What broken attribution is really costing you (in dollars)

Category: Marketing & analytics | Reading time: 7 min

It's easy to talk about data loss in percentages: 25–40% of conversions missing; recovery rates of 50%. These numbers are accurate, but they're abstract – and we know abstract numbers don't change behaviour the way dollar figures do.

So let's make it concrete with real numbers that will make everyone on your team understand what’s happening. Below is a worked example following one  business – a mid-sized eCommerce brand – through exactly how broken client-side tracking compounds into lost revenue, wasted spend and bad marketing decisions.

Meet the business

Aurora Coffee Co.* sells specialty coffee subscriptions online (i.e. they have no store).

• Monthly ad spend: $15,000 (split across Google Ads and Meta)
• Average reported monthly conversions (from GA4/ad platforms): 300
• Average reported cost per acquisition (CPA): $50
• Average order value: $80
• Audience profile: 55% mobile, with roughly a third of total traffic on Safari (this is consistent with a consumer brand in Australia, the UK or the US).

These are the numbers Aurora's marketing team sees every month. On the surface, performance looks reasonable and the campaign is performing adequately.

Here's what's actually happening underneath.

Quantifying the data loss

Based on Aurora's audience mix – over half mobile, a third on Safari, with normal levels of ad blocker usage on desktop – a realistic data loss estimate sits in the 30% range. This is consistent with industry benchmarks (for the purpose of this example, we’ll ignore that real recovery rates seen across comparable Tiide accounts are actually often higher than 30%… we’re being conservative here).

Real conversions occurring: Approximately 429
Conversions being reported: 300
Conversions going uncounted: 129

That's 129 real customers completing real purchases every month that Aurora's ad platforms never find out about.

What this does to reported metrics

If the true conversion count is 429 rather than 300, every metric calculated from that 300 figure is also wrong.

True CPA: $15,000 spend ÷ 429 actual conversions = $34.97

Compare that to the reported CPA of $50. Aurora's marketing team believes they're paying $50 to acquire a customer. They're actually paying closer to $35 – a 30% gap between perceived and actual efficiency.

True revenue and ROAS: Reported revenue, based on the 300 counted conversions, is 300 × $80 = $24,000 against $15,000 spend – a reported paid-channel ROAS of 1.6x.

True revenue, based on the 429 conversions that actually occurred, is 429 × $80 = $34,320. True paid-channel ROAS is $34,320 ÷ $15,000 = 2.29x.

Aurora's marketing team is underestimating their own paid channel performance by more than 40%. The campaigns are working better than anyone in the business believes.

The compounding effect on Smart Bidding

This is where the cost moves from a reporting inaccuracy into an active performance problem.

Google's Smart Bidding and Meta's Advantage+ optimize based on the conversion signals they receive – in this case, 300 reported conversions rather than the 429 that actually occurred. The algorithm is learning from a dataset that's missing nearly a third of the evidence.

In practice, this means the algorithm:

• Under-bids on audience segments disproportionately represented in the missing 129 conversions – likely Safari and mobile users, who in this case make up over half of Aurora's traffic.
• Allocates budget toward channels and placements that appear to convert well in the visible data, even if the invisible conversions tell a different story.
• Takes longer to exit the learning phase, because it has less signal per dollar spent to learn from.
• Produces a less accurate model of "who converts". weakening lookalike audience quality and broad targeting performance.

Conservative estimates from advertisers who've implemented server-side tracking suggest bidding efficiency improvements in the order of 10–20% once the algorithm receives complete signal. Applied to Aurora's $15,000 monthly spend, even a 10% efficiency improvement represents:

$15,000 × 10% = $1,500 per month in improved bidding efficiency

That's $1,500 a month – not from spending more, but from the existing budget working harder because the algorithm can finally see what it's optimising for.

How this affects creative decisions

Six months ago, Aurora's team paused a video ad that was reportedly underperforming – a $52 cost-per-result against a $50 target, compared to $38 on their best-performing static ad. The team concluded the video had fatigued and commissioned a replacement.

Production cost for the new creative: $4,200 (agency fee, filming, editing).

Here's the problem: the original video was disproportionately reaching mobile Safari users – a younger, more engaged demographic that responds well to video content but converts on longer purchase cycles, often beyond Safari's seven-day cookie window. A meaningful share of that video's actual conversions were never attributed to it. Its true cost-per-result was likely much closer to the $38 benchmark than the reported $52.

The team didn't pause an underperforming creative. They paused a working one – and spent $4,200 to solve a problem that didn't exist.

This is a one-off cost in this example, but it's not a one-off pattern. Every business running active creative testing and rotation is at risk of repeating this mistake on a regular basis when tracking is incomplete.

The retargeting gap

Aurora runs a retargeting campaign targeting site visitors who didn't convert, budgeted at $2,000 a month with a reported CPA of $28.

Under client-side tracking, retargeting audiences are built from pixel-fired events. Safari's blocking of third-party cookies means visitors browsing on Safari – roughly a third of Aurora's traffic– are largely excluded from the retargeting pool entirely. Aurora is retargeting a meaningfully smaller audience than the people who actually engaged with their site.

Two costs stack here:

Wasted impressions on existing customers. Without complete purchase event data, some buyers don't get excluded from retargeting and continue receiving ads for a coffee subscription they've already bought. Estimated waste: 5–10% of retargeting spend, or roughly $100–200 per month.

Under-utilized audience pool. The retargeting campaign is working with a smaller, less representative pool than it should have, meaning the $2,000 budget reaches fewer of the right people – a less measurable but real efficiency loss.

A summary of costs

Here's what the broken pixel is costing Aurora Coffee Co. every month, conservatively:

To be clear about what these two figures mean: the $10,320 in unrecognized revenue was already being earned – it's not new money, it's money that existed but wasn't visible, which means decisions weren't being made with the benefit of knowing it was there. The $2,300–2,400 in direct waste (inefficient bidding, unnecessary creative spend, retargeting waste) is the actual cost of operating on incomplete data, and it compounds every month it isn't addressed.

What changes with server-side tracking

Implementing server-side tracking doesn't change Aurora's actual business performance. It changes what's visible.

Within the first few weeks, Aurora's reported conversions would likely rise from 300 toward something closer to the true 429 figure – not because more people are buying, but because the platforms can finally see what was always happening.
Reported CPA drops from $50 toward $35. Reported ROAS improves to reflect the 2.29x true paid-channel return. The Smart Bidding algorithm, now working with a more complete dataset, begins recovering a portion of that estimated 10% efficiency gain over the following weeks as it relearns on better signal.

Costly creative decisions becomes avoidable going forward – future creative evaluations are made against complete data, reducing (though not eliminating) the risk of cutting something that's actually working.

For a relatively modest monthly cost, Aurora recovers visibility into roughly $10,000+ in revenue that was already being generated, and unlocks somewhere in the order of $2,000+ a month in efficiency gains across bidding and retargeting – gains that compound as the algorithms continue learning from more complete data over time.

The takeaway

This example uses a $15,000 monthly spend, but the mechanics scale up and down proportionally. A business spending $5,000 a month sees smaller absolute numbers but the same percentage impact. A business spending $50,000 a month sees the same dynamics at ten times the scale – meaning the dollar cost of inaction is proportionally larger, not smaller.

The pattern holds regardless of size: a meaningful share of real conversions go uncounted, the algorithms optimizing your spend are working from incomplete signal, creative and budget decisions get made on a partial picture, and all of it compounds month after month until someone fixes the underlying measurement problem.

The fix itself is comparatively small – with plans from $24USD per month.

Calculate your own estimated data loss →

*The numbers in this case study are real but the client name has been changed.

The data loss rate used in this example – 30% – is a conservative rate based on industry benchmarks. Your actual rate will vary depending on your browser mix, device split and audience demographics. The only way to know your true number is to measure it.