Audience targeting has largely been automated away. What still separates a campaign that pays back is the work the platform cannot do for you.
A decade ago, media buying skill meant knowing which audiences to stack and which to exclude. Most of that is now done better by the platform than by a person. The lever that remains — and the one that still separates campaigns that pay back from campaigns that quietly drain — is the creative and the offer behind it.
This is not a claim that account structure no longer matters. It is a claim about where the remaining marginal return sits, and it has moved.
The platform learns from what you feed it
Automated bidding optimises towards the signal you give it. Give it a weak conversion event — a page view, a newsletter signup, a form start — and it will find you cheap versions of a worthless action, very efficiently. Give it a revenue-weighted signal and the same machinery works in your favour.
This is why measurement is a media decision, not a reporting one. The event you optimise towards is effectively your strategy, stated in a way the system can act on. Businesses that pass back actual order value, or a qualified-lead event fired by the CRM rather than by the thank-you page, consistently outperform those optimising to whatever was easiest to implement.
Where volume is too low to train on the real event — a handful of qualified leads a month — optimise to the nearest reliable proxy further up the funnel and check the relationship holds. The failure mode is optimising to a proxy that stops correlating with revenue and never noticing.
Run creative as a testing programme
Treat creative like a pipeline rather than a project. A steady flow of distinct concepts — not fourteen crops of the same asset — tested against each other, with the losers retired and the winners iterated on.
The variable that matters most is the concept, not the execution. Two versions of the same idea with different colours will tell you almost nothing. Two genuinely different arguments for why someone should buy will tell you a great deal.
Build a concept library rather than a campaign calendar. Most accounts have three or four arguments that reliably work — price, speed, risk removal, status — and the productive question each month is which argument to express next, not what to post.
The offer is part of the creative
A great deal of creative testing is really offer testing that nobody labelled as such. Changing "book a consultation" to "get a fixed quote in 24 hours" is not a copy change; it is a different promise with different delivery implications, and it will usually outperform anything achievable by rewriting a headline.
This is the part an agency cannot do alone, because the offer is a commercial decision. It is also the highest-return conversation available in most paid accounts, and it tends not to happen because it does not look like media work.
Beware the dashboard
Platforms are extremely good at reporting a number that looks like success. The question that matters is incrementality: would this revenue have happened anyway? Brand campaigns that appear to have a spectacular return are frequently taking credit for demand you already had.
You do not need a perfect answer. Geo holdouts and periodic pauses will tell you more than another attribution model will. Turn a campaign off in two comparable regions for three weeks and watch total revenue rather than platform-reported conversions — it is crude, and it is real evidence.
Watch branded search alongside spend, too. If a campaign is genuinely creating demand, it shows up as people searching your name; if it is only harvesting demand, it does not.
What good looks like month to month
A healthy paid programme produces a small number of new creative concepts, a clear record of which were retired and why, a stable and trusted conversion signal, and at least one incrementality check a quarter. It does not produce a weekly report of platform-reported ROAS with no commentary.
If your paid media reporting has looked identical for six months, the account is being maintained rather than improved, and the returns will decay quietly as creative fatigues.
Common questions
- How often should you refresh ad creative?
- When performance decays rather than on a fixed schedule, which in practice means watching frequency and cost per acquisition on each concept and retiring it when both deteriorate together. High-spend social accounts often need new concepts every few weeks; lower-spend search accounts can run far longer. Refreshing on a calendar wastes concepts that are still working.
- What conversion event should you optimise towards?
- The one closest to revenue that produces enough volume for the platform to learn from — ideally actual order value, or a qualified-lead event fired by your CRM rather than by a thank-you page. Where volume is too low, optimise to the nearest reliable proxy and verify periodically that the proxy still correlates with revenue.
- How do you measure whether paid advertising is incremental?
- By withholding it somewhere and comparing. Geo holdouts — pausing a campaign in comparable regions while it continues elsewhere — and scheduled pauses measured against total revenue rather than platform-reported conversions are the most practical methods for most businesses. Attribution modelling alone cannot answer the question, because it only sees the traffic the platform touched.