.png)
Say you're spending $20,000 a month on Google Ads and you ask three providers what they'd charge. The traditional agency quotes 15% of spend, so $3,000 a month, plus a $5,000 onboarding fee and a 12-month contract. An AI agency quotes $2,000 flat and tells you their models do the heavy lifting. The DIY option costs you a $149/month optimization tool and about six hours a week of somebody's attention. Most buyers line those three numbers up in a spreadsheet, pick the cheapest one they can stomach, and get the decision wrong. The management fee is the smallest number in the comparison. The one that decides the outcome is how much of your $20,000 gets spent on decisions nobody has revisited in eleven days.
I used to argue that any account under about $30k a month should skip agencies entirely: turn on Smart Bidding, mine your search terms weekly, refresh copy every quarter, keep the 15% in your pocket. I gave that advice for years and it was half right. The half that held up is that most of what agencies bill as strategy is maintenance. The half I got wrong is the assumption that maintenance is cheap when you do it yourself. It isn't. It's cheap in cash and expensive in latency, and latency in a live auction has a price you can measure: every day a junk search term keeps spending, every week a losing ad group keeps its budget, every Smart Bidding target still set to a CPA you picked in March.
So this is a comparison of the three models by the only thing that separates them: who or what makes the change, and how long the gap is between noticing something and fixing it. I'll put real prices next to each, including the ones that don't appear on a rate card, and I'll name who should stay exactly where they are. If you're spending under $3,000 a month, you can stop reading now and go do it yourself. There is no management model on earth that earns its fee out of a $3k budget.
A 15% retainer buys you a fixed slice of one person's week. At $20k spend that's $3,000 a month, and if you assume the agency is running at a healthy margin, the labour inside it is maybe six to ten hours. Ask what happens in those hours and the honest answer, in most shops I've seen from the inside, is: a Monday bid and budget review, a search terms pass if there's time, a monthly report, and the call where they present the report. The account gets touched on a cadence set by the agency's staffing, not by your auction. And the fee scales with the thing you want to control. If you double spend to $40k, the work does not double, but the invoice does. That's the part clients notice around month nine.
The other cost line in a retainer is the one nobody puts in the proposal: the onboarding fee and the lock-in. Five thousand up front and a 12-month term means the first three months are a sunk-cost trap. If performance is flat in month two, you don't switch, because you've already paid for the runway. I've watched businesses stay eight months past the point they knew it wasn't working, purely because leaving felt like admitting the five grand was wasted. It already was.
DIY in 2026 means Smart Bidding doing the within-campaign math, a tool like Adalysis at $149 a month flagging ad tests and Quality Score drops, maybe a script or two you found on a blog, and you or a marketing manager checking in a few times a week. On paper it's a $150 monthly line item. In practice you're paying somebody's salary for six hours a week. At a $70k-a-year marketing manager, that's roughly $2,000 a month of loaded cost, which puts DIY within a few hundred dollars of the agency you didn't hire. Nobody accounts for it because it hides inside a headcount you already have.
The real cost is the gap between recommendation and action. Every tool in that stack surfaces things: this search term is wasting money, this ad group's CTR fell off a cliff, this asset group is eating your branded traffic. None of them press the button. So the value of the recommendation decays while it sits in a queue behind a product launch and two client emergencies. I've run accounts where the notification list had 40 open items and the person responsible was three weeks behind. The tool did its job perfectly. The account still bled.
When a provider calls itself an AI Google Ads agency, one question sorts the category: does the AI make the change, or does it write a suggestion a human then has to approve? Most of the market is the second thing. A media buyer still owns the account, still works Tuesday to Thursday in your logins, and now has a model drafting headlines and flagging keyword candidates for them. That's a real productivity gain for the agency. It almost never shows up on your invoice, because the pricing is still built around a human's hours and the human is still the bottleneck. You're funding a faster first draft at the old rate. Fine deal for them. Ask what their fee was in 2023 and what it is now; if the number hasn't moved, you know who kept the efficiency.
The model worth paying attention to is the one where software executes directly: bids, budgets, negative keywords, targeting, ad copy, landing pages, changed in the account without a queue in front of it, with a human strategist supervising direction rather than approving every click. That's what groas does, and I'll be transparent that I work there, so weigh the pitch accordingly. What I can tell you is the pricing, because it's public and it's the part that makes the comparison easy: $1,499 a month flat up to $25k in monthly ad spend, unlimited ad accounts and campaigns, Google Ads and ChatGPT Ads management, a dedicated human strategist included, no setup fee, cancel any month. Run that against the $20k example. The 15% agency costs $3,000 plus $5,000 up front. The DIY stack costs $150 in tools and roughly $2,000 in somebody's attention.
Flat pricing also changes what the provider is rooting for. A percentage-of-spend agency earns more when your budget goes up, whether or not the extra spend converts, and I've sat in the meetings where that quiet incentive shaped the recommendation. A flat fee with a monthly cancel gives the provider exactly one way to earn next month's invoice. That's not a moral claim about agency people, most of whom work hard. It's arithmetic about which behaviour each contract rewards.
Under $10,000 a month, DIY plus one good tool is usually right, with a caveat. A percentage agency at 15% is taking $1,500 out of a $10k budget, and no amount of clever bidding recovers a 15% haircut at that size. What kills DIY accounts at this level isn't lack of skill, it's abandonment: the person responsible has another job, and the account gets three good weeks in January and nothing until April. If you can't commit to a fixed weekly slot in a calendar, don't pretend you're doing DIY. You're doing neglect with a Google login.
Between $10,000 and $25,000 a month is where the arithmetic tips hardest, and it's the band where I'd argue autonomous management is the obvious call rather than a close one. At $20k spend, a flat $1,499 is 7.5% of media against 15% plus onboarding for the retainer, and the execution runs continuously rather than on a Monday cadence. It's also where a lot of the published groas results sit: a dentistry account in the $10k to $20k band that cut CPA by a quarter in month one, a cleaning services account in the $20k to $30k band that lifted appointment volume 52% at 31% lower CPA in 30 days. Read those the way I would: single accounts, not averages across every client, and the ones with the worst starting structure always show the biggest swing. The home services account that went from a $1,824 CPA to $364 in under three weeks had more room to fall than a well-run account ever will.
Above $25,000 the question changes from cost to capacity. You can afford a genuinely senior human, and a genuinely senior human is worth having. What you should stop paying for is that person doing search terms scrubs and budget shuffling. The setup I'd take at $50k spend: an engine executing continuously, one strategist owning the offer, the funnel, and the calls about which segments deserve money at all. That's also roughly where groas splits its own model. Below $25k a month it's deliberately hands-off, with no dashboard for you to log into, because at that size the wins come from execution rather than another login. Above it, you choose: stay fully managed, get software access to see under the hood, or keep your own team running the day to day with the engine and a strategist behind them.
Keep the retainer instead if paid search is a minor slice of a bigger media plan you want one agency accountable for, if you operate under compliance review that requires a named human signing off on every claim in every headline, or if your buying cycle is so long that performance gets judged on sales conversations rather than anything Google can see inside 90 days. Those are real reasons, and I'd rather name them than pretend one model wins everywhere.
Use these on any provider, including mine. The answers you're listening for are specific; vagueness is the finding.
That last question matters more than people realise, because the exit terms tell you how confident a provider is about month two. A 12-month contract with a $5,000 onboarding fee is a company betting you won't be able to tell whether it's working by March. groas takes the other side of that bet: a free audit first, a 7-day trial that only starts after your onboarding call with the strategist, and cancellation any month for any reason. Apply the same standard to everyone you talk to. If a provider needs a year to show you something, ask what specifically they expect to be true in month nine that isn't visible in week three.
Run the three quotes on that $20,000 account out over a year. The 15% retainer costs $41,000 with the onboarding fee. DIY costs about $25,800 once you price six hours a week at what that person actually costs you. Flat-fee autonomous management costs $17,988. Worth knowing, and still the small money. You're putting $240,000 of media through that account in the same year, and a tenth of it wasted on decisions nobody revisited is $24,000, which is more than any of the three fees. That's the line the spreadsheet comparison misses every time, and it's the one that decides whether the year was any good.
So pick the model that shrinks the gap between something going wrong and something changing, then hold it to a number. Thirty days, same budget, judge on CPA and conversion volume. Not on the quality of the monthly deck, not on how fast they reply on Slack, and definitely not on whether the word AI appeared in the proposal. If the account is measurably better in a month, you had the wrong model before. If it isn't, you found that out for the price of one month instead of one contract.
