What founders should measure before increasing paid media spend

The request usually arrives in the same shape. The ads are working, or they look like they are, and the question is how quickly the budget can go up.

It is a fair question. It is also the wrong first one, because raising spend does not change what is happening in a campaign. It buys more of it. If the campaign is producing customers, more budget produces more customers until something saturates. If it is producing activity that resembles customers, more budget produces more of that, and the reporting keeps looking healthy for as long as nobody checks the other end.

So before the number goes up, there are four things worth being able to answer.

1. What counts as a conversion, and can you prove it fired

Not what the ad platform reports. What your business would accept as a real outcome, tracked in a way you have verified yourself by completing the action and watching it register.

This sounds like a formality. In practice it is the single most common reason a scaled campaign disappoints, and the failure modes are boringly consistent:

• The event fires on page load rather than on submit. A thank-you page with the conversion in the page template counts every bounce onto it, including refreshes and back-button returns. Cost per lead looks magnificent.
• The event fires but carries no value. Optimising for purchases without passing value and currency means the platform optimises for count, so it will happily find you many small orders instead of fewer large ones.
• Nothing is attached to the thing people actually do. A WhatsApp button or a tel: link with no event fires nothing, which quietly teaches the platform to stop finding the people who use them.
• The browser event is being dropped. Ad blockers, ITP and consent refusals remove a meaningful share of client-side events. If you are running only a pixel, you are optimising on a sample and do not know its size.

Verify it, do not assume it. Complete the action yourself and watch it arrive: GA4’s DebugView, Tag Assistant, and the platform’s own event manager or test events tool all show it in near real time. Check for duplicates while you are there — a tag firing from both a template and a tag manager will double every conversion.

If the budget is going to be material, move the important events server-side — Meta’s Conversions API, Google’s enhanced conversions, or your own server endpoint — and deduplicate against the browser event with a shared event_id so the same conversion is not counted twice. It is a day of work and it changes what every number after it means.

And test it again after every site release. Tracking does not break loudly.

2. What a customer is worth to you

Not revenue per order. What you keep, and over what period.

A single figure is enough to start: average order value, times contribution margin, times the number of purchases a typical customer makes in a year. It will be rough. Rough is fine — the purpose is not precision, it is having a ceiling. Without one, “our cost per acquisition went up” is a sentence with no consequence attached, because nobody in the room knows which side of profitable it landed on.

Two refinements worth making early:

Use new-customer CAC, not blended. Total spend divided by total orders flatters you with repeat purchases you did not pay for again.

Know your payback period. A CAC you can only justify over eighteen months is a cash-flow decision as much as a marketing one. If you are funding ads out of revenue, the relevant question is not whether the customer is profitable eventually — it is whether they pay you back before the next invoice.

3. What happens after the click

This is the part budget cannot fix.

If the landing page takes six seconds on a mid-range phone on 4G, if the form asks for nine fields, if the product page does not answer the question the ad raised — then more spend is buying more people the chance to leave. The campaign is not underperforming. It is performing exactly as well as the page allows.

Worth measuring before you touch the budget:

• Landing page conversion rate, by device. The blended figure hides a mobile problem behind desktop performance almost every time.
• Largest Contentful Paint and Interaction to Next Paint from field data, not a lab score. Paid traffic is disproportionately mobile and disproportionately impatient.
• Form abandonment — which field they stop at. One required field in the wrong place can cost more than a bad audience.
• Message match. If the ad promised a price and the page opens with a brand story, the bounce is not the visitor’s fault.

Multiplying conversion rate by two costs nothing per click afterwards. Doubling the budget costs double, forever.

4. Whether the demand is yours already

The most flattering campaigns are usually the ones reaching people who were going to buy anyway — your own brand name, your existing customers, people who already had the product in a basket. They report a wonderful return, because the sale was already coming.

There is nothing wrong with running them. The mistake is reading them as growth. Before scaling, separate:

• Branded from non-branded search. They are different businesses in one report.
• New customers from returning ones. Most platforms will report this if you ask them to; in Google Ads it is a bid setting as well as a segment.
• Prospecting from retargeting. A retargeting pool is finite by definition. Its ROAS stays beautiful and its volume never scales.

If the decision is big enough, test rather than infer. A geo holdout — hold out a set of comparable regions for a few weeks and compare total sales, not platform-reported sales — or a platform conversion lift study will tell you how much of the reported return would have happened anyway. It is the only method here that survives an attribution argument, because it does not use attribution at all.

While you are at it, look at where your attribution settings actually sit. Click and view windows, and last-click versus data-driven, change the reported numbers substantially without anything in the business changing at all. Know which settings produced the figure you are about to make a decision on.

The order that tends to hold

1. Verify the conversion is real, and server-side if it matters.
2. Know what a new customer is worth, and how long they take to pay back.
3. Fix the page before the budget.
4. Separate new demand from demand you already had.
5. Then raise spend — in increments you can read.

That last point matters more than it looks. Double a budget and you have changed everything at once: the auction you are in, the audience the platform reaches for, and the pace at which the campaign learns. Large edits push an ad set back into learning, and its performance during that period is not a verdict on anything. Move in steps small enough that the result is still attributable, and leave each step alone long enough to produce a signal rather than a wobble.

What this is not

This is not an argument for spending less. Paid media is the fastest way to find out whether a proposition works, and a business that can profitably buy customers has a genuine advantage over one that cannot.

It is an argument for knowing which of those two you are before you commit more money to finding out.