Paid Media Managed as a Portfolio

Because automated bidding optimises toward whatever you have labelled a conversion, and it does so with real efficiency — so a badly defined conversion means the platform systematically buys the wrong traffic and gets progressively better at it. In the accounts we audit, some meaningful tracking flaw is present more often than not: conversions firing on form page views rather than submissions, thank-you pages counted repeatedly on refresh, phone conversions counting every connected call including wrong numbers, and duplicate tags double-counting. Each makes reported performance look better than reality while actively degrading the account, which is a bad combination because the numbers give nobody a reason to investigate. Until this is trustworthy, every decision about bids, budget and keywords is being made against fiction.

If the budget can only support one properly, start with search — it captures existing demand, so it converts more efficiently, requires no creative production pipeline, and tells you within weeks which messages and keywords actually work. That knowledge then makes everything else cheaper, including your organic content strategy. Move into social once search has plateaued against available query volume, which is a real ceiling: you cannot spend more on search than there are people searching, no matter how much budget you add. That plateau is the signal that you need demand generation rather than demand capture. The exception is a genuinely new category or product where nobody is searching yet, in which case search has nothing to capture and social is the only option available.

Partly by standardising what can be standardised, and partly by being explicit that one number cannot settle it. The standardisable part matters and is frequently missing: one conversion definition, one attribution window, deduplicated across channels so a lead touching both is not counted twice, reported as cost per qualified lead everywhere. That removes most false comparisons. What it cannot resolve is that last-click attribution systematically favours the demand-capture channel, because someone influenced by three social ads who then searches your name arrives credited to branded search. Judged purely on last click, social usually loses that comparison and gets cut — which is exactly the wrong outcome, since what gets cut is the thing feeding the channel that looks efficient. Where budget allows we resolve this with holdout testing; where it does not, we read branded search volume as a demand-generation indicator and say plainly that part of the judgment is inference.

We tell you before taking the account, and this happens more often than the industry norm would suggest. The clearest case is auction economics: if your category's clicks cost sixty dollars and your close rate and average order value cannot support that, no amount of management skill fixes the arithmetic, and taking the retainer would mean reporting on impression share while you lose money. The second case is budget scale — automated bidding needs roughly fifteen to thirty conversions monthly per campaign to optimise reliably, and below that threshold you are paying management fees on an account that cannot learn. In both situations the honest recommendation is usually a narrower campaign on a small set of terms you can genuinely win, with the remaining budget going to organic or a different channel entirely.