Here's a thought experiment. Ten Meta Ads agencies pitch your DTC brand. They all have case studies. They all promise ROAS improvements. They all have smart media buyers who "know the platform."
Now imagine you could see inside their operations. What would you find?
The same tools. The same workflows. The same dashboards. The same creative templates. The same attribution platforms telling them the same half-truths about what's working.
If ten agencies use identical toolsets, how different can their results really be?
The Standard Agency Tech Stack
Walk into almost any performance marketing agency managing Meta Ads in 2026 and you'll find some version of this:
Campaign Management: Meta Ads Manager (the native platform). Maybe a bulk editor. Maybe Revealbot or Smartly.io for automation rules.
Creative Production: Canva for static images. AdCreative.ai or Pencil for "AI-generated" variations. A freelance designer on Fiverr or a small in-house team recycling the same templates.
Attribution: Triple Whale, Northbeam, or Hyros. Possibly still just trusting Meta's in-platform numbers.
Reporting: A Google Sheets template or a dashboard tool like Supermetrics, DashThis, or AgencyAnalytics. Screenshots of Ads Manager pasted into a slide deck.
Landing Pages: Unbounce, Instapage, or Shogun. Maybe just sending traffic directly to the product page.
Copy: ChatGPT with a prompt template, or a copywriter who writes the same PAS (Problem-Agitate-Solve) formula for every client.
This stack isn't bad, exactly. These are competent tools. The problem is that they're commodity tools. Every agency has access to the same ones, at the same price, with the same capabilities. When your toolset is identical to your competitor's toolset, the only differentiator left is the person using it.
And here's the uncomfortable truth: most media buyers are more similar than they are different. They've read the same blogs, taken the same courses, follow the same Twitter accounts, and apply the same "best practices." The variation in outcomes between one competent media buyer and another, when both are using the same tools, is narrower than anyone in this industry wants to admit.
Where Commodity Tools Actually Fail
The standard stack doesn't just produce average results. It actively prevents certain kinds of work from happening at all.
Creative Is the Biggest Lever (and the Most Neglected)
Everyone in Meta Ads knows that creative is the #1 performance lever. Post-iOS 14, targeting is largely automated. Bidding is algorithmic. The only variable that meaningfully separates campaigns is the creative.
And yet, most agencies produce creative using template-based tools designed for speed, not effectiveness.
Here's what happens when you use Canva or AdCreative.ai as your creative engine: you get polished-looking ads that look like every other polished-looking ad in the feed. They're competent. They're on-brand. And they perform like the average of everything else those tools produce -- because they are the average. The templates constrain the output. The AI suggestions are trained on the same data everyone else's AI suggestions are trained on.
What's missing is the intelligence layer. Who is this ad for, specifically? What awareness level are they at? What cognitive biases are most likely to influence this audience segment? What are competitors running right now, and how do we position against it?
No off-the-shelf creative tool answers those questions. They skip straight to "pick a template and type your headline." That's like skipping the diagnosis and jumping straight to prescribing medication.
Attribution: The Comfortable Lie
Here's a number that should bother you: Meta routinely reports ROAS figures that are 30-50% higher than reality. Google does the same thing. Both platforms have financial incentives to take credit for as many conversions as possible, and their attribution models are designed accordingly.
The standard agency response is to buy a third-party attribution tool -- Triple Whale, Northbeam, or similar. These tools are better than platform-reported numbers. But they're still black boxes. You're paying $300-500/month to trust someone else's attribution model instead of building your own.
We took a different approach. We built a reporting system that pulls directly from Shopify's order data, matches orders by UTM source against actual ad spend, and calculates real ROAS from real revenue. No modeled conversions. No statistical projections. Actual money in, actual money out.
Is this approach less sophisticated than a machine-learning attribution model? Maybe. Is it more honest? Absolutely. And when you're making spending decisions with a client's money, honest matters more than sophisticated.
Landing Pages: The Forgotten Conversion Point
Most agencies either send traffic to a client's existing product page (and pray the site converts) or spin up a generic Unbounce page using a template.
Both approaches have the same problem: they treat the landing page as an afterthought rather than a critical piece of the conversion chain. A landing page should message-match the ad that drove the click. It should have one goal, one CTA, one narrative thread. And it should be testable -- headlines, CTAs, hero images, layout -- without requiring a developer.
Template page builders optimize for ease of creation, not conversion rate. They give you drag-and-drop flexibility at the cost of opinionated structure. You want opinionated structure on a landing page. You want the system to enforce "one conversion goal per page" and "headline matches the ad" because those constraints produce results.
What Custom Infrastructure Actually Looks Like
When we say "custom tools," we don't mean vague or theoretical. We mean specific systems we built and use daily. Here's what that looks like in practice.
Campaign Management From the Terminal
We built a command-line interface for the Meta Ads API. Instead of clicking through Ads Manager to create a campaign -- navigating nested menus, setting the same defaults repeatedly, hoping you didn't fat-finger a budget field -- we define campaigns in structured JSON briefs and push them from the terminal.
One command creates a campaign, its ad sets, and its ads. Everything starts paused. Every push gets a dry-run preview first. The CLI handles API versioning, authentication, rate limiting, and the dozens of edge cases the Meta API throws at you.
Why does this matter for clients? Speed and consistency. A campaign structure that takes 45 minutes to build in Ads Manager takes 5 minutes from the terminal. And there's no "I accidentally set the wrong attribution window" because the brief is version-controlled and reviewed before it ships.
A Creative Intelligence Pipeline
This is where the gap between commodity and custom is widest.
Before we create a single ad, we run an intelligence pipeline:
Market Research: We analyze customer voice data -- reviews, Reddit threads, forum posts, social comments -- to understand how your audience actually talks about their problems. We map competitors: their positioning, their offers, their creative patterns, their messaging gaps. We assess market sophistication to determine how aware your audience already is.
Persona Development: The research gets segmented into detailed personas. Not the decorative "Meet Marketing Mary" personas that traditional agencies create. Ours include awareness levels, layered pain points (surface problem, underlying cause, emotional driver), cognitive bias profiles, and exact language patterns. Each persona directly informs which ad angles, hooks, and visual styles we test.
Copy Generation: Ad copy is written using behavioral science frameworks -- specific cognitive biases matched to specific audience segments. Loss aversion for high-awareness audiences who know the alternatives. Social proof for skeptics. Anchoring for price-sensitive segments. Every copy variant includes a behavioral science breakdown explaining why it should work, so we can learn from results regardless of whether the ad wins or loses.
Image Prompt Generation: Visual creative is generated from prompts designed to trigger specific emotional responses in specific audience segments. The prompts incorporate brand guidelines, competitive positioning, and the cognitive bias stack for each persona.
This pipeline doesn't replace creative judgment. It replaces the guesswork that most agencies disguise as "strategy." When we launch a creative test, every variable is intentional and trackable. When something wins, we know why. When something loses, we can iterate precisely instead of guessing again.
Zero-Friction Split Testing
We built a split testing system backed by Supabase. A browser-based dashboard lets us define A/B tests on live landing pages -- swap a headline, change a CTA, test a new hero image -- and the changes take effect immediately. No code deployment. No developer tickets. No waiting three days for a headline test to go live.
Visitors get assigned to variants automatically. Results calculate statistical significance in real time. We can test a new landing page headline in 30 seconds.
Most agencies run landing page tests quarterly, if at all. We test weekly because the infrastructure makes it trivial.
Attribution You Can Actually Verify
Our reporting system is a single command. ./report.py shelskys generates a consolidated weekly report across Meta and Google with real Shopify revenue data. No copying numbers between platforms. No spreadsheet formulas. No dashboard that breaks when Meta changes their API.
For true ROAS, we match Shopify orders by UTM parameters to actual ad spend. The numbers are lower than what Meta reports. They're also real. Clients who've never seen their actual ROAS -- as opposed to the number Meta claims their ROAS is -- tend to find this clarifying.
How to Tell If Your Agency Is Actually Different
You don't need to understand the technical details of what we build. But you should be asking your current agency (or any agency you're evaluating) these questions:
"What tools do you use that your competitors don't have access to?" If the answer is a list of SaaS subscriptions, those tools aren't a differentiator. Any agency with a credit card can sign up for the same ones.
"How do you generate creative concepts?" If the answer starts with "our creative team brainstorms" or "we use AI to generate variations," dig deeper. What informs the brainstorm? What data drives the AI? If there's no research pipeline feeding into creative decisions, you're paying for intuition -- which is fine, until it isn't.
"How do you calculate ROAS?" If they point to Meta's in-platform reporting, they're giving you Meta's numbers, not yours. If they use a third-party attribution tool, ask them to explain how it works. If they can't, they're trusting a black box.
"How fast can you test a new landing page headline?" If the answer involves a developer, a ticket, and a deployment schedule, the answer is "too slow." Headline testing should take minutes, not days.
"What happens to the learnings when a creative test fails?" If failed tests just get paused and forgotten, there's no compounding knowledge. Every test -- winner or loser -- should produce a documented insight that informs the next test. That requires a system, not just a media buyer with good instincts.
The Real Differentiator Isn't Strategy. It's Infrastructure.
Every agency claims to have a proprietary strategy. Press them on what that means and it usually comes down to "we're experienced and we know what works." That might be true. But experience without infrastructure produces inconsistent results. It works when the experienced person is paying attention and breaks when they're busy with another client.
Infrastructure scales. It enforces consistency regardless of who's working on the account or what day of the week it is. It produces compounding returns because every test, every campaign, every client engagement improves the system itself.
When we build a research pipeline for one client, the pipeline gets better. When we create a creative testing framework for one brand, the framework sharpens. The tools improve across every client simultaneously because the intelligence is embedded in the system, not just in someone's head.
That's what separates an agency that builds its own tools from one that buys the same SaaS stack as everyone else. It's not that the tools are magic. It's that the tools compound.
What This Means For Your Brand
If you're spending $10K+ per month on Meta Ads through an agency, you deserve to know what's actually happening under the hood. Not the slide deck version -- the real version. What tools are they using? Are those tools the same ones every other agency uses? Is your creative being generated from templates or from research? Is your attribution honest or flattering?
These aren't gotcha questions. They're the basic due diligence that separates brands who grow from brands who churn through agencies every six months wondering why nothing changes.
We built our infrastructure specifically because we wanted the answers to those questions to be different. Not different in a marketing sense. Different in an engineering sense. The tools our clients' campaigns run on don't exist anywhere else, because we wrote them.
If that sounds like a better way to run Meta Ads, let's talk.