into a separate product page per angle instead of sending pre-sell traffic to one generic page. post ↗
• Creative output
  @zachmstuckHe asked how many concepts or scripts per week a strategist should produce, how many statics per designer, and how many videos per editor. One reply: set the number by how many you can properly test per week, and with AI that bar is high. Zach's point was that you still need a realistic weekly expectation that doesn't sacrifice quality. thread ↗
• Collections
  @zachmstuckA customer couldn't tell his "performance" and "trade" socks apart. His fix: collection pages need to explain the differences between product lines. reply ↗

Meta's weekend budget flex, and why one CBO isn't enough

• Budgets
  @herrmanndigitalMeta's "flexible" budgets raise spend 50–60% on weekend days whatever the brand, and performance rarely holds up. They now cut spend 30–40% going into weekends. He wants rules that let advertisers choose which days Meta can flex. post ↗
• Structure
  @scalesurfer_Agreeing with Brandon Fink that media buying still matters: creative is a huge lever, but launching everything into one CBO and leaving it isn't a strategy. quote ↗

A quick product page test, and the incremental attribution argument

• PDP
  @natelagosAn easy test for this week: below the buy box, replace "Specs & Details", "Features" or "How it's made" with a persuasive headline. For a premium version of a usually cheap product he's testing "[Material] that doesn't suck", "Finally, a [product] that won't quit on you" and "We fixed your [product]". post ↗
• Attribution
  @herrmanndigital"You all zigged too hard on incremental attribution and nuked your audience. Just come back." Barry Hott pushed back, and Herrmann said it was sarcasm. On Oct 6 he posted a summary of his actual view (see Oct 6). post ↗
• Meta issues
  @herrmanndigitalMeta's duplicate button was broken all day, about 50 days before BFCM. Have a backup way to build campaigns ready. post ↗
• Reporting
  @herrmanndigitalPhil Kiel wants Meta to offer audience segment reporting through its API. Herrmann thinks it may already be in Advanced Analytics and is checking with Meta. reply ↗

Auniva's ads go dark, Herrmann's IA playbook, and testing outside the ad account

• Platform risk
  @ecomfrrAuniva's ads are off because Shopify is running a compliance audit on the store. Some ads are "on the grey/black hat side", so they turned them off to give the store its best chance of passing. Cody's reply: "I hope this is satire. If not this is a wild thing to admit." post ↗ Cody ↗
• IA playbook
  @herrmanndigitalHe reposted a summary of his own past posts on incremental attribution. IA is a lever for some accounts, not a default. It suits brands with big retention audiences, strong social followings, product drops or long consideration cycles. Start with one campaign at about 20% of spend, new ads only and no creative overlap, then ramp to 50–60% and later about 95%. Give it 21–28 days, or 30+ if a lot of the volume comes from views. Keep standard attribution for remarketing to existing customers. One 8-week test cut cost per incremental acquisition by about $27. Using IA everywhere shrinks the audience Meta will buy. post ↗
• Testing
  @natelagosThe ad account isn't your only place to test creative. Test on the website, organic, email and SMS. "Top of funnel success comes from bottom of funnel learnings." He's also seeing wins from making the whole site more persuasive, whatever the visitor's awareness level. post ↗ post ↗ reply ↗
• AI uploads
  @herrmanndigitalFil asked why buyers aren't uploading ads with AI through Meta's API ("under a minute from thought to live"). Herrmann can't for publicly traded clients: getting approval isn't worth it, so third-party tools are only for his other brands. reply ↗
• Legal
  @scalesurfer_Alex Beller (Postscript) says California's new SB 690 shuts down the wiretap suits brands got for using the Meta pixel, which came with $10K–$25K demand letters. ScaleSurfer: "next up; fake discount class action lawsuits". Make sure your compare-at prices are real. reply ↗ Beller ↗
• Meta issues
  @scalesurfer_"There's always people complaining about Meta performance even on days where some brands are doing their best days." reply ↗
• Attribution
  @codyplof"I'll always remember where I was the day of: 9/11, Covid started, the IA vs 7dc1dev wars broke out." A sign of how loud the attribution argument got this week. post ↗