Nearly every acquisition report we are handed contains a number that looks precise and is not true. Not because anyone is being dishonest — because the default measurement in every ad platform answers a slightly different question from the one you think you asked.
The question the platform is actually answering
When your ad platform reports a conversion, it is answering: "did someone who touched this ad go on to convert?"
The question you care about is: "would that person have converted anyway?"
These are wildly different questions, and the gap between them is where most acquisition budget goes wrong. No amount of dashboard sophistication closes it, because the platform cannot observe the counterfactual. Only an experiment can.
Who this systematically flatters
The bias is not random, which is what makes it dangerous. It reliably favours channels that intercept demand near the point of purchase.
Branded search is the clearest case. Someone who already decided to buy from you searches your name, clicks the ad sitting above your own organic listing, and converts. Last-touch credits that channel with the sale. In many cases the same person would have scrolled two centimetres and clicked the free result.
Retargeting is close behind. You are showing ads to people who already visited your site — a population selected precisely because they were interested. A meaningful share convert regardless, and the ads collect credit for it.
Meanwhile the channels that created the demand — the content someone read three weeks ago, the podcast they heard, the comparison page they found — are credited with nothing, because they were not the last thing touched. So budget shifts away from them, demand creation slows, and six months later the harvesting channels get more expensive because there is less demand to harvest. The failure mode is slow and looks like the market getting harder.
Optimise hard enough toward last-touch and you will eventually be spending your entire budget buying back customers you already had.
The other error: measuring the wrong outcome
The second problem is subtler. Even with perfect attribution, most accounts are optimising toward the wrong event.
Ad platforms optimise for whatever you feed them. Feed them form submissions and they will find you the people most likely to submit forms — which is not the same population as people likely to buy. Feed a SaaS platform trial signups and it will locate the cheapest, least serious signups in the market with impressive efficiency.
The fix is to pass the event that actually matters to you:
- Home services — the booked job, with ticket value, not the form fill
- SaaS — activation or first payment, not signup
- Insurance — the bound policy, not the quote request
- Healthcare — the attended appointment, not the booking
- Fitness — the retained member at day 60, not the trial
This nearly always requires server-side conversion tracking and CRM stitching, because the real outcome happens days or weeks later, somewhere the browser cannot see. It is more work to set up. It is also the difference between an ad account that finds you customers and one that finds you form fills.
The uncomfortable test
Take last month's platform-reported conversions and add them up. Now take the number of actual new customers from your finance system. If the first number is meaningfully larger, you have double-counting, view-through inflation, or both — and every efficiency metric built on it inherits the error.
What to do instead
Run a holdout
Turn a channel off in some regions and leave it on in others. Compare total conversions, not attributed ones. It is the only method that answers the counterfactual question directly, and it is far cheaper to run than most people assume.
Branded search is the natural first candidate, because the result is frequently dramatic in one direction or the other and the test costs almost nothing. Sometimes it turns out to be genuinely defensive against competitors bidding your name, and worth every dollar. Sometimes it is paying for traffic that was already yours.
Report blended, not just platform
Total acquisition spend divided by total new customers. Crude, immune to attribution games, and the number your finance team is already using whether or not marketing quotes it. When blended CAC and platform CAC diverge, platform CAC is the one that is lying.
Match the measurement window to the buying cycle
A 7-day conversion window on a product with a 60-day evaluation cycle will make every upper-funnel channel look worthless. Set windows against your actual sales cycle, which you can measure from your own CRM data rather than guessing.
Accept some uncertainty
Perfect attribution does not exist and pursuing it wastes real money. The goal is not a precise number; it is being roughly right about direction — which channels are creating demand, which are harvesting it, and whether the overall machine is profitable.
Rough and honest beats precise and wrong. A business that knows its blended CAC within 15% and has run two holdouts is in a far better position than one with a beautiful multi-touch model built on last-touch data.