The Fastest Way to Cut Wasted Ad Spend by Channel

A practitioner method for finding dead channels using your GA4 export, plus how the same audit looks in Triple Whale, Northbeam, and a one-off causal read.

Updated 8 September 2026 · Joris van Huët, founder, Causality Engine

The fastest way to find wasted ad spend is to pull your GA4 export, put each channel's platform-reported ROAS next to its causal incremental ROAS, flag the channels whose incremental contribution confidence interval includes zero, check that the channel is measurable at your scale at all, then cut those in steps and re-read after two weeks. No new pixel. No engineering ticket. You are working with data you already have.

That is the whole method. The rest of this is how to run it without fooling yourself, and how the same exercise plays out depending on whether you already have a pixel-based platform running or you just want one clean read.

Two numbers to write down before you cut anything

Start from the fact the dashboards are built around: the party that sells you the advertising also measures whether it worked, and nobody audits the result. That is not fraud. It is an incentive structure, and it does its work without anyone being wicked. So keep one number the seller does not produce, your commerce platform's orders, and compute two ratios against it.

The first is the claim ratio: every platform's claimed conversions added up, divided by the orders you actually shipped. Anything above 1 is the amount by which your suppliers collectively believe they did more work than exists. Track its movement monthly. When it rises, the platforms have become more generous with themselves, and any performance improvement you are reading is partly definitional.

The second is coverage: the conversions your analytics could attribute to any source, divided by the orders your commerce platform recorded. If coverage fell from 0.8 to 0.6 over a year, your reported cost per acquisition rose a third from the decay of your own visibility alone, before any change in media price or conversion rate. In our own analytics warehouse, 48.6% of sessions between 17 October 2025 and 2 September 2026 had no resolvable source. That is one company's census, not a benchmark. Yours will differ, which is the point of computing it.

Neither number needs a vendor. Both take an hour. Every step below reads differently once you hold them.

Why platform numbers keep sending people the wrong way

Every ad platform reports conversions under its own attribution rules. Meta counts a conversion it touched. Google counts one it touched. GA4's data-driven model splits credit across the path it observed. All of those are real measurements. The problem is that they answer "which channel got credit?" and the question you actually need answered in a budget meeting is "how much revenue disappears if I turn this channel off?"

Those are different questions. A customer can see three ads, get an email, and buy because they were always going to buy. Credit gets handed out. Nothing was caused.

I have watched brands cut a channel because last-click called it dead, then watch revenue sag two weeks later because that channel was doing upper-funnel work nobody was crediting. One of Causality Engine's published cases describes exactly this: last-click said cut Pinterest, the causal read said keep it live. Cutting on reported credit alone is how you build attribution debt.

The method, step by step

1. Export the right period from GA4

You want a historical window long enough to hold signal. Pull date, channel or source/medium, spend where GA4 has it, purchase events, revenue, and any campaign identifiers the export carries. If you have geography, product category, promotions, or seasonality markers available, keep them. Those act as controls.

GA4 can export raw, unsampled event data to BigQuery, with daily date-partitioned events_YYYYMMDD tables. If you are running the Causality Engine workflow, you skip the warehouse and hand it a GA4 CSV instead. You pick the period, upload the file, and get a per-channel read back. A longer history produces a cleaner one. You can add a Shopify export, but GA4 already carries the conversion events, so the CSV on its own is enough.

2. Build two views of every channel, side by side

Do not blend the numbers. Put them next to each other.

Channel-level view	What it measures	How to use it
Platform-reported result	Conversions, revenue, or ROAS the platform claims	Where the platform says credit belongs
Causal result	Estimated incremental revenue the channel caused	Revenue attributable to the channel, not merely associated
Confidence interval	The range of uncertainty around the causal estimate	Whether the lift is distinguishable from zero

The gap between the first two columns is where the waste hides. When platform-reported ROAS is high and causal incremental ROAS is low, the platform is claiming credit for demand it did not create.

Here is what a filled-in read looks like. These are illustrative figures, not a benchmark you should expect to hit:

Channel	Platform ROAS	Causal incremental ROAS	Incremental CI	Decision
Meta	4.2x	2.1x	Above zero	Keep, maybe optimize
Google	3.8x	4.7x	Above zero	Protect or test an increase
TikTok	2.9x	1.4x	Includes zero	Reduce gradually, re-read
Email	1.8x	3.2x	Above zero	Do not cut on platform ROAS alone

Notice email. Platform ROAS makes it look weak. Causal contribution says it is one of the strongest things in the mix. That inversion is common, and it is the exact reason blended-only reads mislead. If you want to understand why the reported number and the incremental ROAS diverge this hard, that gap is the whole story.

3. Read the interval, not the point estimate

The decision rule that matters:

If a channel's incremental-contribution confidence interval includes zero, the current data does not clearly establish positive incremental lift. That channel becomes a candidate for a controlled spend reduction.

This is more defensible than cutting on a low point estimate. A channel showing 1.5x incremental ROAS with a very wide interval deserves more caution than one showing 1.1x with a tight interval clearly above zero. The tight one is doing predictable work. The wide one, you do not actually know yet.

Language discipline here saves you in the budget meeting:

• Say "the interval includes zero," not "the channel does nothing."
• Say "the data does not clearly establish positive lift," not "the channel is worthless."
• Apply the rule to incremental contribution, where zero is a natural reference point, not mechanically to a raw ROAS interval.

One more discipline, and it is the one that separates measurement from theatre. An interval that includes zero can mean the channel does nothing, or it can mean the effect you are looking for is smaller than your data can distinguish from noise. The floor is arithmetic, not a vendor guideline. Multiply the channel's share of revenue (spend divided by revenue) by the return you would honestly defend for it; that is roughly how much total revenue would move if the channel stopped. Compare it with the smallest lift your data can tell apart from noise. For a typical DTC brand with six months of history and an eight-week test that is about 8%. If the first number is smaller than the second, no method can answer the question at your scale, ours included, and the right move is to manage that channel on judgement, openly, rather than on a null result read as a verdict.

4. Cut in steps, then re-read

Rank channels by evidence of positive contribution. Mark the zero-containing ones that clear the floor as reduction candidates. Then:

1. Reduce a candidate in measured steps rather than switching it off cold.
2. Hold the rest of the media mix as stable as you can.
3. Write down the date, the spend change, and anything else that moved: promos, pricing, big creative or audience swaps.
4. Re-read after two weeks.

The two-week checkpoint is an operating cadence, not a law of physics. Low-volume channels, long consideration cycles, and seasonal demand all need more time. Treat it as a first look.

What you are watching for:

• Revenue holds while spend falls. The channel likely carried waste or was cannibalizing demand you would have captured anyway.
• Revenue drops materially. Restore or reassess. That channel was doing real work.
• The next interval moves decisively above zero. Your first read was too uncertain, or the cut crossed a real demand threshold.
• The interval still includes zero. The case for reallocating that budget gets stronger.

I am not going to hand you a savings percentage. Anyone who promises "recover 30% of wasted spend" before seeing your account is guessing. How much you can pull depends on your mix, your demand conditions, and how much measurement uncertainty you are sitting on.

The same audit in Triple Whale, Northbeam, and a one-off causal read

The method above is portable. What changes is your starting point and what the tool is built to answer.

Pixel-based platforms: Triple Whale and Northbeam

Both run on a pixel that continuously collects first-party customer-journey and order data, then assigns conversion credit using attribution models you select.

Triple Whale's measurement layer is built around the Triple Pixel, which it describes as identity-resolution technology using first-party and server-side data. Its models include Triple Attribution, Linear All, Linear Paid, First Click, Last Click, Clicks and Deterministic Views, and Total Impact. Triple Whale labels these as attribution models. It also sells incrementality and marketing-mix capabilities inside its Enterprise tier, so do not treat every Triple Whale output as either causal or non-causal without naming the specific method producing the number. The pixel has to be installed across the store, including the order-confirmation page.

Northbeam works on three inputs: first-party pixel data, server-to-server order data, and platform spend. Its Clicks and Deterministic Views model matches verified ad-platform events to first-party pixel and order data using identifiers like order ID and hashed email. It documents multi-touch attribution, platform attribution, and modeled-view variants. The pixel goes in the site header or through Google Tag Manager, plus theme and post-purchase scripts on Shopify. Northbeam warns that orders which do not fire the pixel show up as unattributed, so tracking coverage directly affects what you see.

For the wasted-spend audit, both look like this: install and maintain the pixel, let it collect, pick a model, monitor attributed ROAS by channel, flag candidates, and then run a separate incrementality or causal analysis before you call a low-credit channel proven waste. The pixel tells you what interactions happened. It does not, on its own, establish the counterfactual.

One-off causal read: Causality Engine

This starts from history you already have. You export GA4, upload it, and get a per-channel causal read: incremental ROAS, causal contribution, confidence intervals, and a platform-reported versus causal comparison for every channel. The question it is built to answer is counterfactual. What revenue would have happened without each channel? No pixel, no SDK, no DNS change, no developer ticket.

Criterion	Causality Engine	Triple Whale	Northbeam
Measurement approach	Causal inference on GA4 history	Pixel-based multi-touch attribution (incrementality in Enterprise)	Pixel-based multi-touch attribution
Primary input	GA4 CSV (Shopify optional)	Triple Pixel first-party data	Northbeam pixel + order + spend data
Pixel required	No	Yes	Yes
Confidence intervals per channel	Yes	Not publicly confirmed for standard models	Not publicly confirmed
Reported vs causal comparison	Yes, per channel	Not publicly confirmed	Not publicly confirmed
Operating pattern	One-off read, or continuous Pro	Ongoing	Ongoing
Entry price	€99 per read, or €299/mo Pro	Free tier; paid from $219/mo, structure varies	Listed plans require contact; e.g. Professional $3,500/mo

The pricing lines are not directly comparable and I would not pretend they are. The €99 Causality Engine read is a one-time audit. Triple Whale and Northbeam are ongoing measurement products with pixel installation, continuing data collection, and different plan and contract structures. Triple Whale's public page shows more than one pricing display and mentions a possible 12-month commitment on paid plans. Northbeam's listed plans route through sales, so it is not a self-serve comparison to a €99 read.

For the specific job in this article, pull what already happened and decide where to cut, the one-off causal read is the cleanest fit. You are not standing up new tracking to audit last quarter. You are reading last quarter directly.

One more thing, and it applies to us as much as to the two products above. Between 14 August and 2 September 2026 we audited thirty-one commercial measurement vendors for a published validation of their method against randomised experiments, with the sample, the design and the discrepancies disclosed. We found none. Hold Causality Engine to the same question. The read states its counterfactual and its interval, and it is a model on observational data, not an experiment. Ask us, and every other vendor, which of your channels are not measurable at your current spend. The honest answer is a list, and if you are not handed one you have learned something anyway.

What I would do first

If you have €5,000 or more in monthly paid spend and meaningful GA4 history, run a single causal read before you touch a pixel platform decision. That fit floor is Causality Engine's own stated guideline, and it is honest about why: below that spend, the model may not separate signal from noise tightly enough to defend a budget move. It is a vendor guideline, not an industry standard. The arithmetic in step 3 is the real test, and it applies to every method, including ours.

Then put reported and causal side by side, find your zero-containing channels, and make one staged cut. Do not cut four things at once. You will never learn which change did what.

If you already run Triple Whale or Northbeam for daily monitoring, keep them. They are good at campaign diagnostics, creative comparison, and ongoing budget management. Add a causal read as the layer that tells you whether a low-credit channel is genuinely disposable or just under-credited. The two layers do different jobs.

FAQ

Does a confidence interval that includes zero mean the channel is useless?

No. It means the current data does not clearly separate positive lift from no lift. The channel might be working and just be too noisy to prove at your volume, or it might be genuinely marginal. That is why the method calls for a staged cut and a re-read rather than an immediate shutoff. You reduce spend to gather cleaner evidence, not to punish the channel.

Are platform-reported conversions fake?

No, and calling them fake misreads the problem. They are real measurements under each platform's attribution rules. The issue is that credited conversions and incremental conversions answer different questions. A channel can honestly receive credit for conversions it merely touched without having caused them.

Can I just use Triple Whale or Northbeam for this and skip a separate causal read?

You can start there, but their standard outputs are attribution models, not counterfactual estimates. Triple Whale offers incrementality inside Enterprise and Northbeam documents multi-touch attribution, so the right move is to name the specific method producing your number. If you want a clean causal contribution with a per-channel interval and you are auditing history you already have, a one-off GA4 read gets you there without installing anything.

How long should I wait before re-reading after a cut?

Two weeks is a practical first checkpoint, not a guarantee. If the channel has low conversion volume, a long purchase lag, or strong seasonality, you will need longer before the next read means much. Watch whether revenue holds as spend falls, and give slow channels more runway before you decide.

Sources and further reading

• The Price of Being Found (Causality Engine, Edition 2.10, September 2026): the coverage rate and the claim ratio (Chapters 8 and 9), the measurability floor (Chapter 15), the vendor audit (Chapter 17)
• Incremental ROAS explained
• True ROAS versus reported ROAS
• Geo testing for incrementality
• Are my Meta ads actually working?
• Sixty-second versions of these ideas on YouTube Shorts

Vendor prices and features quoted in this article were taken from each vendor's own website on 8 September 2026 and may have changed since. Check the vendor's pricing page before relying on a figure.