.png)
Here's a conversation I've had more times than I can count. Someone's paying $4,000 a month to an agency, the performance chart has been flat for two quarters, and every monthly call ends with the same words: "the market's competitive right now, but we're optimizing." They want to know if they're being managed or being milked. And they've started hearing that AI can do the same job, so they want to know if that's real or just the next version of the sales pitch they're already tired of.
So let me be blunt about the thing most agencies won't say out loud. A $4k retainer buys you a fixed number of human hours per month, and after the first 90 days, most of those hours go to reporting, not optimization. I spent the better part of a decade on the other side of that arrangement, building match type structures by hand and mining negative keywords at 1am, and I can tell you the honest truth: a huge share of the work that gets billed as "strategy" is mechanical maintenance a spreadsheet could supervise. That doesn't automatically mean your agency is ripping you off. Sometimes a plateau is genuinely the market. The problem is that from where you sit, neglect and normal conditions look identical. This piece is about telling them apart, and then, if you decide to switch, knowing exactly what the first 30 days feel like so you're not walking in blind.
Start with the data, not the vibe. Pull your last twelve months of monthly conversions, CPA, and impression share. A real market plateau has a signature: your impression share is already high (say 70%+ on your core terms), your CPCs are creeping up across the board, and your competitors' ads have visibly multiplied. That's a ceiling you hit because there's only so much demand to buy. Neglect has a different signature entirely. Neglect looks like a search terms report full of junk you're still paying for, ad copy that hasn't changed since onboarding, one Smart Bidding strategy that got set and forgotten, and a landing page that's identical for every keyword regardless of what the searcher actually typed.
The tell I trust most is the search terms report. Ask your agency for the last 90 days of it. If you see garbage queries you're clearly paying for and nobody's added them as negatives, that's not a competitive market. That's someone not doing the 1am work. When I audited accounts, wasted spend on irrelevant terms was routinely 20-30% of budget, and it's the single fastest thing to fix. If that report is clean and tight, give your agency more credit. If it's a swamp, you have your answer.
Before you fire anyone, ask five specific questions and listen for specific answers. One: show me the search terms report and the negatives you've added in the last 90 days. Two: what changed in the account last month, action by action, not "we optimized bids." Three: which landing pages are we sending paid traffic to, and do they match the ad? Four: what's our impression share on our top ten terms, and where's the headroom? Five: if I gave you another $1,000 a month, what specifically would you do with it? Vague answers to specific questions tell you everything. If they can't name what they did, they probably didn't do much.
Let's do the math nobody puts in the deck. At $4k/month you're spending $48k a year on management. If your ad spend is, say, $20k/month, that management fee is a 20% tax on your media before a single click converts. For that, you're buying the output of a human, or more often a junior media buyer offshore, who can physically touch your account a few hours a week. That's the real constraint. A person can only run so many manual A/B tests, review so many search terms, and rebuild so many landing pages in a week. You're paying full rate for that ceiling.
Autonomous PPC changes what's doing the work, not whether work gets done. The reason I ended up writing about this instead of doing it by hand is that the mechanical layer, the bids, budgets, negative keywords, and copy tests, is exactly the part a model trained on enormous volumes of spend does better than a tired human at 1am. groas, the platform I work with, was built by people who ran a recommendations tool on hundreds of client accounts and realised the human clicking the buttons was the bottleneck, so they trained models on $500B+ in profitable ad spend to execute every action directly, around the clock. The part I'd push back on if anyone oversold it: this doesn't mean humans add zero value. It means humans should be doing the judgment work, the offer strategy, the "should we even be bidding on this segment" calls, not the button-clicking. groas keeps a senior strategist on top of the engine for exactly that reason.
One thing worth being clear-eyed about on cost. groas charges a flat monthly fee with no percentage of ad spend, no setup fee, and no long-term commitment, month to month. That structure matters more than the headline number, because a percentage-of-spend model quietly rewards your agency for spending more, not converting better. When the fee is flat and the contract is monthly, the only way to earn next month is to perform this month. That's the incentive alignment the old model never had, and it's the reason the 12-month comparison usually isn't close once you factor in what you stop wasting on unmanaged spend.
The reason people stay with a bad agency isn't loyalty. It's fear of the handover. Everyone's heard the horror story where someone switched, performance cratered for a month, and they crawled back. So here's the honest week-by-week, because knowing the shape of it is most of what de-risks it.
Week 1 is audit, access, and goals. With groas the front door is an application, not a signup: you book a short call, they review your account and goals, and if it's a fit you get a flat quote and an onboarding date. Most accounts are live within the first week. What's happening under the hood is the unglamorous part that matters most: connecting the account, mapping your real conversion actions (not the vanity ones), and dropping the tracking snippet on your site directly or via GTM. This is also where you catch the landing page problem. groas takes your existing page and deploys dynamic versions that adapt to each search intent, so the person searching "emergency plumber near me" and the one searching "water heater install cost" don't hit the same generic page. That alone tends to be where the neglected accounts were bleeding conversions.
Week 2 is the learning period, budgets, and guardrails. This is the part people panic about, so set expectations now: any change to a Google Ads account goes through a learning phase, and Smart Bidding's version of it is roughly two weeks of the algorithm feeling around, often at your expense. It's the single most common reason a fresh setup looks worse before it looks better. Autonomous execution shortens the pain because the engine isn't waiting for a human to notice a bad signal and react on Tuesday's check-in, it's adjusting continuously. But it doesn't repeal physics. Expect week two to be noisier than week four, set your guardrails on budget caps and target CPA before it starts, and don't judge anything by the day-three numbers. If someone promises you no learning period at all, they're selling.
Weeks 3 to 4 are the first real optimizations, and where reporting should start earning trust. By now the junk queries are getting blocked, the cheaper high-quality traffic is getting found, and the copy tests are running in parallel rather than one-at-a-time. What you should see in your inbox is a weekly report of every action taken, every change, every win, in plain language. That's the standard I'd hold any autonomous service to: not a dashboard of metrics you have to interpret, but a record of what it actually did and why. If you can't tell what changed and what it cost or saved you, you've just swapped one black box for another.
There are two ways to make the handover, and the right one depends on your spend and your nerve. A clean handover means groas takes over the existing campaigns on day one. It's faster and it avoids splitting your conversion data across two setups, which matters more than people think: Smart Bidding needs signal density, and running two half-fed campaigns side by side starves both. For most accounts under $20k/month, clean handover is what I'd do. Parallel running, where you keep the old campaigns live at reduced budget while the new setup ramps, feels safer and occasionally is, but it costs more, muddies the data, and tempts you to declare a winner at day five off half the sample. If you parallel run, do it for a defined two-week window and commit to killing the loser at the end, not letting it drag on because you're attached to it.
Here are the metrics to actually watch in month one, and the ones to ignore. Watch CPA and conversion volume together, never one alone, because it's trivial to cut CPA by shrinking to only your brand terms and calling it a win while volume quietly collapses. Watch impression share on your core terms to know whether there's headroom or you're hitting the ceiling I described earlier. Watch the search terms report the same way I told you to audit your agency, that's your early-warning system for wasted spend. Ignore day-to-day CPC swings, ignore click-through rate as a standalone number, and ignore anything measured before the learning phase clears. The red flags that mean call someone: CPA still climbing at the end of week four with no explanation in the report, or a report that's gone vague. Specific accounts on the platform have posted things like CPA cut 27% or appointment volume up 52% at a lower CPA inside 30 days; your mileage will vary by industry and starting point, but the shape you want to see is volume holding or growing while CPA trends down by week four.
So back to the person on that flat-lined monthly call. If your agency's search terms report is clean, your impression share is already high, and they can name exactly what they did last month, you don't have an agency problem, you have a market you've mostly saturated, and switching won't conjure demand that isn't there. But if that report is a swamp, the copy hasn't moved in a year, and "we're optimizing" is the whole answer to every question, you're not paying for management. You're paying for the reporting on the absence of it. The switch is less frightening than the story in your head: one week to go live, two weeks of expected noise, and a written record by week four of what actually changed. The only real question is how many more $4k invoices you want to clear before you ask for it.
