"Can I just use AI to run my ads and skip the agency?"
Some version of that question shows up more than almost anything else.
What people usually mean is a third-party AI tool that connects to your Google or Meta account, and it acts as your media buyer. It builds the campaigns, makes the changes, finds the winners, and scales them while you go run the business.
My short answer is that these tools can do pieces of the job well. I haven't seen one do the whole job. And most people asking the question are focused on the wrong AI. The one that decides whether your money goes to the right people is already running inside the ad platforms, and almost nobody is feeding it properly.
What third-party AI tools do well for paid ads
I want to be fair here, because a lot of agencies get defensive about this and it shows.
Search term reviews are a good fit for AI to do. I've seen AI scrub a Google Ads search terms report and come back with solid recommendations on what to add as a keyword and what to add as a negative. It's a real time saver. But it still takes a strategist to make the call, because the tool doesn't know which of those terms have true intent behind them, or just look good on the surface.
Audits are the same. AI can run a generic account audit and pull out insights in minutes that would have taken a person hours. That speeds up account reviews and a lot of routine checks. But it's only effective in the hands of someone who already knows what they're doing, because that person can tell which insights matter and which ones are noise (similar to Google's Recommendations tab they're have for years).
Ad copy is a good use too, with direction. We use large language models to write a lot of our search ads. We give it the target keywords and our own rules for how an ad should be built. Those ads consistently get higher click-through rates than the hand-built ones Google scores as "excellent." Then we test them against each other in regular A/B tests like we'd normally do.
So AI is real leverage on the analysis and production side. The question is whether it can run the account.
The AI media buyer that runs your account by itself doesn't exist yet
Here's my honest assessment. I don't think these tools are doing what people hope they do, where the AI runs on its own, gets smarter every week, and just makes you money. I haven't seen that happen.
Most of the AI ad tools I've looked at were built for ecommerce. That makes sense, because a purchase is a clean signal and the tool can see it. Lead generation is an afterthought. When I've asked these tools about offline conversion tracking, I've ended up explaining the concept to them, which tells me it isn't built into the product. A tool that can only see form fills will optimize toward form fills, and I'll get to why that's a problem below.
I also haven't seen an AI tool run strategic tests and learn from them. Deciding what to test, why, and what the result means for the next test is still a person's job.
And I haven't seen one create solid ads on its own. It writes good ads when it has strategic direction going in and someone on the back end judging whether what it made is any good. Without those two things, you get volume, not quality. It also can't tell you which problem your buyer will actually pay to solve, or which offer is strong enough to get them to raise their hand. That comes from testing messages against your real market and reading the results against pipeline.
Handing an AI tool the keys can get the account shut down
This one doesn't get talked about enough. I've heard from colleagues and other agency owners about businesses whose ad accounts were shut down after AI tools, or other automation, made too many changes too quickly, in a way that didn't look like a human was doing it.
I can't tell you how often that still happens. From what I've read, the platforms don't ban anyone for using AI itself. They act on behavior, like changes made faster than a person could make them, or connections the platform hasn't approved.
The consequence is what makes it serious. Under Google's circumventing systems policy, a suspension happens on detection and without prior warning, and creating new accounts to get back in is itself a violation. You don't just start over with a fresh account. For a business that depends on paid media for pipeline, that's not a risk worth taking to save a few hours a week.
How we use AI at Gradari
So here's where we've landed for now.
We use AI to pull insights out of accounts faster, check that budgets are pacing the way they should, and flag anything that's changing or needs a closer look. It's leverage on the analysis side, and it's made the team faster.
Every change inside an account, we make ourselves. Our AI tools don't have write access to client accounts. Before we'd ever let one touch a client account, we'd test it on our own account first and prove it out there.
That will probably change as the tools get better and as platform policies allow AI more into the system. It isn't worth gambling a client's account to find out early.
The AI most people overlook is already inside Google and Meta
Here's the part I think matters more than any outside tool.
The ad platforms are already running on AI, and it's very good. When Google decides who to bid on, it's running a cluster analysis across millions of data points on every person, far more than any media buyer, or any outside tool, will ever see. The ten or twenty levers we get to pull in the interface are a small control panel bolted onto a machine doing the real work in the background.
Bidding, audience expansion, placements, and budget pacing are mostly the platform's job now. Fighting it on those usually loses. The real lever is what you teach it to look for.
It optimizes toward whatever you call a conversion
Most accounts tell the platform to get as many conversions as possible for as cheap as possible. That's the default, and it's what a non-marketer, or an ecommerce-first AI tool, will almost always pick.
The problem is auction economics. The people you actually want, decision makers with budget, are the same people every other advertiser wants, so they cost more to reach. Cheap conversions are cheap for a reason.
In Google's eyes all conversions are equal, because you haven't given it anything better. So it finds the cheapest form fills it can, and it's very good at finding them. I call this the Reverse Optimization Trap, where the platform keeps getting better at finding leads you don't want.
A higher lead or MQL rate isn't proof it's working either. One of our clients watched their MQL rate spike to around 50% after an account restart, then settle near 25%. The 25% was the healthier number. At 50%, the platform had learned to find people who looked qualified on the form and then didn't turn into sales opportunities. When we started optimizing for SQLs, the lead to MQL rate dropped, but the MQL to SQL rate skyrocketed and we started driving more revenue. This will be unique for every business.
It can't see what happens after the form fill
The platform only knows what you send it. If the only thing it ever sees is "someone submitted a form," that's the only thing it can optimize toward. It has no idea which of those people took a sales call, became an opportunity, or signed.
The fix is offline conversion tracking, which means sending qualified and closed outcomes from your CRM back to the ad platform, matched to the original click. Once the platform gets that, the same intelligence that was finding junk starts finding buyers.
Almost nobody does this. I've audited more than 140 lead gen ad accounts, and one of them was feeding real CRM outcomes back to the ad platform when I first looked. If you've ever wondered why your ad platform and your CRM never agree, this is usually where it starts.
No outside AI tool fixes this for you, because it isn't an ad account problem. It's plumbing between your forms, your CRM, and the platform.
Bad data is worse than no data
When tracking is off, nothing looks wrong for a while. Then it falls apart all at once, and it looks like an audience problem.
A partner of ours was running webinar registration ads for an enterprise client. Cost per registrant held steady around $25, then jumped to about $112 overnight and stayed there for a week. Nothing on the page had changed.
The registration event was firing roughly 15 times per registrant, and on other buttons too. The algorithm was learning from thousands of false signals, got confused, and drifted toward broader, cheaper, worse traffic.
The tell was that a duplicate campaign pointed at an event that never fires at all brought registrants back at a normal cost. No signal beat a bad signal. The platform's AI was working exactly as designed, on data that didn't describe the outcome.
An AI tool sitting on top of that account would have seen the same dashboard everyone else did, full of conversions. It's the conversions themselves that were wrong.
Creative is now part of the targeting
On Meta especially, the creative is doing a lot of the targeting now. Different ads find different people. So the number of genuinely different ads you can produce starts to cap how well the platform can find buyers. We've written about what to do when your only winning ad wears out.
The platforms also keep adding AI settings that act on their own. Google's AI Max can test different landing pages on your site automatically. I've had it surface a page that converted better than the one we picked. I've also had it send budget to a page that had nothing to do with the offer. Someone still has to be watching.
So can you run paid ads with an AI tool and no agency?
Sometimes, yes. Here's how I'd think about it.
It can work when these are true.
- Your sale is simple and closes on the spot, so a form fill or a purchase really is the outcome you want.
- Your spend is small enough that learning on your own budget is an acceptable cost.
- Someone who knows paid media reviews what the tool recommends before anything changes.
It gets risky when these are true.
- A lead is not a sale, and someone on your team qualifies leads before they turn into revenue.
- Your sales cycle runs longer than a few days.
- Your CRM and your ad platform don't talk to each other.
- The plan involves giving an AI tool permission to make changes in the account on its own.
Being able to run it yourself and it being the right use of your time are two different things. I wrote more about that tradeoff in how to decide between an agency and an in-house team.
What to set up before you let any AI spend your budget
Whether you use a tool, hire someone, or do it yourself, this is the order I'd go in.
- Audit what you count as a conversion. Open the conversion actions in your account and confirm each one fires once, on a real action, from a real person. Most of the problems I find start here.
- Define a qualified lead with sales, not just marketing. Write it down. It's the definition the platform will eventually optimize toward, so it has to match what actually closes.
- Connect your CRM to the ad platform. Pass the click ID through your forms into the CRM, then send qualified and closed stages back as offline conversions. Once that data exists, here's how to track unqualified leads without wrecking your bidding.
- Consolidate so each campaign gets enough signal. The platforms need roughly 15 to 30 qualified conversions a month to learn well. Spread that across ten campaigns and you starve all of them.
- Then let the platform's AI do what it's good at. Broad match, Performance Max, and Advantage+ all work far better once they're learning from the right outcome.
Expect cost per lead to go up and lead volume to go down when you do this. Lead quality goes up by a lot more, and your cost per customer drops. That trade is the whole point.
Yes, this is the whole playbook
If you want a second set of eyes on your setup, our free system review looks at exactly this, what your ad platforms are learning from and whether it matches what actually turns into revenue. You can also see how we run the full system.
Sources
About the author
Kyle Rutledge
Owner
I’m Kyle, founder of Gradari, a paid ads lead generation agency that helps B2B and SaaS companies stop wasting budget on low-quality leads and start building systems that actually drive growth.