How much of your ad budget is wasted (and how to find it)

Ask a room of media buyers how much of the average account is wasted and you get a shrug and a range. The best-sourced figure we know of is WordStream's: about $1,127 a month spent on clicks that did not convert, against a median monthly spend of about $3,127. Roughly a third of the budget, going to queries nobody meant to buy, placements nobody chose and audiences you are paying twice to reach.

Read that number as context, not as a diagnosis. The accounts in the study came through a free grading tool, so they likely lean toward the unoptimised end, and “wasted” there means clicks that did not convert, which is a loose definition. Your account is not the median. The useful part is the shape underneath: the waste is not hiding in some sophisticated attribution model. It is sitting in reports you already have access to. Here is the playbook, roughly in the order the leaks tend to give money back.

1. Wasted search terms: the largest pool you can stop today

Start here, because it is usually the biggest single pool and the fastest to fix. Broad and phrase match keywords pull in queries you never intended to bid on. The search terms report is where you see what people actually typed to trigger your ad, and it is routinely full of junk: wrong intent, “free”, “jobs” and “DIY” modifiers, and searches for products you don't sell.

Every click on one of those is spend you can stop today with a negative keyword. Pull the search terms report over the last 30 to 60 days, sort by cost, and read down until the queries stop looking like buyers. On a neglected account this is where the afternoon pays for itself.

How to price it: total the cost of the queries you are about to add as negatives over the window, and scale it to a month. It runs larger on a broad-match-heavy account that has never been groomed and smaller on one that is already tightly matched.

2. The location default: “interest” vs “in”

This one is a single setting, and it quietly bleeds local and regional accounts. The location option Google recommends for Search, Presence or interest, shows your ads to people who are physically in your target area or who have merely shown interest in it. For a plumber in Manchester, that can mean paying for clicks from someone in another country googling “Manchester” for a trip.

Switch it to Presence: people in or regularly in your targeted locations and you cut the tourists and the researchers. Google's own location targeting help says to consider Presence when you only want to reach people in specific locations. The broad option widens reach, and for a business that serves only one area, that extra reach is mostly people who cannot buy.

How to price it: segment by user location and total the spend that came from outside the area you serve. It is negligible for genuinely national or global advertisers, which is exactly why it is easy to miss on the accounts it hurts most.

3. Placement leakage: the leak in the average

Display, video and Performance Max spread your budget across placements you never hand-picked: apps, made-for-ads sites and low-quality partner inventory that eats impressions and returns nothing. A healthy blended number can hide a placement quietly burning cash underneath it, which is the exact dynamic in placement drift: the leak hiding in your average.

Pull the placements report (or the Performance Max placement and channel reporting), sort by cost, and look for spend with clicks but no conversions. Exclude the worst offenders, and exclude app placements if you are not deliberately advertising in apps. On accounts running Display or Performance Max at scale, this is often the second-biggest pool after search terms. How to audit a Performance Max campaign walks through where that spend hides.

How to price it: the spend on placements with clicks and no conversions over the window, scaled to a month.

4. Device bids: paying a premium for the worse device

Segment any mature campaign by device and the conversion rates rarely match. Mobile often converts worse than desktop for considered purchases; the reverse is often true for impulse and local. If your bids treat every device the same, you are overpaying on the one that converts worse and underbidding the one that converts better.

On manual bidding, look at cost per conversion by device over a meaningful window and adjust to match reality: down on the laggard, up on the leader. It is a smaller pool than the first three, but it is pure margin, because you are not cutting volume, just repricing it. On Smart Bidding (Target CPA, Target ROAS, Maximize conversions or conversion value), Google's bid adjustment rules ignore manual device adjustments except −100%, which switches a device off, and Target CPA treats a device adjustment as a change to the target. There the lever is the target and the conversion data you feed it, not the bid.

How to price it: the gap between each device's cost per conversion and the account's, times that device's conversions, over a month.

5. Audience saturation: paying full CPA to reach converters

If you have no audience exclusions, you are paying full price to show ads to people who already bought. Retargeting that keeps chasing existing customers, prospecting that re-serves recent converters, and broad campaigns with no exclusion list all spend real money reaching people who are done. You pay the full CPA to acquire someone you already acquired.

Add exclusions: existing-customer lists, recent purchasers out of prospecting, and converters out of top-of-funnel campaigns. Retargeting is not wrong. Unbounded retargeting saturates, and the incremental return collapses while the spend does not.

How to price it: the spend that reached people who had already converted, over the window. It concentrates in accounts leaning hard on retargeting.

6. Branded-search double-pay

The most argued-about leak on the list. When someone searches your brand name, they were likely going to find you organically, so a branded click can be spend for a visit you would have had for free. Not always: if others bid on your name, or your organic listing is buried, branded search earns its keep. But a lot of accounts run it on autopilot and never test whether the paid clicks are incremental. Performance Max can blur this further by taking credit for branded demand.

The test is simple: pause branded search for a controlled window and watch whether total branded traffic and conversions actually drop, or just shift to organic. If they shift, you were double-paying.

How to price it: only the test prices it. The leak is the branded spend whose conversions moved to organic when you paused, which is why it is a test, not a rule.

Attach a number to every leak

Here is the framing that turns a checklist into a decision: every leak above can carry a monthly price, and the only question worth asking is which one gives back the most, first. A generic “you have wasted spend” is noise. “Your search terms are leaking this much a month, and your location setting is on the wrong default” is something you can act on before lunch.

That is the whole idea behind recoverable spend: not a vague health score, but a ranked list of leaks with a dollar figure and a fix attached to each. It is how you decide what to touch, and it is the framing our product is built around. Don't tell someone their account is unhealthy; tell them where the money is and what closing the gap is worth.

Pricing also disciplines whoever produced the list. In rough order of money recovered per hour spent:

1. Check the signal before anything else. If the conversion the account optimises toward is not the conversion the business cares about, every efficiency number downstream is a proxy, and the ranked list you build on it is ranked wrong.
2. Price each candidate leak as a monthly range. Spend in the window, times the share you would actually reclaim, scaled to a month. Write both ends of the range down; if you cannot, you do not understand the leak yet.
3. Put what you can't price on a watch list. Write down the condition that would make each one worth acting on, so it comes back when the condition is met instead of nagging you every week.
4. Re-run the same list a month later. A leak that reappears unchanged is not a finding any more. It is money that has already gone.

A 60% swing on a campaign spending $50 a day is noise wearing a large percentage; a 4% drift on the campaign carrying half the budget is the thing that matters. That is why the price comes before the percentage.

What one run surfaced on a live account

Here is what that method looks like when software runs it. A single run on one live account produced 60 decisions, each with evidence attached and each priced. Not “issues found”: decisions, sorted by what they are worth per month, so the list ranks itself and you work top down instead of reading all of it.

It is a range. $2,892 to $5,867, not a single figure. The spread is the honest expression of what the account's own history can support: pricing a leak requires assuming what happens to the money afterwards, and that assumption has a floor and a ceiling.

Only 34 of 60 are “fix now”. Thirteen are opportunities, seven are experiments, and six are being watched rather than acted on. A tool that returns 60 urgent problems has not read an account; it has run a checklist.

192 pieces of evidence for 60 decisions. Roughly three per decision. Every verdict names the rows it came from, which is the only reason you can argue with it in your own Ads Manager instead of taking it on faith. It is one account's output, not a benchmark for yours.

Sometimes the leak is somewhere else

The most useful output we have seen on a second live account was a diagnosis that moved the problem. On a Google read of 61 campaigns and $7,825 (₺378,800) over 28 days, in August 2026:

Both of the obvious “recover wasted spend” moves were available, and neither fit: nothing was over-pacing, so there was no case to cut, and adding budget to campaigns held at 10% impression share buys a larger share of the same loss. The money was not being wasted at the budget layer at all. A waste percentage that cannot reach that conclusion is measuring the wrong thing confidently.

The catch: doing this once isn't enough

You can work this list in an afternoon and claw back a chunk of budget. The problem is that waste regenerates. New junk search terms accumulate every week. Performance Max drifts toward cheap inventory the moment you stop feeding it exclusions. A new campaign launches on the location default because that is what defaults are for. The afternoon audit is a snapshot; the leak is a stream.

This is where continuous beats periodic. An AI media buyer reads these reports the way you would (search terms, placements, location, device, audiences), but on a schedule rather than once a quarter. It ranks each leak by what it is costing now and keeps the evidence, next move and later result together. It removes the hours of digging that stop you doing this as often as you should.

If you would rather run the manual version first, the step-by-step is in how to audit a Google Ads account, and the Meta equivalent is the Meta Ads account audit.

Frequently asked