Quick math up top, because that's the deal here.
If you've searched "how long does it take to make money with ai" and gotten nothing but vague answers, I don't blame you. I didn't find one either, so here's my actual timeline instead: six months, two dead products, and $1,260 a month at the end of it.
Revenue last month was $8,340. Facebook ate $4,890 of that in ad spend, which is a brutal ratio and I'll get into why later. Product cost and shipping ran $1,975. Payment processing and app fees took another $215. What's left is $1,260, and that's the number I actually keep.
I'm Owen, 31, and until February I adjusted property claims for an insurance company in Tulsa, the kind of job where you spend your day arguing with contractors about roof shingles. My whole department got folded into a "claims automation initiative," which is corporate for: a model reads the photos now and you don't. Two months of severance, then a lot of free time.
A video showed up in my feed a few weeks later. A guy walks through a five-step method: use Claude to find a market, then a specific problem inside it, then a product, then validate it with real numbers, then build a plan for when the ads inevitably stop working. He claims to have found something doing $1,000 a day with it. I did not find anything doing $1,000 a day. What I found, after two dead ends and about six months of actually working the method in order, is a small store that nets $1,260 a month.
Product one was a posture corrector. I picked it because it looked good in the example ads, which I now know is exactly the wrong reason to pick anything. Three weeks and about $340 in test spend later, it had sold four units. Product two was a pet grooming glove, a better instinct but still the wrong move, because I skipped the validation step out of impatience and spent $290 finding out nobody was buying it from me specifically. I quit that one too.
One thing before I get into what actually worked. I spent close to five weeks bouncing around dropshipping tutorials on YouTube trying to piece together what a "good" ad number even looks like, and every video had a different opinion. A structured course laid it out once, in order, and it took an afternoon to understand what I'd been guessing at for over a month. YouTube is fine once you already know what you're looking for. It is a maze when you do not.
One more thing, and it's the single most useful part of this whole post, so don't skim past the end. Most people who test a product and lose money blame the product. It's usually not the product. There's a specific, checkable way to tell what actually went wrong, and I'm saving it for last on purpose.
I stopped picking the product first
The biggest thing I got backwards on the first two tries: I went straight to "what should I sell" instead of "who am I selling to." A posture corrector shown to someone mid-scroll is a different business than the same posture corrector shown to someone who just spent forty minutes searching why their back hurts after work. Same object, different market, very different odds.
So on attempt three I picked a market and left it alone: pain and mobility. Not original, it's about as unoriginal as markets get, but people in actual physical discomfort search with intent and buy with intent. US retail ecommerce keeps growing every quarter according to the Census Bureau's own retail data, and pain and mobility is a permanent, recession-proof slice of it. That's most of the game right there, and I skipped it twice.
Finding a problem before a product
With a market picked, I asked Claude to get more specific: what's a narrow emotional problem inside "pain and mobility" that people are quietly frustrated by every day, something with a name they already type into Google because they're already living with it. Not "back pain" broadly. The exact ache in the bottom of your heel the first ten steps out of bed is a problem. "People with foot pain" is just a market wearing a problem's clothes.
I ended up with a shortlist that included plantar fasciitis, tennis elbow, and a neck-and-shoulder-tension cluster. This step didn't require genius, as far as I can tell. It required specificity, and I'd skipped that part twice already.
Checking who's already winning before I spend anything
Here's the part I wish someone had told me before product one and two: you don't need a clever idea, you need a product that's already proven itself somewhere. Meta's own Ad Library is free and public, and searching a niche there shows you which advertisers are currently running a lot of active ads, which is a decent proxy for "this is working for someone right now." A page running well over a hundred ads in your niche didn't get there by accident.
That's where I found it: a page running well over a hundred active ads around a fairly specific plantar-fasciitis angle. Different product than either of my first two entirely, and the ad volume alone told me something neither of my failures ever had going for them.
Doing the math before believing anything
Ad volume proves someone's spending money, not that they're making any. So the next step was a rough public estimate using a free traffic-estimator browser extension (accuracy varies, treat it as a ballpark, not a fact). Their site was pulling several hundred thousand visits a month. I multiplied that by a conservative 2% conversion rate and an estimated order value, landed on a revenue estimate north of six figures a month, and figured even if I was off by half, it was still a real business underneath the ads.
Then I set two rules before testing anything, and I've kept both since. At least 3x cost of goods sold at my price point, so a product costing $15 landed doesn't get sold under $45. And a minimum of $30 gross margin per unit, because ad costs keep climbing and the $20 margins that worked a couple years ago don't survive today's CPMs. My winning product cleared both with room to spare.
Where I actually learned to do the Claude side of this properly
I'd used Claude before this in the vague, unstructured way most people do: ask it something, get a mediocre answer, move on. What changed was a short course on Mindwand that walked through prompting for actual research tasks in 15-minute daily lessons, which fit a laid-off guy's wide-open schedule fine. Skool has communities doing similar things, and Coursera has structured courses too if that format suits you better. Pick one. Finish it.
The part I saved for last: is it the ads or the funnel
Most people who test a product and it doesn't sell decide the product was wrong and quit. My read, after watching my own first two failures up close, is that it's almost never the product. It's your ads or your funnel, and there's a specific way to tell which.
If your cost per click is elevated (I aim for around $1.50 in the US), check your click-through rate first. Under 3% and your creative is the problem, no amount of funnel tweaking fixes a boring ad. If your CTR is healthy, closer to 4 or 5%, but your CPC is still high, look at your CPM. Anything north of $80 in the US is worth investigating, and that usually loops back to the creative anyway, since the platform rewards ads people actually engage with.
If the ads check out, look at the funnel. An add-to-cart rate under 10% of landing page visitors usually means your traffic isn't as qualified as you think, or your page isn't answering the objection your ad raised. Cart-to-purchase completion should clear something like 30%. Below that, something's breaking between "I want this" and "I paid for this," and it's usually checkout friction or a payment processor issue nobody bothered to check.
Looking back, my first two products probably had problems on both sides. This one had neither, which is most of why it worked when the other two didn't.
What $1,260 a month actually costs me
Averaged across all six months, counting the two dead products and every hour spent reading traffic estimates and scrolling an ad library, I land at something like $11 an hour. That number climbs every month the winning product keeps running, since most of the setup work is already done, but I won't pretend the early average is impressive. It isn't.
I don't think this replaces a job, at least not six months in. It's a real $1,260 a month that exists because I finally ran a process in order instead of skipping to the part where you spend money. If you're still asking how long it takes to make money with AI doing something like this, my honest answer is: longer than the video made it look, and probably shorter than doing it with no process at all. If you want to know which lane actually fits you before you spend a dollar on ads, the Mindwand quiz takes about three minutes and points you at one, and Skool and Coursera remain solid options for the learning half. Either way, run the process in order. That's the only part I'd call non-negotiable.
For a broader read on what these methods actually pay, see our make money with AI breakdown, or our ranked take on 5 AI side hustles.