Choosing a Google Ads bidding strategy is the decision most UK advertisers make least deliberately. It is usually made twice: once at setup, by accepting whatever the campaign creation flow suggested, and once again when an account representative or the Recommendations tab proposes switching to something automated. Neither moment involves anyone establishing whether the account has the conversion volume or the data quality for the strategy being adopted, which are the two things that determine whether it will work.
This guide covers that decision properly. It starts with how Smart Bidding actually reaches a bid, because the mechanism explains both its genuine advantage over a human and the specific circumstances in which it fails. It then covers the two prerequisites nobody checks, when manual bidding still outperforms automation and why, how to survive the learning period without intervening and making it worse, how to set targets that tighten efficiency rather than starving a campaign of volume, and how to match strategy to conversion volume and commercial goal rather than to whatever the interface recommends. Attribution and conversion data quality are the foundation all of this rests on, and we cover those separately in our guide to Google Ads attribution.
How Smart Bidding actually reaches a bid
The mechanism matters because it explains the whole decision. Smart Bidding sets a bid at auction time, for each individual auction, based on a prediction of how likely that particular query from that particular person is to convert.
The signals it evaluates at that moment include device, physical location and location intent, time of day and day of week, operating system and browser, language, whether the person is on a remarketing list, the specific search term rather than just the matched keyword, and characteristics of the ad and landing page. It combines these into a conversion probability estimate and bids accordingly — higher where conversion looks likely, lower or not at all where it does not.
That is something a human genuinely cannot do. A manual bidder sets one bid per keyword, perhaps adjusted by device and schedule, and that bid applies equally to a promising query and an unpromising one. Auction-time bidding differentiates between them at a granularity no manual structure can express. Where there is enough data for the prediction to be reliable, this is a real and substantial advantage, and it is why automation has become the default rather than a fashion.
What it cannot see
The system optimises toward the conversions you tell it about, weighted by the values you give them. It has no visibility of anything outside that. It does not know your profit margin, so a conversion worth £40 of margin and one worth £400 look identical unless you pass values. It does not know lead quality, so a time-waster who filled in a form counts exactly as much as a qualified buyer. It does not know your capacity, so it will happily generate enquiries you cannot service. And it does not know that your phone line is unstaffed at weekends, if telephone calls are what actually convert.
Every one of those blind spots is fixable by feeding better data in — values rather than counts, offline outcomes rather than form submissions, schedules that reflect reality. But they are blind spots by default, and an advertiser who switches to automation without addressing them has not automated their judgement; they have removed it.
The two prerequisites
Everything above depends on two conditions. The first is conversion volume sufficient for the prediction to be statistically meaningful. The second is conversion data that measures something the business actually values. Where either is missing, automation does not degrade gracefully — it optimises confidently toward the wrong thing, which is worse than a mediocre manual bid because it is harder to notice.
Before changing any bid strategy, audit what you are counting as a conversion. Open the Conversions view and look at every active action: what triggers it, whether it is set to count every conversion or one per click, and whether it is included in the “Conversions” column that bidding optimises toward. The common faults are counting page views as conversions, counting the same enquiry twice through two actions, and including low-value micro-conversions alongside genuine enquiries so the model optimises for whichever is easiest to generate. Fixing this takes an afternoon and it determines whether every subsequent bidding decision is built on something real.
Manual against automated bidding
The comparison below highlights the automated column, because for accounts that meet the two prerequisites it genuinely does outperform manual bidding and the gap is not close. The important qualification is that those prerequisites do the deciding: below meaningful conversion volume, or on unreliable conversion data, the manual column is the better answer and switching to automation makes the account worse rather than better.
Manual CPC
You set the bids, adjusted by rules
Smart Bidding
Auction-time bids against a target
The row that decides most real cases is the fourth. Manual bidding is indifferent to conversion tracking quality because it does not consult it — a manual campaign with broken conversion tracking is merely unmeasured, which is a reporting problem. Smart Bidding consults nothing else, so the same broken tracking becomes an optimisation problem: the system spends the budget pursuing whatever the faulty data rewards. This is the mechanism behind most accounts that got worse after switching to automation, and it is routinely misdiagnosed as automation being unsuitable for the business.
The management time row deserves a caution, because the way automation is sold implies it approaches zero. It does not. The work changes rather than disappearing: less bid adjustment, more attention to conversion data quality, search term review, negative keywords, target setting, and feeding values and offline outcomes back in. Accounts that switch to Smart Bidding and reduce management effort proportionally tend to drift, because the remaining work is the work that matters.
On control: the loss is real and it is frequently overstated. You retain control of budgets, targets, geography, schedules, audiences, keywords, negatives and what counts as a conversion — which is most of the levers that matter. What you give up is the individual bid, which is the one lever a human was worst at setting.
Where bidding decisions go wrong
The grid below groups the recurring failures we find when auditing UK Google Ads accounts. The badges reflect how much budget each issue typically wastes rather than how technically complex it is.
The first card is where the money actually goes. An account counting page views as conversions will report a flattering cost per conversion and, under Smart Bidding, will actively pursue the traffic most likely to produce page views — which is not the traffic most likely to produce customers. The same logic applies to double-counting: if an enquiry fires two conversion actions, the reported cost per conversion is half the real figure, and any target set against it is twice as aggressive as intended.
The third card’s second row is the most common self-inflicted wound in automated bidding. An advertiser currently achieving a £120 cost per acquisition sets a target of £60 because that is what the business would like. The system responds rationally: it bids only where it predicts conversion at or near £60, which is a small fraction of available auctions, so impression share collapses and volume disappears. The campaign is not broken and the target is not being missed — it is being met, on almost no traffic.
Auto-apply recommendations deserves its own mention. It allows Google to change your account automatically, including bid strategies and targets in some configurations. There are settings within it that are genuinely useful and others that will restructure decisions you made deliberately. Reviewing what is enabled takes five minutes and is worth doing on any inherited account.
Google Ads bidding in UK accounts — the numbers
The figures below reflect what we find when auditing UK Google Ads accounts spending between roughly £1,500 and £40,000 a month. They describe accounts that are actively managed rather than abandoned ones.
The first figure is the one that should be addressed before any bidding decision. In more than half of accounts, something is being counted that the business would not pay for: a contact page view, a brochure download treated as equivalent to an enquiry, a newsletter signup sitting in the same conversion column as a sales lead. Under manual bidding this distorts reporting. Under Smart Bidding it distorts spending, because the system will pursue whatever is cheapest to produce among the things you have told it to value.
The thirty-conversion figure needs an honest caveat. Google formerly published minimum conversion thresholds for automated strategies and has since removed them from the interface, and the strategies will run at lower volumes. What was removed is the requirement, not the statistics: a model predicting conversion probability from eight data points a month is working with very little, and the observable result is erratic bidding and targets that are hit in one fortnight and missed by a wide margin in the next. Treat thirty monthly conversions per strategy as the point below which you should expect volatility rather than as a rule you are breaking.
The last figure is where the largest untapped gain sits for UK lead generation advertisers. Fewer than one in five feed back what happened to the leads — which ones became opportunities, which closed, what they were worth. Without that, the system optimises for lead count, and lead count is not the objective. Feeding outcomes back turns Smart Bidding from a volume optimiser into something aimed at revenue, and it is the single change most likely to improve a lead generation account that already has clean conversion tracking.
What the strategies actually optimise for
The chart below shows how often we find each bid strategy in use across audited UK accounts. It is worth comparing against what each one is designed to do, because the distribution suggests a good deal of selection by default rather than by intent.
The second row is the one worth examining in your own account. Maximise conversions without a target will spend the entire daily budget pursuing the largest possible number of conversions, with no efficiency constraint at all. That is occasionally the correct choice — when you genuinely want maximum volume at whatever cost the budget implies, or during an initial data-gathering period. It is frequently not, and it is the default that campaign creation nudges you toward, which is why a quarter of accounts are running it.
The target ROAS row at eight per cent is low, and for many UK advertisers correctly so. Value-based strategies require conversion values that mean something, which for e-commerce is straightforward and for lead generation requires the offline feedback most accounts do not have. Adopting target ROAS without real values attached produces optimisation toward a number you invented, which is worse than optimising toward conversion count honestly.
The portfolio figure at six per cent represents a genuine missed opportunity. A portfolio strategy pools conversion data across several campaigns, which is precisely the remedy for an account where individual campaigns each sit below meaningful volume. Three campaigns with twelve conversions a month each will behave erratically on separate targets and considerably better on one shared strategy. It is a setting rather than a project, and most accounts that would benefit have never considered it.
Benchmarks — bidding practice against what we find
The figures below show how often each practice is present across audited UK Google Ads accounts. They describe operating discipline rather than account size.
Adoption of sound bidding practice in UK accounts
Eighty-eight per cent against forty-three per cent is the gap that matters most. Almost every account has conversion tracking; fewer than half are tracking only things the business would actually pay for. Since Smart Bidding consults nothing except that data, the second figure is the effective ceiling on how well automation can possibly perform across this population.
The learning period row at twenty-two per cent explains a great deal of reported dissatisfaction with automated bidding. Roughly four accounts in five intervene during the period when the strategy is calibrating, usually by adjusting the target because early performance looks alarming. Each adjustment restarts the calibration, so the account experiences permanent learning-period volatility and concludes that automation does not work for its sector.
Data exclusions at twelve per cent are worth knowing about specifically. If conversion tracking broke for four days, or a site outage produced a period of unrepresentative data, you can tell Google to disregard that window rather than let it poison the model. Similarly, seasonality adjustments let you signal a short, known spike — a sale, a campaign, a known event — rather than waiting for the system to discover it. Both are free, both take minutes, and almost nobody uses them.
Moving an account onto automated bidding
The sequence below is how to migrate an existing UK account from manual to automated bidding without the volatility that makes most advertisers reverse the decision. The ordering is deliberate: the data work happens before the strategy change, not alongside it.
The two phases people skip are weeks one and two, because they involve no visible change to the campaigns. Skipping them is what produces the common pattern of switching to Smart Bidding, seeing performance deteriorate, and concluding the account is unsuitable for automation — when the actual cause was that the system was handed a target derived from a cost per acquisition that was measured wrongly.
The step-size discipline in weeks seven to twelve matters more than it looks. Large target changes are treated as significant strategy changes and can trigger renewed learning, so a single aggressive tightening both destabilises the campaign and obscures whether the target was achievable. Ten to fifteen per cent at a time takes longer and converges reliably.
Automation readiness — where most UK accounts sit
Combining the assessment areas gives an indication of how much value an account would get from automated bidding today, as opposed to after some preparation. A low score is not an argument for staying manual permanently; it is an indication of what to fix first.
A score in the high thirties usually has a consistent shape. Tracking exists, which is why the score is not lower. Data quality scores poorly because something is being counted that should not be. Value differentiation scores badly because almost nobody does it. Volume scores variably depending on account size. Operating discipline scores worst, because the learning period is routinely interrupted and targets are routinely set from ambition.
What makes this benchmark unusual is that the remedies are almost entirely free. Cleaning up conversion actions costs an afternoon. Setting a target from measured performance rather than aspiration costs nothing. Respecting the learning period costs patience. Reviewing auto-apply settings takes five minutes. The only item requiring real work is offline outcome feedback, and that is also the item with the largest upside.
The caveat: an account spending a few hundred pounds a month with two or three conversions is not going to reach a high score and should not try. Below meaningful volume the honest answer is manual bidding or maximise clicks with a sensible bid cap, tight keyword control, and acceptance that sophisticated optimisation requires data the account does not generate. Trying to run target CPA on four conversions a month produces worse outcomes than a competent human setting bids by hand.
Setting targets without starving the campaign
Target setting is where most automated bidding fails after the data problems are resolved, and the failure is almost always in the same direction.
Start from measurement, not ambition
Set the initial target at the cost per acquisition you are actually achieving. This feels pointless — why automate if not to improve? — and it is the step that makes improvement possible. The system needs to establish stable operation before it can be pushed, and a target it cannot reach produces a campaign that simply declines to bid.
Understand what an aggressive target actually does
A target well below achievable performance does not make the system try harder. It narrows the set of auctions where the predicted cost per conversion falls within the target, and the system bids only in those. Impression share collapses, volume falls away, and the reported cost per acquisition may well hit the target — on a fraction of the traffic. The campaign looks efficient and delivers almost nothing, which is the most misleading failure state in the platform because every visible metric except volume improves.
Watch impression share as the constraint signal
As you tighten in steps, impression share is the indicator to watch. While it holds roughly steady, the tightening is genuine efficiency gain. When it starts dropping materially, you have reached the point where further tightening trades volume for efficiency. Whether that trade is worth making is a commercial decision about whether you would rather have fewer cheaper customers or more expensive ones — but it should be a decision rather than a surprise.
For value strategies, the values must be real
Target ROAS only works if the conversion values reflect something true. For e-commerce that is usually transaction value and is straightforward. For lead generation it requires estimating what different lead types are worth, ideally informed by actual closed business. Inventing values to enable a value-based strategy produces confident optimisation toward an arbitrary number, and the resulting apparent return on ad spend is a measurement of your own assumptions.
One further practical note: budget and target interact. A target CPA campaign constrained by too small a daily budget cannot reach the volume at which its predictions stabilise, so thin budgets and automated targets combine badly. If the budget is genuinely limited, fewer campaigns with adequate budget each will outperform many campaigns each starved, and this is another argument for consolidation in smaller accounts.
When manual bidding still wins
Automation is the right default for accounts that meet the prerequisites, and there remain circumstances where a competent human outperforms it. These are not edge cases; several are common among UK SME advertisers.
Genuinely low conversion volume
Below roughly thirty conversions a month per strategy, the prediction has very little to work from and behaves erratically. A human with category knowledge, setting bids against keywords they understand, will frequently do better — not because they bid more accurately per auction, but because they are not making confident predictions from eight data points. Consolidating campaigns or using a portfolio strategy to pool data is the first thing to try; manual is the fallback when pooling is not possible.
Brand campaigns
Bidding on your own company name is a different problem. Intent is already extremely high, competition is usually limited, and the objective is to appear cheaply rather than to bid up to a target cost per acquisition. Smart Bidding will sometimes pay considerably more than necessary for traffic that was going to convert regardless, because it correctly predicts high conversion probability and bids accordingly. Manual CPC with a modest bid, or target impression share, generally holds the position for less.
Unreliable or unfixed conversion data
If tracking is broken, contested or in the middle of being rebuilt, manual bidding is the safer state. Manual is merely unmeasured; automated bidding on faulty data actively pursues the wrong outcomes with the full budget. Stay manual until the data is trustworthy, then switch — in that order, never the reverse.
Knowledge the system does not have
Where you know something material that is not in the data, manual control lets you act on it. A trade business whose phone is unstaffed after 5pm and at weekends, where calls are the conversion, knows something Smart Bidding will only discover slowly and imperfectly. The better answer is usually to encode that knowledge as schedules and exclusions and still automate, but where it cannot be encoded, manual control is legitimate.
The table below summarises which strategy suits which situation. It is a starting point rather than a rule, and the volume column is the constraint that overrides the rest.
| Situation | Reasonable strategy | Conversion volume needed |
|---|---|---|
| New campaign, no conversion history | Maximise clicks with a bid cap, or manual CPC | None — gathering data |
| Established, adequate volume, cost efficiency is the goal | Maximise conversions with a target CPA | Around 30+ monthly |
| E-commerce with real transaction values | Maximise conversion value with a target ROAS | Around 50+ monthly |
| Several small campaigns, each below threshold | Portfolio strategy pooling their data | 30+ combined |
| Brand terms | Manual CPC or target impression share | Not applicable |
| Very low volume, or tracking not yet trustworthy | Manual CPC with tight keyword control | Not applicable |
Note the first row. A brand new campaign has no conversion history, so an automated target has nothing to predict from. Running maximise clicks with a sensible bid cap for the first few weeks gathers the data that makes a subsequent switch viable, and is a more honest approach than setting a target CPA on day one and interpreting the resulting volatility as a problem with the keywords.
UK cost benchmarks that frame the decision
Bidding strategy interacts with what clicks cost, because a sector with expensive clicks reaches meaningful conversion volume at a much higher spend. The indicative UK figures below are for 2026, exclude VAT, and vary widely by competitiveness and location — treat them as a basis for sizing rather than as quotations. Fuller cost context sits in our guide to Google Ads cost in the UK.
| Sector | Indicative UK cost per click | Typical cost per acquisition | Monthly spend for ~30 conversions |
|---|---|---|---|
| Local trades and home services | £1.50–6.00 | £20–70 | £600–2,100 |
| B2B professional services | £3.00–12.00 | £60–220 | £1,800–6,600 |
| IT and technology services | £4.00–14.00 | £80–300 | £2,400–9,000 |
| Legal and insurance | £8.00–28.00 | £150–600 | £4,500–18,000 |
| E-commerce, general retail | £0.40–2.50 | £8–45 | £240–1,350 |
The final column is the one that should inform strategy selection, and it is rarely calculated. An advertiser in legal services spending £2,000 a month is very unlikely to generate thirty conversions, which means automated targets will behave erratically no matter how well the account is built. That is a structural consequence of click costs rather than a failure of management, and the honest options are consolidating to fewer campaigns, pooling with a portfolio strategy, accepting volatility, or bidding manually.
Conversely an e-commerce advertiser can reach meaningful volume on a few hundred pounds a month, which is why automated and value-based strategies are near-universal in retail and comparatively rare in high-value lead generation. The same platform advice does not apply equally across those two situations, and generic guidance that recommends target ROAS to everybody is written for the retail case.
One consequence worth drawing out: for expensive-click sectors, improving conversion rate is often a better route to viable automation than any bidding change, because it raises conversion volume at constant spend. Landing page and enquiry-flow work is unglamorous next to bid strategy and frequently has a larger effect on what strategies become available to you.
The number that determines whether automation can work
If one figure decides whether an account will get value from automated bidding, it is not spend, volume or sector. It is whether the conversions being counted are things the business would actually pay for.
Forty-three per cent means that in well over half of UK accounts, Smart Bidding is being asked to optimise toward a target that includes something the business does not want more of. The system will do exactly that, efficiently and at scale, and the reporting will look reasonable throughout because the cost per conversion is calculated against the same inflated denominator.
This is why the figure matters more than volume. An account with thin conversion volume and clean data has a known limitation and a sensible fallback. An account with ample volume and contaminated data has a confident optimiser pointed slightly wrong, spending a full budget, and producing numbers that justify continuing. The second situation is more expensive and much harder to notice.
Checking it is genuinely quick. List every active conversion action, and for each one ask whether you would pay money for one more of those events. A submitted enquiry from a qualified prospect, yes. A completed purchase, obviously. A contact page view, no. A brochure download, possibly, but not at the same value as an enquiry — and if both sit in the bidding column at equal weight, you have told the system they are equivalent. Where the answer is no, remove it from the conversions column used for bidding while keeping it as a secondary action for reporting, which is a setting rather than a deletion.
Value-based bidding, and why lead generation needs it most
Once conversion data is clean and volume is adequate, the largest remaining gain in most UK accounts is telling the system that conversions are not all worth the same. Only about a quarter of accounts do this, and for lead generation businesses it is the difference between optimising for enquiries and optimising for revenue.
The problem with counting conversions equally is straightforward. If your enquiries range from a £400 one-off job to a £40,000 annual contract, and both register as one conversion, the system will optimise toward whichever is cheaper to generate — which is almost always the small one. It is doing precisely what you asked. Over months this drags the account toward high-volume, low-value traffic while every visible metric improves, because cost per conversion falls as the average conversion gets smaller.
Two practical routes
The first is assigning estimated values at the point of conversion, differentiated by what the enquiry appears to be. A quote request for a large installation carries a higher value than a general contact form; a demo booking carries more than a newsletter signup. These are estimates and they do not need to be precise — they need to be directionally right and consistently applied, because the system is learning relative worth rather than absolute figures.
The second, and considerably stronger, is feeding actual outcomes back. When a lead becomes an opportunity or closes, that result is imported against the original click, so the system learns which traffic produced revenue rather than which produced form fills. This requires the CRM and the ads account to be connected and somebody to maintain the flow, which is why fewer than one in five UK lead generation accounts do it. It is also why those that do tend to pull away from competitors who are still optimising for lead count.
What it requires of the business
Honesty about lead quality, and a sales process that records outcomes reliably enough to feed back. That second condition is the real blocker in most organisations: if leads are worked from an inbox and outcomes live in somebody’s memory, there is nothing to import. The prerequisite for advanced bidding is therefore a functioning CRM discipline, which is not an advertising problem and is frequently where the project actually stalls.
Where telephone enquiries matter — and for most UK trades, professional services and healthcare advertisers they matter a great deal — call tracking is part of this foundation rather than an optional refinement. An account where half the enquiries arrive by phone and only web forms are tracked is optimising on half its data, and the half it cannot see may be the more valuable one. We cover the setup in our guide to call tracking for PPC campaigns.
The 12-point bidding strategy checklist
Items one to four establish whether automation can work at all. Items five to eight select and launch the strategy. Items nine to twelve operate it.
- Audit every conversion action against the question “would we pay for one more of these?” Remove anything that fails from the conversions column used for bidding, keeping it as a secondary action for reporting.
- Resolve double-counting and check counting settings. One enquiry firing two actions halves your apparent cost per acquisition and doubles the aggressiveness of any target set against it.
- Add call tracking where telephone enquiries are a real route. Roughly half of accounts that need it do not have it, and those accounts are optimising on partial data.
- Measure conversion lag. How long conversions take to arrive after the click. This sets the minimum window before any performance judgement is valid.
- Count monthly conversions per campaign. Comfortably above thirty can carry its own target. Below that, consolidate, pool into a portfolio strategy, or stay manual.
- Choose the strategy from the goal, not the Recommendations tab. Cost efficiency points to target CPA; real transaction values point to target ROAS; brand terms point to manual or impression share; no history points to maximise clicks with a cap.
- Set the initial target at measured performance, not aspiration. The objective of the first phase is stability. A target below achievable performance narrows bidding until volume disappears.
- Change one thing at a time. Switch the bid strategy without simultaneously altering budgets, structure or keywords, or you will not know what caused the result.
- Leave it alone for the learning period. Seven to fourteen days. Negative keywords are acceptable; touching the target restarts calibration, which is what four accounts in five do.
- Tighten in ten to fifteen per cent steps, watching impression share. Steady impression share means genuine efficiency gain. Falling impression share means you are now trading volume for efficiency.
- Review auto-apply recommendations and switch off what you did not choose. Some settings will change bid strategies and targets automatically. Five minutes on any inherited account.
- Add conversion values, then offline outcomes. Differentiated values first, then actual closed business fed back, so the system optimises for revenue rather than enquiry count.
If only three items are completed, make them one, seven and nine. Auditing what counts as a conversion determines whether automation is pointed at anything real — and it fails in over half of accounts. Setting the target from measurement rather than ambition prevents the most common failure state, where a campaign hits its target on almost no traffic. And leaving the learning period alone is free, requires only patience, and is the difference between a strategy that stabilises and one stuck in permanent recalibration.
What this looks like in practice
A UK commercial cleaning contractor with 60 staff had been running Google Ads for three years at about £4,200 a month, managed in-house by a marketing executive. Performance had been acceptable and then deteriorated over about five months: reported cost per acquisition rose from roughly £74 to £118 while enquiry volume was broadly flat, and the sales team was independently complaining that the enquiries had become less serious.
The account had been switched from manual CPC to target CPA nine months earlier on the recommendation of a Google representative, with a target of £65 — below the £74 the account was achieving at the time. The switch was made in the same fortnight as a landing page redesign and a budget increase.
The audit found three compounding problems. Four conversion actions were active and all four sat in the bidding column: enquiry form submissions, a quote calculator completion, contact page views, and a PDF brochure download. Contact page views alone accounted for a little over a third of counted conversions, which meant the reported cost per acquisition had never been the real one and the £65 target had in reality been far more aggressive than it appeared. Second, telephone calls — which the sales team said were the better enquiries — were not tracked at all. Third, the target had been edited five times in nine months, twice within a fortnight of each other, so the strategy had spent much of the period recalibrating.
Remediation followed the sequence above. Page views and brochure downloads were moved to secondary actions, leaving enquiries and quote calculator completions in the bidding column. Call tracking was implemented, which immediately revealed that calls represented about 40 per cent of genuine enquiries. With the conversion set corrected, the true historic cost per acquisition turned out to be around £145 rather than £118 — a worse number, and the first honest one the account had produced.
The target was reset to £145 and left alone for a fortnight. Volume recovered as the campaign stopped being artificially constrained. Over the following ten weeks the target was tightened in four steps to £118, at which point impression share began falling and the tightening stopped. Six months later, estimated values were added to differentiate contract enquiries from one-off jobs.
The number went up before it went down, and that was the hardest part to explain internally. We had been reporting a hundred and eighteen pounds a lead to the board for months and had to go back and say the real figure was a hundred and forty-five, and that the reason we now knew was that we had stopped counting people looking at our contact page as leads. Nobody enjoyed that meeting. Everything after it was easier.
Two points generalise. The first is that the original switch was made with three changes at once, which made the subsequent decline impossible to attribute — and the decline was not caused by automation but by automation faithfully pursuing a contaminated target. The second is that correcting conversion data usually makes reported performance look worse before it gets better, and an organisation that has not been prepared for that will interpret the correction as the problem.
Common bidding strategy mistakes
The errors below recur across UK accounts. Nearly all are decisions about data or process rather than about which strategy is theoretically best.
- Automating before fixing conversion tracking. Manual bidding on bad data is merely unmeasured; automated bidding on bad data spends the full budget pursuing the wrong outcomes, confidently. Fix the data first, in that order, never the reverse.
- Counting things the business would not pay for. Contact page views, brochure downloads and newsletter signups sitting in the bidding column at the same weight as enquiries. Present in over half of audited accounts.
- Setting the target at the desired cost per acquisition. A target below achievable performance does not make the system try harder; it narrows bidding until impression share and volume collapse while the target is nominally met.
- Intervening during the learning period. Roughly four accounts in five adjust the target while the strategy is calibrating, restarting it each time and producing permanent volatility they then attribute to automation.
- Changing several things at once. Switching bid strategy alongside a budget change and a landing page redesign makes the outcome unattributable, which is how automation gets blamed for unrelated declines.
- Running automated targets on thin volume. Below about thirty monthly conversions the prediction has too little to work from. Consolidate, pool into a portfolio strategy, or stay manual — do not simply accept the volatility.
- Using Smart Bidding on brand terms. High predicted conversion probability means high bids for traffic that was converting anyway. Manual CPC or target impression share usually holds the position for less.
- Valuing every conversion equally. Where enquiries range from small jobs to major contracts, equal weighting drags the account toward whatever is cheapest to generate, which is the small end.
- Leaving auto-apply recommendations enabled. Some settings change bid strategies and targets without your involvement, undoing decisions made deliberately.
- Judging performance inside a week, or without allowing for conversion lag. If conversions take ten days to arrive, a seven-day assessment is measuring an incomplete picture and will systematically understate results.
Treat platform recommendations as informed suggestions from a party whose commercial interest is correlated with your spend rather than with your profit. That is not an accusation of bad faith — many recommendations are sensible, account representatives are frequently knowledgeable, and the platform genuinely benefits from advertisers succeeding. But a proposal to raise budgets, broaden match types or adopt a strategy that spends more is not neutral advice, and the person making it cannot see your margins, your capacity or your lead quality. Ask what the recommendation assumes about your cost per acquisition and where that figure came from. If it came from your own contaminated conversion data, the recommendation inherits the contamination.
At a glance — Google Ads bidding summary
| Question | Short answer |
|---|---|
| How does Smart Bidding decide a bid? | At auction time, predicting conversion probability for that specific query from signals like device, location, time, audience and search term |
| What is its genuine advantage? | Per-auction bid differentiation, which no manual structure can express |
| What can it not see? | Your margins, lead quality, capacity and anything outside the conversion data you feed it |
| The two prerequisites | Adequate conversion volume, and conversion data measuring something the business values |
| Practical volume threshold | Around 30 conversions monthly per strategy. Google removed the stated minimum, not the statistics. |
| Accounts with clean conversion data | About 43 per cent — which caps how well automation can perform across the rest |
| When manual still wins | Very low volume, brand terms, unfixed tracking, and where you hold knowledge that cannot be encoded |
| Learning period | 7–14 days. Negative keywords are fine; touching the target restarts it. |
| How to set the first target | At measured current performance, not the figure you would like |
| What an aggressive target does | Narrows bidding until impression share and volume collapse, while nominally hitting the target |
| How to tighten safely | 10–15 per cent steps, one to two weeks apart, watching impression share as the constraint signal |
| Remedy for several small campaigns | A portfolio strategy pooling their conversion data — used by only about 6 per cent of accounts |
| Biggest untapped gain in lead generation | Feeding offline outcomes back so the system optimises for revenue rather than enquiry count |
| Underused free tools | Seasonality adjustments and data exclusions — used by around 12 per cent |
| How to read platform recommendations | As informed suggestions from a party whose interest correlates with spend rather than profit |
How Cloudswitched approaches bid management
Cloudswitched manages Google Ads for UK organisations, and on any inherited account the first work is the conversion audit rather than the bid strategy — because in more than half of accounts the numbers every subsequent decision would rest on are not measuring what the business values. In practice that means establishing what should count as a conversion, adding call tracking where telephone enquiries matter, measuring conversion lag and real cost per acquisition, then matching strategy to the volume each campaign actually generates. Targets are set from measurement and tightened in steps, the learning period is left alone, and for lead generation clients we work toward feeding closed business back so bidding optimises for revenue rather than enquiry count. Where an account genuinely lacks the volume for automation, we will say so and bid manually.
Find out what your real cost per acquisition is
We audit what your account is counting before touching a bid, then match strategy to the conversion volume you actually have — including saying when manual is the better answer.
Talk to a Google Ads SpecialistFrequently Asked Questions
Is automated bidding better than manual bidding?
For accounts that meet two prerequisites, yes, and the gap is not close. Smart Bidding sets a bid for every individual auction based on a prediction of conversion probability from signals including device, location, time, audience membership and the specific search term — granularity no manual structure can express. The prerequisites are adequate conversion volume for the prediction to be statistically meaningful, and conversion data that measures something the business actually values. Where either is missing, automation does not degrade gracefully; it optimises confidently toward the wrong thing, which is worse than a mediocre manual bid because it is harder to notice.
How many conversions do I need for Smart Bidding?
Treat around 30 a month per strategy as the point below which you should expect volatility. Google formerly published minimum thresholds and has removed them from the interface, and the strategies will run at lower volumes — but what was removed is the requirement, not the statistics. A model predicting conversion probability from eight data points a month is working with very little, and the observable result is erratic bidding with targets hit in one fortnight and badly missed the next. Below the threshold, consolidate campaigns, pool them into a portfolio strategy, or stay manual.
When should I still use manual bidding?
Four situations. Genuinely low conversion volume, where a human with category knowledge beats confident predictions from very little data. Brand campaigns, where intent is already high and Smart Bidding may pay well above what is needed for traffic that was converting anyway — manual CPC or target impression share usually holds the position for less. Where conversion tracking is broken or being rebuilt, because manual is merely unmeasured while automation actively pursues faulty signals. And where you hold material knowledge that cannot be encoded as a schedule, exclusion or audience.
What is the learning period and what should I do during it?
Roughly seven to fourteen days after a strategy change, during which the system calibrates and performance is volatile — cost per acquisition often overshoots and click costs move unpredictably. What to do is very close to nothing. Adding negative keywords and pausing obviously irrelevant search terms is acceptable. Adjusting the target, switching strategy or restructuring restarts calibration. About four accounts in five intervene during this window, usually because early performance looks alarming, which produces permanent learning-period volatility that then gets attributed to automation being unsuitable.
What resets the learning period?
Changing the bid strategy itself, changing the target substantially, altering which conversion actions are used for bidding, and major structural changes to the campaign. Small budget adjustments, adding negative keywords and ad copy changes generally do not, or have limited effect. The practical rule when tightening a target is to move in ten to fifteen per cent steps with one to two weeks between changes — large jumps can be treated as significant changes, which both destabilises the campaign and obscures whether the new target was achievable.
How should I set my target CPA?
At the cost per acquisition you are actually achieving, measured over a meaningful period with a correct conversion set. This feels pointless because the reason for automating is usually to improve, and it is what makes improvement possible: the system needs stable operation before it can be pushed. Then tighten in ten to fifteen per cent steps, watching impression share. While impression share holds steady the tightening is genuine efficiency gain; when it starts falling materially you have reached the point where further tightening trades volume for efficiency, which should be a commercial decision rather than a surprise.
Why did my campaign stop spending after I set a target?
Almost certainly because the target is below what the campaign can achieve. An aggressive target does not make the system work harder — it narrows the set of auctions where predicted cost per conversion falls within the target, and the system bids only there. Impression share collapses, volume disappears, and the reported cost per acquisition may well hit the target on a fraction of the traffic. This is the most misleading failure state in the platform, because every visible metric except volume improves. Raise the target to measured performance and tighten gradually from there.
Should I use target CPA or target ROAS?
Target CPA where the goal is cost efficiency per conversion and conversions are broadly comparable in worth. Target ROAS where you have conversion values that mean something real — straightforward for e-commerce with transaction values, and for lead generation requiring either estimated values by enquiry type or, better, actual closed business fed back. Adopting target ROAS without real values produces confident optimisation toward a number you invented, and the resulting return on ad spend figure is a measurement of your own assumptions. Target ROAS also typically wants more conversion volume than target CPA before it stabilises.
What is a portfolio bid strategy and should I use one?
A portfolio strategy applies one bid strategy across several campaigns, pooling their conversion data. It is the right answer when individual campaigns each sit below meaningful volume: three campaigns with twelve conversions a month each will behave erratically on separate targets and considerably better on one shared strategy. Only about 6 per cent of audited accounts use them, which makes this one of the more common missed opportunities in smaller UK accounts. It is a setting rather than a project, though it does mean the campaigns share one efficiency target rather than having individual ones.
Why did performance get worse after switching to Smart Bidding?
Most commonly one of three causes. The conversion data was contaminated, so the target was derived from a cost per acquisition that was never real — if page views were being counted, the apparent figure was flattering and any target set against it was far more aggressive than intended. Or the learning period was interrupted, so the strategy never stabilised. Or several things changed at once, making the decline unattributable and automation the convenient explanation. Audit the conversion set first; in over half of accounts that is where the answer is.
How do I make bidding optimise for revenue rather than lead count?
Feed value back. The weaker version assigns estimated values at conversion, differentiated by what the enquiry appears to be worth — these need to be directionally right and consistently applied rather than precise, because the system learns relative worth. The stronger version imports actual outcomes: when a lead becomes an opportunity or closes, that result is attributed to the original click so the system learns which traffic produced revenue. Fewer than one in five UK lead generation accounts do this, and it requires a CRM discipline that records outcomes reliably — which is usually where the project actually stalls.
Should I follow Google’s recommendations?
Read them, do not auto-apply them. Many are sensible and account representatives are frequently knowledgeable, but a recommendation to raise budgets, broaden match types or adopt a strategy that spends more comes from a party whose commercial interest correlates with your spend rather than your profit, and who cannot see your margins, capacity or lead quality. The useful question is what the recommendation assumes about your cost per acquisition and where that figure came from — because if it came from your own contaminated conversion data, the recommendation inherits the contamination. Also review the auto-apply settings on any inherited account; some will change bid strategies and targets without your involvement.
Related reading
More guidance on measuring and managing UK digital marketing and technology:
Bid strategy is a data decision before it is a settings decision
Cloudswitched audits what your account counts, establishes your real cost per acquisition, and matches strategy to the conversion volume you actually generate — then leaves the learning period alone.
Talk to a Google Ads Specialist
CloudSwitched
London-based managed IT services provider offering support, cloud solutions and cybersecurity for SMEs.
