Most of the ads you launch are going to fail. That is just the ugly truth about media buying right now. We get emotionally invested in a specific creative, convinced it's going to absolutely crush, and the market just completely rejects it.

Last year, we dropped roughly $280,000 on ads. That spend helped us generate over $2 million in revenue, sitting at a blended ROAS of around 7.5. Almost everything learned during that scale came from screwing things up completely first. The platforms are getting harder to predict, and lead quality is a massive bottleneck for almost everyone.

Here is the raw reality of what's working and what is completely broken right now, based on launching nearly 900 individual creatives.

1. Volume is the only answer to a low hit rate
Out of those 900 ads launched, the vast majority were absolute trash. Our average hit rate-meaning the ads that actually see sustainable success-is maybe 20% on a good day. Two out of ten succeed. Even as someone who does this professionally, beating those odds is nearly impossible. You just have to build a machine that creates enough volume to test heavily and find the outliers. Stop crying over a failed ad and just test the next one.

2. Broad targeting is king
Meta's targeting algorithm is terrifyingly smart now, especially after the Andromeda update late last year. We waste so much time agonizing over lookalikes versus interest stacking. Stop overcomplicating things. Broad targeting (or Advantage+) completely removes the nightmare of decision fatigue. Pick one broad audience and focus all your mental energy on making better creatives. Let the machine do the heavy lifting of finding the buyers.

3. Stop feeding your pixel garbage leads
This was the biggest unlock by far. Whatever hits your 'Results' column dictates who the platform goes after next. If you let unqualified funnel-hackers, fake emails, and broke leads fire your pixel, the algorithm just goes out and finds more of them. We implemented a manual review pipeline in GoHighLevel to stop this. When a lead books a call, a human (or a VA) quickly checks if they are actually qualified. If they pass the check, we drag them to a pipeline stage that manually fires the pixel and tags the contact.

Why the tag? Because if that same lead reschedules three times, you don't want the pixel firing three times and ruining your campaign data. Guard your pixel with your life. Someone in the thread asked if this manual feedback loop can be done with Hyros and Close CRM. Absolutely. You just need a webhook to delay the conversion event until the lead is actively verified.

4. Blended ROAS > In-platform metrics
In-platform attribution is an absolute joke right now. Even click-based tracking tools miss the big picture. We routinely had ads showing a mediocre 4.2 ROAS on the ad level, but our actual blended ROAS across the business was north of 7.5. People see an ad, don't click, but look you up later on Google, binge your Instagram, and then convert. Ad platforms are heavily leaning into ad sequencing-showing users a specific mix of different creatives before they finally bite. If your top-line revenue and booked calls are up, stop panicking and killing ads just because the direct attribution looks weak. Zoom the hell out.

5. Launch weekly, or die
Having one winning hero ad is terrifying. The day it fatigues, your revenue tanks overnight. This happened to us early on, one ad carried the whole account for six months, and when it died, we were in full panic mode trying to replace it. Now, we force a weekly launch cycle. Block out dedicated days for ad research, scripting, filming, and launching. If you are spending over 500 bucks a day, you cannot afford to skip a week of creative testing.

6. Ruthlessly consolidate your campaigns
The days of running 15 different campaigns simultaneously are dead. Simplification is the entire game now. For accounts spending under $500 a day, one campaign, one ad set, and 10 to 20 ads inside it is all you need. All those different creatives might hit different pain points, but putting them in one single ad set consolidates the algorithmic learning phase. Keep your footprint tiny and let the budget flow to the winners.

7. Blank pages are for amateurs
Writing ad scripts from scratch takes way too much time. Use AI, but don't let it write generic marketing garbage. Pulling winning ad transcripts from the Ad Library is step one. Feeding them into an AI along with your own VSL transcript and website copy gives the prompt deep context. Then, just ask the AI to map your specific offer onto the proven structure of the winning ad. You never copy the original word-for-word, but it gives you the pacing and hook framework instantly.

8. Treat filming like a muscle
Batching an entire month of ads into one miserable, eight-hour shoot does not work. Your energy drops, your delivery sounds completely robotic, and the ads flop. Pick one day a week, every single week, to film. Delivery on camera is a skill you have to practice consistently to stay sharp.

9. Carousels act as a lead filter
Static image ads will always get you the cheapest leads, but the quality is usually terrible because there is zero friction. Video ads are great, but not everyone wants to be on camera. Carousels are the ultimate middle ground for lead generation. You force the user to swipe through 5 to 10 slides of education or pain-agitation before they ever see a call to action. The cost per click is way higher, but the lead quality is night and day because you've pre-qualified their intent before they leave the platform.

10. Scale the angle, not just the ad
When an ad finally hits, amateur buyers immediately try to test 20 completely new concepts. Don't do that. When you find a winning angle, ruthlessly extract every drop of juice from it. Remake that exact same script as a talking-head video. Turn it into a B-roll format. Make a static graphic out of it. Turn it into a Twitter-style screenshot ad. Format diversity is how you scale a winning concept without burning your budget on unproven psychology.

It's a brutal ecosystem right now. But if you stop fighting the platform's automation and start fiercely protecting the data you feed it, scaling is still very possible.