Paid advertising is now a baseline skill for marketers, and the internet is full of articles teaching you how to work each platform's dashboard. That's not what this is. Instead, I want to break down the underlying logic of how digital ad systems are designed: the core concepts and optimization principles that apply no matter which platform you're running on.
Data, data, and more data
This is the most important idea in the whole piece, and also the one people skip past fastest, so I'm going to say it three times. Why does data matter so much? You have to look at how ad platforms make money, and where their business models are ultimately headed. Start by asking what Google's and Meta's vision for advertising really is.
"Control your buying impulses at a subconscious level: whatever they push, you buy."
To achieve that vision, their most valuable asset isn't their data centers or how many Nvidia chips they've grabbed. It's how much data they hold: every online and offline purchase, everything you browse, every conversation you have with friends, your body language, even your micro-expressions, all used to predict what you're about to buy.
This whole surveillance-capitalism argument is a long, complicated rabbit hole. If you're curious, read Shoshana Zuboff's The Age of Surveillance Capitalism. But here's the one conclusion you need to walk away with: the real asset you're handing these ad companies isn't your ad spend, it's your data.
The core principle: data is king
"Data" here carries two meanings: telling the platform who you are and who's buying.
The first is simply the business information you fill in when setting up your ad account, and the only rule that matters is more detail, more fields, always better. This isn't a one-and-done task, either. Take Google as an example. My business profile has gone through multiple redesigns, and every redesign adds a few new fields to fill in. Keep maintaining your business profile without complaint, and fill in every new field completely. This isn't only about data accuracy. The more diligently you update it, the more the platform tends to reward you with resources.
Take it a step further and you should be feeding your product data to the platform in a proper feed format, keeping product information updated in real time. That's how you unlock more automated placement tools, and this kind of automated product advertising is usually the format that drives the most revenue.
The second meaning centers on connecting your conversion data, a bit more technically demanding, but the single most important thing you have to get right. Whether it's through GTM or server-to-server postbacks, if you can't figure it out yourself, get your dev team to help. Make sure your basic data connections are solid before you touch any ad settings, otherwise you're just firing blind forever.
A couple of small but important warnings here: don't ask friends and family who aren't your actual customers to like your page, and don't buy junk traffic through black-hat tactics. Contaminating your own data is pure downside. There's no upside to it at all.
Bottom line: fill in every field, and make sure your conversion events are properly connected.
The counterintuitive truth about audiences
Most people have some baseline mental picture of their target audience. But when you're actually running ads, throw out every assumption you've made about who your audience is, and feed the platform as much of your product and customer data as you possibly can.
Most marketers have heard the classic "beer and diapers" story. It's largely an urban legend statistically, an interesting anecdote without real statistical significance, but the underlying logic does hold up in digital advertising.
Here's a case from my time at an agency. I was running ads for a women's lingerie brand. Performance was middling at first. We were hitting the target ROAS (return on ad spend), but the numbers weren't strong enough to convince the client to raise their budget.
Then one day I made a single audience change, and performance took off. What I did was remove the gender targeting that had been locked to women only. It went against everything the product's positioning implied, and yet it sent sales climbing. If you know your marketing theory, you already know what this means: the people buying the product and the people using it weren't the same audience. Exactly like beer and diapers.
So should I just not set any audience targeting at all?
Sort of yes, sort of no. It depends on the ad format and platform. For search ads, lock down country and region and leave everything else alone; that's the safest approach. At most, set some loose target audience signals to point the system in the right direction, but never lock the targeting down too tight.
For display network placements, you'll generally still want to set an audience, depending on how mature that network is. Here's another example, and a fun one: back when LINE first started running LAP ads, the targeting system was so primitive that everyone's performance was mediocre across the board. Checking an audience filter meant your ads barely delivered at all, and leaving it unchecked kept CPC stubbornly high. So I did something a little reckless: I checked every single audience segment available. In theory that should behave the same as leaving it unchecked. Instead, CPC dropped by a flat 75%. That kind of illogical trick can work wonders on an immature ad system, though of course it stopped working before long. Ha.
If you're new and only running ads on the two major platforms, treat that last story as pure entertainment and ignore it entirely. On a mature ad system, the most effective audience setting is closely tied to your own data, lookalike audiences being the standout example. Test your own percentage range for lookalikes; there's no universal number, and it varies a lot by industry. And if you're just getting started with too little data, don't bother using lookalikes yet, similar to the idea of a minimum viable sample size in statistics. You want at least 2,500 unique users at a minimum, and ideally 10,000-plus before you start relying on lookalike audiences.
Ad platforms will also usually offer some built-in interest categories or audience targeting options. There's really only one rule here: go broad. These audience categories are almost always structured like a tree, and no matter how well you think you know your users, start by testing the broadest top-level category first. Who's to say someone into cosplay games won't also click on a strategy sim?
E-commerce advertisers have another important tool: the marketing funnel. Most people have heard of this concept but don't realize how directly it applies to digital advertising. This is essentially what remarketing is. It gets fairly involved once you dig in, so I'll save that breakdown for another time.
Bottom line: don't box in your imagination of who your audience is, a smaller pool isn't automatically a better one.
Climbing out of ad design hell
Ad creative is usually the part that gives media buyers the biggest headache. Good creative drives higher click-through rates, and higher CTR means more chances to convert. I won't get into brand guideline conventions or classic advertising design theory here; go build a good relationship with your designer and trust their eye. What follows are a few ideas specifically about how digital ad creative behaves as a category.
The 3-6-9 rule for ad creative
If you know your way around Japanese mahjong tile efficiency, you've probably heard of the "147, 258, 369" number groupings, but the "369" here has nothing to do with tile efficiency. I'm just borrowing the numbers. Simply put, digital advertising is a system that learns and optimizes over time, and the same "feed it plenty of data" principle from earlier applies just as much to creative. Never assume a single image is good enough and run only that one.
Stick to this rule: if it's a format that shows one image at a time, design at least three distinct creative variations. Same logic for something like search ads that display three lines of copy at once, write at least six sets of copy and let the system optimize across them. If you can, just fill every available slot the platform gives you. More is always better here.
Design mobile-first
This should go without saying by now. Only old-timers are still browsing on desktop. Most people complete their online purchases on their phones these days. Stick to these three ad sizes: 300x300, 300x250, 320x100, and prioritize designing for those. Always preview how the creative renders on mobile, and aim for something clean with a clear single message. Don't cram in a wall of text trying to hit every selling point at once.
For video, design for silence. Even when someone's browsing with sound off, your core message needs to land within the first three seconds. Ideally the first second, honestly. Attention spans keep shrinking in the TikTok era.
Automate whenever you can
Here's a slightly more advanced point. Whether it's DPA ads or HTML5 formats, whenever a platform offers you an automated creative option, use it. That said, this only applies to conversion-focused campaigns. If the goal is brand awareness or getting a specific message across, stick with manually designed creative.
Out of ideas?
If you're stuck on design direction or copy even in the age of AI, just go copy someone. Same way Palworld borrows from Pokémon, or Genshin Impact borrows from Zelda. Go look at what's already working in the market, ideally your direct competitors' ads. Whatever design the market has already rewarded is, by definition, a good design. Pull from a handful of proven winners, break down what makes them work, internalize the pattern, and you can rough out something the market is likely to respond to.
Bottom line: design multiple creative sets for the system to learn from, prioritize mobile, and automate wherever you can.
The bidding mechanism is absurdly complicated
Real-time bidding is probably the strangest part of the whole digital ad system, and it doesn't work like the auction model most of us intuitively picture. Most digital ad platforms run on a second-price auction: in short, the highest bidder wins, but only pays what the second-highest bidder offered. It's a little roundabout and a little confusing, and the bid itself isn't the only variable at play. Your account and ad quality matter, and so does your budget. You don't need to fully master the mechanics, just get the gist of it.
Where do you find cheap ad prices?
There's only one path to good pricing: keep the platform happy. And I don't mean sending your account rep a gift basket. I mean exactly what I said earlier: give the ad system the data it wants. The more you feed it, the higher your account quality climbs, and the system rewards that with better pricing. It really is that simple.
What tier am I at?
A brand-new ad account and one that's been nurtured for five or ten years are operating on completely different levels. The most direct measure is how much you've spent on the account, how much conversion data you've fed it, and how many impressions it's served. This part is just reality: an account with no history is going to underperform, so don't expect to walk in and get great pricing bidding on CPA (cost per action) right out of the gate.
In the early stage of an account, bid on CPM or CPC to build up your account's data history. Spend broadly, feed the system your conversion data in parallel, and it will gradually learn who's actually buying from you. Bottom line: you have to pay your dues and put in the time. There's no shortcut around it.
Should I try automated bidding?
Short answer up front: that's a toy for advanced players. For accounts with history, automated bidding often lands you cheaper, better pricing, because the system has more freedom to read complex buyer signals. But at a stage where the ad system barely knows you yet, letting automated bidding loose just raises your odds of getting yelled at by your boss. I'd rather underbid and have ads deliver slowly than gamble on that. Modern bidding systems are less likely to blow through budget wildly than they used to be, but personally, I still wouldn't take that bet.
How do I set my bid?
Work backward from your goal. Say you want to know how much you're willing to spend per paying customer. Use your conversion rate to work out how many clicks you need to buy, then use your click-through rate to work out how many impressions you need to buy those clicks. That gives you your target CPM. The more granular your funnel gets, and the more layers you add, the more precise this backward math becomes. But these three basic layers alone cover most situations. It's just simple arithmetic.
Should I trust the bid strategy suggestions the system gives me?
Never.
Bottom line: know exactly where your account stands, the more the system likes you, the sweeter the pricing gets.
How to optimize your ads
Now, finally, ad optimization. Honestly, if you've done everything above properly, your ads shouldn't perform too badly to begin with, but every so often something tanks your performance out of nowhere anyway. Based on how these systems are designed, here are a few tricks for pulling a campaign back from the brink.
Always check your data first
The moment performance suddenly drops, the first thing to check is whether your data connection has broken somewhere. Setting it up once isn't a set-it-and-forget-it situation. You'll run into weird issues that tank your matching rate out of nowhere, and that's a serious problem. Get your dev team into a room immediately; fixing this is priority one.
Let your ads run a full 24 hours
This applies when your budget is too low for the ad to spend through the full day. At that point you have two options: raise the budget or lower it, and raising it is almost always the better move. Raising the budget does more than pull in more impressions and clicks. It gives the system more room to test and learn, and it's not unusual to see your CPA actually drop after a budget increase. Counterintuitive, but it happens.
A campaign that was running fine suddenly dies
This usually shows up after a budget runs dry. If your ad account's top-up balance isn't stable and you accidentally let the budget hit zero, restarting the campaign often doesn't bring the delivery volume back. There's one strange little trick that helps here: duplicate the campaign exactly and turn off the original. Don't ask me why this works. Nobody really knows.
Don't adjust too often
An ad campaign is like a houseplant. It's fragile, and it dies easily if you keep fussing with it. Leave it alone as much as possible. Wait at least 24 hours minimum between changes; once a week is a healthier cadence if things are going well. Trust me, constantly poking at it will only kill it faster. If you genuinely have a fresh idea you want to test, use the platform's A/B testing feature instead. Give it time to learn, and go focus on something else in the meantime.
Bottom line: always check your data first, and give the system enough budget and enough patience.
Wrap-up
Every platform has its own quirks and details that differ at the margins, and working those out comes down to your own research skills and willingness to keep testing. Running digital ads, at the end of the day, is a lot like dating the system: treat it well, feed it the data it wants, and it rewards you with better pricing and better performance. Every stage has its own challenge. Data connection is the foundation, bidding strategy takes time to build real experience, and creative is a battle of ideas. But what actually matters most is having the right mental model, staying curious, being able to take the heat when things go wrong, and having patience.