Splitting ad budget without reliable attribution: how to decide

Consent rules, browser restrictions and platform self-reporting have left most shops without a channel-level number they fully believe. The common responses are both bad: spend everything where attribution still works, which over-invests in the bottom of the funnel, or ignore measurement entirely and go by feel.

There is a middle path that uses the evidence still available, which is more than it first appears.

Start with the baseline you already have

Before allocating, know what happens when you do nothing. Most shops have never measured this and it changes the whole conversation.

1. Total revenue by week, for a year, from orders.
2. Direct and organic revenue as a share of it. This is roughly your floor — what arrives without paid media.
3. New versus returning customer revenue. Paid media mostly buys the first; email and retention mostly grow the second, and confusing them is how budget gets misallocated.

A shop where 70 % of revenue is returning customers has a different optimal budget from one where it is 20 %, regardless of what any platform reports.

Buy evidence with holdouts

• Geographic holdouts for broad channels: turn a channel off in one comparable region and compare total revenue per customer, not platform conversions.
• Audience holdouts for anything addressable — email, flows, customer match.
• Run them one channel at a time, for at least a month, and long enough to cover your purchase cycle.
• Accept a wide margin of error. "Somewhere between 20 and 50 % of what we thought" is still a far better input than a platform figure you know is inflated.

Think in marginal terms, not average

The question is never "what is this channel's ROAS". It is "what happens to revenue if I move the next €1,000 here". Those have different answers, because channels saturate.

1. Change one channel's budget by 20–30 % and hold everything else steady for a month.
2. Measure total business revenue, not the channel's reported revenue.
3. Keep going while the increase pays; stop when it does not. This is a slow, unglamorous loop and it beats optimising against reported ROAS.
4. Watch frequency and cost per acquisition as the early signs of saturation, before revenue shows it.

Four rules for the part you cannot measure

• Protect the retention budget first. Email and flows to existing customers are the cheapest and most measurable revenue you have. Fund them fully before arguing about paid media.
• Keep a deliberate share in channels measurement cannot follow. If you only spend where you can measure, you will end up spending only on people who were going to buy anyway.
• Do not reallocate on a single month. Noise is larger than most of the differences being argued about.
• Write down the reasoning when you change a split, so next quarter you can tell whether the decision was right or merely followed by a good month.

What to fix before spending more

Two things return more than any reallocation and are routinely skipped:

1. Suppress existing customers from acquisition campaigns, refreshed daily from real orders. Paying to acquire people you already have is the most common waste in ecommerce advertising, and it is a configuration change.
2. Fix the retention flows. A cart recovery, post-purchase and replenishment programme that works is worth more than a 20 % budget increase anywhere, and its effect is measurable with a holdout.

Both improve the honest numbers rather than the reported ones, which is the only kind of improvement that survives the next platform change.