Unpredictable ad results usually start before the first campaign. The three decisions to make first, how to put numbers on them, and how they show which lever to pull.
Spend goes up and enquiries go down. Or enquiries go up and sales stay flat. When that happens, the usual reaction is to blame the channel: the algorithm changed, clicks got more expensive, Apple's privacy changes broke the tracking. Those things are real, but they are rarely the root cause.
Most performance marketing fails because three decisions are never made before the first campaign goes live: what a new customer is worth to you, what counts as a good enquiry, and how you will know whether ad spend is turning into sales. Without those answers every campaign is an educated guess, and even excellent execution cannot make the results predictable or explain them when they change.
The cost of skipping this is not only wasted spend. If nobody knows what a customer is allowed to cost, you cannot tell a good month from an expensive one, and every budget discussion turns into a debate about opinions. This article sets out the three decisions, shows how to put numbers on them from your own figures, and explains how they tell you which lever to pull when results slip.
Channel problems and framework problems look alike in a monthly report, but they behave differently over time. A channel problem usually has a date: something changed, and results moved with it. A framework problem follows you from agency to agency and from channel to channel, because it was never about the channel.
Run through this list honestly. If two or more of these sound familiar, the problem sits before the campaigns, not in them:
Start with money, not media. Customer acquisition cost (CAC) is what it costs you to win one new customer, including advertising, agency fees and tools. The useful question is not what your CAC is today but what it is allowed to be, and the answer comes from the gross profit a customer brings you, not the revenue.
Here is a worked example with round, illustrative numbers. Say you sell a business service and the average new customer brings €12,000 of revenue in their first year at a 50% gross margin, so €6,000 of gross profit. If you are willing to spend up to a third of first-year gross profit to win a customer, your ceiling is €2,000 per customer.
Now work backwards through your sales process. If one in five qualified enquiries becomes a customer, you can pay up to €400 for a qualified enquiry. If half of all enquiries turn out to be qualified, the most you can pay for any enquiry is €200, and that is a number your campaigns can actually be steered by.
If you sell to consumers, the logic is the same with different inputs: average order value, the margin left after product and delivery costs, and how often a customer buys again in the first year. A shop whose customers reorder every two months can afford to pay far more for a first order than one whose customers buy once, even if the first orders look identical.
How much of that gross profit to spend is a business decision, not a marketing one. A company with healthy cash reserves might accept customers who only pay back after a year or more; a company watching its cash closely might want each first order to cover its own cost. Either is defensible, as long as it is written down and everyone works to the same number.
The second decision is a written definition of a qualified enquiry, agreed between whoever runs marketing and whoever handles sales. In most businesses this definition exists only in the sales team's heads, which is how marketing and sales can look at the same week and reach opposite conclusions.
Keep it short and testable. A useful definition names the type and size of customer you want, the problem they need solved, whether they could realistically buy within a sensible time, and anything that rules them out, such as being outside the regions you serve. A furniture maker that supplies hotels might count a hotel group planning a refurbishment within twelve months as qualified, and a private buyer furnishing one room as a welcome sale but not the target.
The definition matters for a technical reason too. Automated bidding on Google and Meta optimises towards whatever you tell it counts as a conversion. If every form fill counts, the systems will find you the cheapest form fills, which often come from the people least likely to buy. Send back the enquiries your sales team marked as qualified, or better still the ones that became customers, and the bidding learns to find more people like them.
The third decision is the measurement chain: the route by which a euro spent on advertising can be traced to a euro of sales. At minimum, every enquiry or order should carry its source into your customer relationship management (CRM) system, where your team records enquiries and deals, and the outcome of each enquiry should be recorded there within days, not months.
Accept from the start that tracking will be incomplete. Safari and Firefox block third-party cookies, many visitors decline consent, and in the European Economic Area Google requires Consent Mode v2 for ads measurement, so part of what the platforms report is modelled rather than counted. Their reports will disagree with each other and with your own records, and that is normal.
The practical answer is one blended figure, checked every month: total marketing spend including fees, divided by the number of new customers in your own records. It ignores which channel claims the credit, so double counting cannot inflate it. If that blended cost per customer stays under the ceiling from decision one, the system is working, whatever each platform says about itself.
Most briefs open with the questions that should come last: Google or LinkedIn, video or static images, €5,000 a month or €10,000. Asked first, these are matters of taste. Asked after the three decisions, most of them answer themselves.
Take the example above, with its ceiling of €400 per qualified enquiry. If a LinkedIn test produces qualified enquiries at €550 and they close at the same rate as the rest, LinkedIn is too expensive in its current form, however good the conversations feel. If Google search produces them at €250, it deserves more budget until the cost of each additional enquiry approaches the ceiling.
The framework also tells you how big a test needs to be. Automated bidding learns from conversions, so a campaign that records three enquiries a month has almost nothing to learn from, and its results are mostly noise. You will usually learn more from two well-funded campaigns than from six thin ones spread across every channel at once.
The real payoff comes when results slip. Because the measurement chain records every step from click to sale, you can see which step changed instead of concluding that the ads are not working:
Each of these points to a different person and a different fix. Without the framework, all four look the same in a monthly report: fewer sales for the same spend, followed by a long meeting about changing channel or agency.
This is why scealy puts this groundwork in place before running any campaign: it is the only reliable way to tell a good month from a lucky one.
The question isn't which channel to run, but whether you have a system that can tell you whether any channel is working.