GOOGLE ADS KEYWORD BIDDING: WHEN I STOP SETTING BIDS BY HAND

I still open accounts where every keyword carries a hand-set max CPC, adjusted weekly, sometimes for years. And I open accounts where someone flipped a campaign to Target CPA in week two and has been apologising to the client ever since. Both are the same mistake: treating google ads keyword bidding as an identity choice rather than a function of how much data the account has produced.

The question I actually answer for each campaign is narrower and more useful: at what point does the algorithm know more about this auction than I do? Below that point my hands are worth something. Above it, they're noise. What follows is where I've found that line across the accounts I manage, the sequence I use to cross it, and what the failures taught me that the successes didn't.

What manual keyword bidding google ads actually gives you

Manual CPC gives you one thing: a ceiling per keyword, applied uniformly to every auction that keyword enters. That's the whole of it. Everything else people attribute to it control, predictability, protection from waste is either a side effect of that ceiling or an illusion built on top of it.

It's worth being precise here, because a lot of the debate happens between people describing different things. When you place bids manually you are not choosing what you pay. You're choosing the maximum you'd tolerate paying. The actual price is decided in the auction, discounted by the thresholds you clear.

The asymmetry that decides everything

A manual bid is a single number covering thousands of auctions that differ from each other in ways I can't see. Google's documentation on Smart Bidding lists what's available at auction time device, physical location, location intent, weekday and time of day, remarketing list membership, browser, operating system, and the actual query text rather than just the matched keyword. I can replicate perhaps four of those with bid adjustments, statically, and only in the combinations I've thought to test.

That asymmetry doesn't matter much when volume is thin. With 40 clicks a week on a keyword, the signal combinations the algorithm would exploit haven't occurred often enough to be learnable, and my flat ceiling costs me almost nothing. It matters enormously at 400 clicks a week, where the same flat number is paying identical amounts for auctions with wildly different conversion probability.

Where the bid actually gets decided

Your bid amount is one input among several. Ad Rank combines it with expected click-through rate, relevance, landing page experience and context which is why two advertisers pay different prices for the same term while naming identical maximums. Ad quality doesn't just move you up the page; it discounts what you're charged to get there.

This matters for the manual-versus-automated question more than it first appears. If a campaign is paying a quality premium, no amount of careful hand-tuning fixes it you're optimising the one variable that isn't the problem. I've recovered more budget on inherited accounts by fixing relevance than by fixing bids, and the relationship between weak quality signals and what you pay is usually visible within an hour of opening the account.

Where manual still wins

Two situations, consistently. First, accounts with no conversion tracking worth trusting automation optimising toward a broken signal is worse than no optimisation at all. Second, the first three to four weeks of a genuinely new account, where I use manual bids as a discovery instrument rather than a performance tool: flat bids across a wide keyword set produce a clean read on which terms have any commercial life, without an algorithm concentrating spend before I know what's there.

My four thresholds for handing over the bids

I don't switch on a calendar. I switch when a campaign clears these four, and I've stopped switching when it clears only three that's where most of the failed migrations I've inherited went wrong.

Threshold 1 Conversion volume, measured at campaign level

My working floor is 30 conversions in 30 days per campaign for conversion-count strategies, and closer to 50 when I'm optimising toward value or a return on ad spend target. Below that, the strategy spends its life in a learning state that never resolves, and every weekly fluctuation reads as a trend. I've tried it under the floor on three accounts because a client insisted. All three took longer to stabilise than they would have taken to accumulate the data first.

Threshold 2 Conversion tracking I'd defend in a meeting

Not "tracking exists" tracking whose numbers I've reconciled against the client's CRM or backend within an acceptable margin. An automated bidding strategy pointed at a bad signal doesn't underperform quietly, it scales the error, and every automation failure I've diagnosed at length turned out to be a measurement failure underneath. Duplicated conversions, form-view events counted as leads, a thank-you page firing twice on refresh. The algorithm optimises faithfully toward whatever you tell it success is, and it does so much faster than a human would.

Threshold 3 A conversion lag shorter than my patience

If the median time from click to conversion is 21 days, the strategy is steering on a month-old picture, and so am I. On long-lag accounts I either hand it a well-correlated micro-conversion a qualified demo request rather than a closed deal or I stay manual longer than the volume threshold alone would suggest.

Threshold 4 Structural stability

I don't hand over a campaign I'm about to restructure. Splitting ad groups, changing match types, or rewriting the keyword set mid-migration means the learning period restarts and I can't attribute the result to anything. Structure first, automate second the reverse order is the most common sequencing error I see, and it's why the manual versus automated comparison so often comes out inconclusive when people run it themselves.

Structure in this context means something specific: the keyword set is settled, the match types are deliberate, the negatives are current, and the ad copies in each group are actually about the terms in that group. If any of those is still moving, the campaign isn't ready regardless of its volume.

The threshold I deliberately ignore

Account age. I've moved six-week-old accounts to automation because they were producing 200 conversions a month, and left three-year-old accounts on manual because they produce eleven. Volume and measurement quality decide it; tenure is decoration. The same goes for spend level a €30,000 account with poor tracking is a worse candidate than a €4,000 account with clean tracking.

The sequence I use when I hand keyword bidding over

The switch itself takes two minutes. Everything that makes it survive happens before and after it.

Step 1 Record the baseline before touching anything

Trailing 30 and 90 day CPC, conversion rate, CPA, and impression share, at campaign level, exported to a sheet. Without this I'm arguing from memory in four weeks' time, and memory always favours whichever direction I was rooting for. I add a note of anything unusual in the baseline window a stockout, a holiday, a competitor going dark because those explain more anomalies than people expect.

Step 2 Set the target from actuals, not from ambition

I set the initial tCPA or tROAS at the campaign's trailing 30-day actual, not at the number the client wants. Starting 40% tighter than current performance doesn't produce that performance; it produces a campaign that stops spending while it hunts for auctions that clear an unreachable bar. I move the target afterwards, 10–15% at a time, with two weeks between moves.

If the campaign has no reliable target to anchor to, I'll start with Maximize conversions and no target at all for the first fortnight, purely to establish what the campaign does when it isn't constrained. That number becomes the anchor for everything after.

Step 3 Migrate one campaign, not the account

One campaign with adequate volume goes first, and I keep a comparable one on manual as a control. It's imperfect the two campaigns aren't identical but it beats the alternative, which is switching everything at once and having no reference point when the account softens for unrelated reasons. Drafts and experiments do this more rigorously when the campaign has the volume to split.

Step 4 Leave it alone for the learning period

Two weeks minimum, and I don't touch targets, budgets, or the keyword set during them. The urge to intervene at day four is strong and I've given in to it enough times to know exactly what it costs: the period restarts and I've learned nothing except that I'm impatient.

What I do keep doing daily is the search terms report, because negatives remain my job regardless of who decides the price. Handing over bid control doesn't hand over query control if anything it raises the stakes on keeping the traffic clean, because the strategy will happily spend more on a bad query that happens to correlate with a tracked conversion.

Step 5 Judge on the right window and the right metric

Thirty days minimum, judged on conversions and cost per conversion, never on CPC. Almost every successful migration I've run shows CPC rising in week one the strategy is paying more for auctions it has identified as better. A client watching CPC alone will call it a failure on day three. I show them the CPA line and the conversion line, and I frame the whole comparison the way I would for any decision about how to bid on keywords in the account: cost per outcome, not cost per click.

Step 6 Reassign your own time

The hours that went into bid adjustments don't disappear, they move. Into search terms and negatives, into asset testing, into the page people arrive on, into feeding better conversion signals back to the system. Advertisers who hand over the bids and then do nothing with the recovered time see the smallest gains, which makes sense the automation replaced their weakest lever, not their strongest.

In practice my weekly routine on an automated account looks almost nothing like my weekly routine on a manual one. Less time tuning specific keywords in the bid table, considerably more in the search terms report and in whatever happens after the click.

What I infer when the switch goes badly

Volume collapses immediately: the target is unreachable, not the strategy wrong. The campaign has decided almost no auction clears the bar I set. I raise the target back toward the trailing actual rather than reverting to manual and concluding the whole approach failed.

CPC rises, conversions rise, CPA holds: the migration is working. This is the normal shape and I've learned to warn clients about it in advance, because it looks alarming for about ten days and generates the email that kills more migrations than any performance problem does.

CPC rises and conversions don't: the conversion signal is unreliable. The system is optimising confidently toward something that isn't real revenue. I go to tracking before I go to targets this is the single most common diagnosis I make on failed automation, and it's almost never what the client expects to hear.

Performance is erratic month over month with no pattern: the volume floor was never really cleared. The campaign is oscillating because each fortnight is a small sample, not because the strategy is unstable.

It works, then degrades after a structural change: I broke it. Adding a large keyword block or splitting the campaign resets what the strategy had learned. I now stage structural changes and bidding changes at least a month apart, and I write down which one I did when.

Ad performance drops in the same week as an unrelated creative refresh: attribute nothing. Two variables changed, so neither result is readable. I stage them apart now.

Manual outperformed over 60 clean days: believe it, and ask why. Uncommon but real, and in my experience it points to thin volume, a long lag, or a niche where human knowledge of which terms are commercially serious genuinely exceeds what the account's data expresses.

What I stopped doing

Adjusting keyword bids weekly on campaigns with real volume. I audited a year of my own adjustments on one account and the net effect was inside noise. I was busy, not effective, and the busyness was legible to the client in a way the results weren't.

Layering aggressive manual bid adjustments underneath an automated strategy. Device and location are already being used at auction time. My static multipliers on top were second-guessing a system with better inputs than mine, using a coarser instrument.

Switching strategies when a campaign has a bad fortnight. Every switch costs a learning period. Two switches in a quarter means the campaign spent the quarter learning rather than performing, and the account ends up with no clean read on either approach.

Treating the decision as permanent in either direction. I've moved campaigns back to manual when volume dropped seasonally below the floor, and forward again when it recovered. It's a threshold, not a conversion experience.

Promising clients that automation lowers CPC. It frequently raises it. What it should lower is cost per conversion, and that's the only promise I make now in writing, before the switch, with the expected week-one shape described.

Building narrow single-keyword groups purely to control bids. That structure existed to give me bid granularity I no longer set by hand. Keeping keywords relevant to their group still matters for relevance and for the ad copy; splitting them to the atom purely for bid control stopped earning its maintenance cost.

The practical takeaway

Bidding by hand is worth doing exactly as long as your knowledge of the account exceeds the account's own data and that crossover happens at a measurable point, not a philosophical one. Thirty conversions a month, tracking you'd defend in a meeting, a lag you can live with, and a structure you're not about to change: clear all four and my hands add nothing that auction-time signals don't add better.

Before you clear them, the opposite holds just as firmly. A thin ads campaign handed to automation early doesn't get smarter, it gets erratic, and the erratic period costs real money while it resolves.

If you take one thing from the migrations I've run: the campaigns where automation failed almost never failed because the algorithm was wrong. They failed because someone handed it a bad conversion signal, an impossible target, or a campaign too thin to learn from and then blamed the strategy for doing precisely what it was told to do.