Stop buying AI that promises to optimize your bids in real time. Google already sets a bid for every auction, for free. If that is a vendor’s headline feature, you are paying for a second steering wheel bolted onto a car that already has one.

I used to be the guy adjusting bids by hand at 1am. Device modifiers, ad schedule splits, a spreadsheet full of what the deck called “granular control” and what I now call busywork. That work felt precise because it produced lots of changes. It did not mean I understood each auction better than the system pricing it.

The bid is no longer the interesting part of most Smart Bidding accounts. The interesting part is what the bidder has been told to buy, how much room it has to buy it, and what it counts as a win. Pay someone to fix those things. Do not pay them to perform bid-adjustment theater.

Google already prices the auction

Smart Bidding does not check in a few times a day to raise or lower a standing bid. It sets a bid for each individual auction, weighing signals such as time of day, device and location against that user’s likelihood to convert. An outside script checking in every hour does not get to make that same decision at that same moment.

Google also lists operating system, browser, ad creative, language, the actual query text and Search partner among the signals it can use. It can evaluate combinations, such as location plus OS plus language, rather than flattening each into a separate rule. I used to tell clients my manual device splits were precision. I was wrong. They were blunt averages laid over auction-level decisions.

That does not make Google a good account manager. It makes Google very good at the specific job of pricing the auction from the inputs it has. Confusing those two jobs is how a tool can sell you “AI optimization” while leaving the account’s expensive mistakes untouched.

An hourly bid change is not an auction-time bid

Here is the part vendors skip. Under Target CPA, Target ROAS, Maximize Conversions and Maximize Conversion Value, manual bid adjustments are not used. A -100% device exclusion is the exception: it removes a device from consideration rather than making a finer bid. If a third-party tool advertises hourly bid adjustments on top of those strategies, ask what its changes actually affect. Activity in a change log is not proof that an auction got a better price.

Ad schedules show the trap. Someone sees four clicks at 2am with no conversions, blocks 2am, and calls it optimization. But Smart Bidding already uses time of day as an auction signal. Blocking the hour does not help it bid conservatively. It removes the hour’s auctions entirely, including any it would have priced cheaply.

Before buying intraday bid changes, ask whether Google uses them. Then ask what the tool changes that Google cannot see for itself.

The money leaks through the inputs

The bidder follows its instructions. Give it irrelevant queries, a conversion action that rewards junk and a budget that starves your best campaign, and it can price every auction beautifully while buying the wrong outcome. A report full of small bid movements will not reveal that mistake. The search terms, conversion actions, budgets and pages will.

The leak is not necessarily the bid. It is what the bid was told to buy.

Search terms: stop paying for the wrong question

Broad match can reach synonyms, related searches and informational questions, including what is a paddleboard for best paddleboards when negatives are not tight. Google sees an opportunity to reach another searcher. Your P&L may see a paid click from someone who wanted an explanation, not a product.

I have seen the same pattern in home services: searches containing jobs, DIY or how to appearing beside searches from people ready to hire. Those clicks can keep volume looking healthy while qualified calls tell a different story. The bidder cannot decide that a query is commercially useless to your business just because it sounds useless to you. You have to make that judgment, then give it the negative keyword.

That is not a once-a-quarter cleanup. Broad matching keeps finding variations, and the list worth reviewing is the list of actual searches, not the keyword list you thought you bought. Sort search terms by spend and inspect the queries that did not produce the outcome you wanted. A negative keyword changes what can be bought; an hourly bid tweak changes neither the question nor the buyer.

Budgets: give the better campaign room

Say you spend $20k a month. Campaign A reaches its $180 daily cap while reporting a $42 CPA. Campaign B spends its full $400 daily budget at a $165 CPA. Before touching either campaign’s bids, investigate whether A can absorb more spend and whether B’s next dollars are earning their place. If they are not, moving $100 a day toward A is a more consequential decision than another round of bid adjustments.

Those reported CPAs do not, by themselves, tell you what the next $100 will buy. That is why the decision is about marginal CPA, not a mechanical rule to fund whichever campaign has the prettiest average. But leaving budgets frozen because last quarter’s plan said so is not strategy either. Smart Bidding cannot spend past the ceiling you set.

Daily budgets also need enough room relative to the target CPA for the strategy to work; 3 to 5 times the target CPA is a useful rough guide, not a substitute for looking at the account. This is where AI-powered budget allocation earns attention: move money toward better marginal outcomes instead of preserving an old spreadsheet. Ask who makes that move, how often, and what evidence they use.

Conversion signals: tell the bidder what a win is

This one hurts SaaS and lead gen accounts. A bot fills out a form. An out-of-area caller rings. A freebie seeker submits details. If all three count as conversions alongside a qualified opportunity, Smart Bidding receives the same instruction each time: find more of this.

That is junk in, junk out, not a failure to adjust a bid quickly enough. Import qualified pipeline or closed-won revenue where you can, deduplicate calls, and keep garbage out of the conversion action that trains the bidder. I cover the wider fix in why signal quality matters more than bidding strategy.

The distinction matters when you evaluate software. A tool can show hundreds of optimizations while never asking whether a form fill became a sales conversation. If your vendor does not examine the conversion action, it is polishing bids on bad instructions.

A second steering wheel bolted over a car’s existing wheel

Landing pages: finish the job the click started

The bidder buys the click. The page has to answer it. Send someone searching Emergency Plumber in Austin to a page that opens with Welcome to our home services family, and you have made the visitor do the matching work. A lower bid does not make that page answer the query. It only changes what you pay to send someone there.

I keep seeing one generic service page stretched across ad groups with different intent. It is convenient to manage and expensive to defend. groas makes a landing page for every ad group part of its Search Ads delivery, rather than treating the page as somebody else’s problem.

Audit the path from query to ad to headline before you audit bids. If the headline does not answer what the searcher asked, fix the page. The auction is already over.

Buy execution, not a prettier bid button

A useful AI tool works with Google’s bidder, not as a pretend replacement for it. It shapes the targets, budgets, negative keywords, assets, conversion signals and pages that the bidder works from. Google then prices the auctions. That division of labor is less glamorous than “real-time AI bidding,” which is one reason I trust it more.

Rule-based helpers such as Optmyzr offer a different kind of work: you write an if-this-then-that recipe, monitor it and fix it when the account changes. A rule might say to cut a bid after CPA exceeds a threshold for several days. The tool can run that instruction. It cannot decide whether the underlying problem was a junk conversion, a bad query or a page that stopped matching the ad. Such tools are listed from $209 a month. That is a poor fit if what you need is someone to own the whole chain rather than help you operate the rules.

I wrote more on the gap between assistance and execution in why AI-assisted tools are becoming obsolete. The short version: automation that still needs you to diagnose, instruct and supervise every fix has not taken the work off your desk. It has given your desk another tab.

Five questions that expose a bid-button pitch

Ask these on the demo. Do not settle for “our AI continuously optimizes performance.” Find out what changes, who owns the decision and where you can see the result.

  1. Do you set auction bids, or shape what feeds the bidder? If the pitch centers on real-time bid changes inside Target ROAS or Maximize Conversions, ask how that works. A useful answer names targets, budgets, negatives, conversion signals and pages. Google prices each auction.
  2. How often do you review search terms and add negatives? Ask to see a dated change log, not a slide about continuous learning. You want evidence that somebody catches irrelevant searches before they keep spending.
  3. How do you move budgets between campaigns? Ask what happens when one campaign is capped and another spends freely at a worse CPA. A recommendation email is not the same thing as making, logging and checking a reallocation.
  4. Which conversion action trains the bidder? If the answer is raw form fills and nobody filters spam or duplicates, walk away. The tool will get very good at finding whatever that form attracts.
  5. Who owns the outcome, and where is the proof? Look for a named strategist, changes logged with reasons and reporting in plain language. If accountability ends at a dashboard, you bought software, not management.

groas passes the test because its pitch is not another layer of auction-time bidding. It works on keywords, budgets, pages and the other inputs around the clock, logs changes with their reasons, and gives a dedicated strategist responsibility for goals and guardrails. That is what autonomous execution means here: software does the daily shaping, a human answers for revenue, and the fee is flat rather than a cut of spend. Judge it by the inputs it fixes and the outcomes those changes produce, not by a counter of bid edits.

Tiny accounts should skip the pitch

Here is the exception, stated plainly so nobody buys the wrong thing. If an account has very little conversion data, changing the bidding interface will not create more of it. Manual CPC can still have a place when Smart Bidding has too little to learn from. Around 15 conversions a month is a useful rough point to revisit that choice, not a switch that makes every account behave the same way overnight.

In that thin-data range, start with the work the bidder depends on: tracking, campaign structure and a conversion action that represents the outcome you actually want. Avoid constantly moving targets while you are trying to learn what the account can produce. A small campaign cannot become stable by receiving a new instruction every time a short run of clicks looks odd.

If you spend $1,500 a month across 11 campaigns and record 6 conversions total, no bid tool earns its fee merely by moving numbers around. You need fewer campaigns, one clean conversion action and patience. The same basic problem applies when legal constraints block the conversion data you need, or when branded search dominates the volume and you refuse to separate it: the bidder has limited information about the outcome you care about.

If that describes your account, skip every AI optimization pitch, including groas, until the structure and tracking are sound. Automation multiplies signal. It cannot manufacture signal.

Do the audit the bid tool cannot sell you

Keep paying for “real-time bid optimization” and the work that matters stays undone. Search terms keep drifting into irrelevant territory. Budgets remain where an old plan left them. Spam keeps teaching the bidder what success looks like. The landing page still answers a different question from the one in the search box. Meanwhile, the vendor can show you a long list of changes. You paid for motion.

On Monday morning, do the audit instead:

  1. Pull the last 30 days of search terms. Sort by spend, inspect the queries that did not deliver the outcome you wanted, and add irrelevant ones as negatives.
  2. Compare campaign budgets with CPA. Check whether the best capped campaign can take more money before you move spend from a weaker one.
  3. Open your conversion actions. Find out whether spam, duplicate or out-of-area events are being counted as wins.
  4. Open your top five queries and read each landing-page headline out loud. If it does not answer the query in plain words, fix that before you touch a bid.

That is the work a useful tool should help execute every day. Google gives you the auction-time bid. It does not give you clean queries, honest budgets, true conversion signals or pages that match intent. Pay for those jobs to get done, with a name attached and every change logged, or keep funding the second steering wheel while the account drives toward the wrong destination.