Menu
Your client’s monthly review is coming up soon. You’ve just seen that your Google Ads CPA increased by 38% since last month, spend hasn’t moved, and the account looks, at a glance, exactly the way it did when CPA was healthy. Same campaigns, same budgets, same bid strategies.
The natural instinct is to start pulling levers. Drop the target, pause the expensive keywords, tighten the bids, and hope something lands before the call. It lets you tell your client you’ve made some updates, but you haven’t actually found the source or put a guardrail in place to stop it happening again next month.
If you want a repeatable, step by step process you can actually follow, this article is for you.
Your Google Ads CPA increased for one of two mechanical reasons: you’re paying more per click, or fewer of those clicks are converting. CPA is just your cost per click divided by your conversion rate.
Seven causes sit behind changes in those two numbers: auction competition, a Quality Score slip, Smart Bidding misfires, match-type drift, a broken landing page, broken conversion tracking, and simple scaling. They group into three diagnostic paths. Before touching anything, work in this order.
Because CPC and conversion rate are averages rather than totals, the two time windows don’t need to be the same length, as long as each has enough clicks to be reliable and each covers comparable conditions, so you’re not holding a peak week against an off-peak month.
Within each path, the checks are ordered by how fast you can run them. Work through them in order and stop when you find the cause.
👉 One shortcut worth knowing before you start: if CPA leapt to an absurd number overnight rather than drifting up over weeks, suspect tracking before anything else. That’s Step 1.
CPA is spend divided by conversions. If conversions are being undercounted, the number inflates on its own, with nothing wrong in the market, your bids, or your pages. This is the cause that wastes the most agency time, because people optimise furiously against a problem that only exists in the reporting.
Tracking can break quietly. A tag dropped during a site migration. A consent-mode change that stopped conversions firing for a slice of users. A GTM edit a developer shipped without telling anyone. Then there’s attribution lag, which fools people constantly: conversions can register days after the click, so a recent window regularly looks worse than it will once those conversions land. Judging this week’s CPA before the conversions catch up is like calling a match at half time.
What makes tracking fiddly to diagnose is that it isn’t one system. A conversion travels a chain: the click might fire a tag in Google Tag Manager, that tag reports to GA4 or straight into Google Ads, phone leads route through a call tracking platform, and maybe offline sales come back from your CRM. A break in any single link reads the same from inside Google Ads, and each link could be owned by a different person who only checks their own. The reliable way through is to take one conversion you can actually identify, a test lead or a known sale, and follow it along the whole chain until you reach the point where it disappears.
That trace used to mean logging into five tools one after another. AI has made tracking diagnosis much faster by connecting the dots and adding a layer of expertise across all the linkage points (GA4, Google Tag Manager, Google Ads, your call tracking, and your CRM). It can very quickly follow a conversion path through every link and tell you where the signal drops, and can even produce a to-do report that can be passed onto web devs, site owners etc.
Either way the principle holds: check the path end to end, not one system in isolation. If you’d rather hand that off wholesale, there’s an example copy-paste AI prompt for the full trace at the end of this article.
Once you’ve found the break, fix it before you touch a bid, and if Smart Bidding has been feeding on broken data, let it re-learn on clean signals before you judge the real CPA underneath.
Did your CPA climb at the same time as your conversion volume, off the back of a budget or bid increase? That’s the price of scale, and it may actually be at an acceptable price. If your spend held steady and impression share didn’t step up, you didn’t scale into this, so skip through to Step 3.
When you raise a budget or increase bids, you’re asking Google to buy more of the market. The catch is that your cheapest, highest-intent conversions were already being captured at the old spend. Everything above that line is less efficient by design.
Moving from, say, 20% to 40% search impression share means winning auctions you used to skip. You’d already captured your best, highest-intent searches at the old spend, so the ones you add at the margin convert less well, and you pay more to win them. Marginal CPC rises and marginal conversion rate falls at the same time. Both halves of the CPA equation move against you at once. That’s not a broken account. That’s just the shape of the demand curve.
The marginal maths can be fiddly to do by hand, so there’s an AI prompt at the end that runs it from your before-and-after numbers.
Judge the expansion on incremental economics, not on the blended average that scaling always lifts. If the marginal conversions clear your profit threshold, hold your nerve and keep the volume. If they don’t, pull back to the impression share band where efficiency held and grow in smaller steps, roughly 20% at a time, so the algorithm and your CPA both stay stable. One caveat before you trust the new number: the same budget increase may have also knocked Smart Bidding into learning (Step 3), so the CPA is still settling. Don’t lock in a verdict until it has.
Smart Bidding only does what you set it up to do, so a lot of CPA inflation comes from a bad setup rather than a bad algorithm. No amount of machine learning fixes instructions that were off target to begin with.
It shows up three ways.
There’s a direction to this that catches people out. Target-based bidding optimises toward the number you set, so your target isn’t a ceiling you hope to beat, it’s the level the system aims for. If a campaign was limited by budget and quietly running below its target, Google will work it up toward that target as it gets the room, and a target sitting above the CPA you actually want will drag your realised cost up to meet it. A target set once and never revisited is one of the quieter reasons a CPA drifts higher. Set it to the CPA you want, not a placeholder you typed in during setup.
If multiple people work on the same account, compounding is a real danger. Three people making changes to one campaign across a fortnight can hold it in permanent learning, where it never settles long enough to bid efficiently. Nobody did anything wrong on their own. The account just never got to sit still. Adpulse stores and shows the previous bid strategy and the date it was changed, so a manager can see a campaign has already been touched this week before adding one more change on top, which is how you avoid the over-optimisation that keeps it stuck.
What to do
If it’s a genuine learning period, the hardest and most correct move is to stop touching it. Every change resets the clock. If it’s starvation, loosen the target to widen the conversion pool, broaden your conversion actions to feed more signal, or move to Maximise Conversions without a CPA constraint until you’ve built enough data to reintroduce a realistic target. If the search terms report turned up a handful of very expensive clicks, negatives on those specific queries could be the first move if they are poor intent (that’s the Step 6 workflow), not a reactive bid cap. Reach for an upper CPC limit only when the overpayment is spread across many terms you don’t want to negate: move the campaign to its own Target CPA portfolio strategy so the cap touches nothing else, and keep it sensibly loose, because a tight limit costs you volume everywhere to fix a problem in a few places.
The most deflating cause of a CPA increase is the one you didn’t create. The auctions you were winning cheaply last quarter have filled up. A new competitor with a fresh budget entered your client’s category, or an existing rival raised bids. The price of a click rises for everyone, and because CPA is built on CPC, yours rises with it while your conversion rate hasn’t twitched.
Nothing in your account changed. The market repriced around you, and the auction passed that straight through to your CPCs. At scale it’s easy to miss, because you notice it only when three accounts in the same vertical drift the same way in the same fortnight.
You have two simple options:
Chasing every impression at the new market rate is how a CPC problem becomes a CPA disaster.
Here’s the mechanic most people don’t fully understand. Quality Score doesn’t just decide whether you show. It decides what you pay to show. Your actual cost per click is, roughly, the Ad Rank of the advertiser beneath you divided by your own Quality Score, plus a cent. Higher Quality Score, lower price for the same position. So when Quality Score falls, your effective CPC rises even if no competitor changed a thing.
A drop from 7 to 5 on a high-volume keyword is enough to push CPCs up and cost you position as well. What makes it slippery is that Quality Score can decline while you sit still. Landing page experience is re-evaluated continuously, so a speed regression after a website deploy can drag the score down with nothing showing in the Ads interface. Google will not send you a memo when this happens.
Current Quality Score is only a snapshot, so the standard column can’t tell you whether a slip caused the rise. Use the historical columns instead.
Expect a lot of blanks, and don’t misread them. A keyword showing no Quality Score isn’t scoring zero and isn’t something to fix. Google only rates a keyword once it has gathered enough recent exact-match search activity, so new and genuine low-volume terms are reported as null. Two other things read as blank for reasons unrelated to quality: Quality Score is a Search-keyword metric only, so Shopping, Dynamic Search Ads, Performance Max, and Display targets never carry one, and the historical columns log a value only for days a keyword actually served, so anything paused or dormant in your range shows gaps. The blanks are rarely your CPA problem. The keywords worth your attention are the high-spend ones that carry a score and a trend, which is exactly why you sort by cost first.
Fix the specific component that dropped, not all three on principle.
One quick win is keyword insertion, the {KeyWord:…} tag in a headline. When the searched term matches, it tightens ad relevance, which often helps both Ad Strength and Quality Score.
Go from this:
To this:
Every point of Quality Score you win back lowers your effective CPC, which lowers CPA on the same spend.
Google has spent years widening what counts as a close variant. Exact match isn’t exact the way it once was, phrase match reaches further, and broad match on a Smart Bidding strategy can wander a long way from the query you had in mind. Your ads get served against searches adjacent to your keywords but lower in intent. Those clicks cost the same but convert at a lower rate.
Spread across a few hundred keywords, a quiet drift in query quality moves a whole account’s CPA without a single metric raising its hand.
Adpulse studied the CPA impact of close variants and found they typically carry an 80% higher CPA than search terms that match the keywords you’re actually bidding on.
Add the consistent offenders as negatives to steer spend back toward converting intent, and tighten match types on your highest-value keywords where the conversion data supports it. The reactive version of this, an audit every few weeks, always runs behind the drift. The reliable version is a search term review built into the standard daily or weekly workflow, so a bad query gets caught in days rather than discovered in a monthly report. For a large export, the AI prompt at the end will triage the terms and draft the negatives in one pass.
Not to be confused with the Step 1 tracking problem, where the conversions actually happened but weren’t recorded, making the CPA only look worse. This step is about conversions that genuinely didn’t happen because the visitor hit a roadblock. The fastest way to tell them apart is to look outside Google Ads: if your CRM or back-end shows the sales still coming through, it’s tracking, but if it confirms they’ve actually dropped, it’s the page.
Here, your targeting is sound, your search terms are on-intent, and the right people are clicking. Then they hit the page and don’t convert, so conversion rate falls and CPA climbs while every in-platform metric insists nothing is wrong.
The triggers usually live outside the Ads account. A checkout step that broke after a plugin update. A form that stopped submitting on mobile. A price rise, a changed offer, an out-of-stock hero product, a page-speed regression from a new tag. A campaign that ran out of budget is visible in Google Ads. A landing page that quietly stopped converting on iOS Safari is not, and it will burn a fortnight of spend at a wrecked CPA before anyone thinks to load the page on their phone.
Route the page and funnel fixes to whoever owns the site, and be specific about what broke and where, because “conversions are down” gets deprioritised while “the mobile checkout has failed since the 3rd” gets fixed. Where the page is technically fine but simply underperforming, that’s a conversion rate optimisation project, and it’s the highest-return one you have.
Remember the equation: lifting conversion rate lowers CPA just as directly as cutting CPC does, and it’s often the cheaper win. For a site with several conversion routes, the prompt at the end maps every path your ads feed and ranks them by spend. It catches structural and analytics-visible breaks, but a device-specific bug like an iOS Safari checkout failure still needs a human on that device. The map just tells you which route to load.
The hard part of a CPA investigation isn’t any single check. It’s that the seven live in seven different places, conversion status in one report, CPC trends in another, Quality Score history somewhere else, search terms somewhere else again, and you have to run all of them across every account before the culprit shows itself. Done by hand across eighty clients, the slow, silent problems are the ones that slip through: the tag that broke on Tuesday, the disapproved ad nobody was told about, the campaign that has been learning for a month. Adpulse watches CPA, conversion tracking, budget pacing, and performance anomalies across every connected account in one view, so the diagnostic surfaces before the client’s monthly review does. If you manage paid media at scale and want that visibility without logging into each account one at a time, Adpulse runs a free 14-day trial with no account or feature limits from the start.
A few of these checks are slow by hand and quick to hand off. The prompts below are written to be pasted into whatever AI assistant your agency is comfortable giving account data to, and they’re deliberately tool-agnostic. The more your assistant can reach, the more of each one it can run. They don’t replace your judgement, and a prompt is only ever as good as the data you feed it, so treat the output as a first pass you sanity-check rather than a verdict.
You have access to my GA4, Google Tag Manager, Google Ads, call tracking platform, and CRM. Here is one conversion I can identify: [describe it, e.g. a test lead submitted on DATE, or order #1234]. Trace it through every system in order: did the tag fire in Tag Manager, did it reach GA4, did it register as a conversion in Google Ads, did the call or lead route through the call tracking tool, and did it land in the CRM? Tell me the exact link in the chain where it stops appearing, and what that points to. Then list any Google Ads conversion actions showing "no recent conversions" or an inactive status, and tell me whether the recent drop lines up with any tag or site change.
You are a paid search analyst. I'll give you Google Ads data for one campaign across two periods: the weeks before a budget increase and the weeks after. For each period I'll provide total cost, total conversions, and total conversion value.
1. Work out CPA, conversion rate, and ROAS for each period.
2. Work out the marginal cost per conversion of the added spend: (cost_after - cost_before) / (conversions_after - conversions_before), and do the same for value to get marginal ROAS.
3. Tell me whether those marginal conversions clear my break-even value of $X (or my target CPA of $Y).
4. Flag anything that makes this unreliable: a seasonality shift, other changes made in the same window, or too few added conversions to trust the number. Do not treat the before-and-after difference as proven incrementality; a causal read needs a controlled budget experiment.
You are a paid search analyst. I'm pasting a Google Ads search terms report with these columns: search term, matched keyword, match type, cost, clicks, conversions, conversion value. My keywords are meant to reach [describe who your customer is and what they're searching for].
1. Flag the search terms that are off-intent or unlikely to convert for this business, with a one-line reason each.
2. Rank them by wasted spend (cost with no or few conversions).
3. Draft a negative keyword list from the clearest offenders, and say whether each should be a phrase or exact match negative.
4. Point out any high-converting terms worth adding as their own keywords.
Don't negate terms that are converting acceptably, even if they look tangential.
You are a conversion rate analyst. My Google Ads conversion rate has dropped while the traffic still looks on-intent. Help me find where visitors are failing to convert. Work in three stages.
1. Map the paths. For the affected campaigns, take the landing page URLs the ads point to and list every route a visitor could take to convert on each: the conversion action (lead form, phone call, live chat, checkout), the steps in that route, and the device and browser combinations that matter, at least mobile and desktop. Rank the routes by the ad spend flowing to them so the costliest are tested first.
2. Test each path. If you can open the pages, walk each route from landing to the final step on both mobile and desktop and tell me the exact point where a visitor would hit a roadblock: a form that won't submit, a broken or slow checkout step, a dead CTA, a consent wall, an out-of-stock or changed-price surprise, a validation error. Walk to the point of submission only, don't complete real purchases or send real leads. Where you can't open a page, inspect the structure and the analytics instead: conversion rate by page, device, and browser, plus page speed or Core Web Vitals for any regression that lines up with the date the CPA rose. Be explicit about limits: you can flag structural and analytics-visible failures, but you can't reliably reproduce a device-specific bug like an iOS Safari checkout failure, so for those, name the route and device and I'll test it myself.
3. Report. For each broken route, give me the path, the device or browser it fails on, the exact failure point, and the share of spend it affects, most costly first.
Don't assume the tag is the problem, that's a separate tracking check. Here I want to know whether a real visitor can actually complete the action.
We love talking PPC and spend most of our time chatting with agencies, so if you have a question please do drop us a line!
Also, Adpulse comes with a 14-day free trial, and right now the first month is only $19.99/mth, so it’s a low-risk way to try it for yourself. Sign up now!
