PPC and paid media managed against profit, not clicks

Paid media is rented attention, and that should change how you buy it

Almost every other marketing investment compounds. A page that earns a ranking keeps earning after the writer has been paid; a brand that becomes known stays known for a while. Paid media does not work like that. The traffic stops the same hour the card stops working, which is not an argument against it — rented attention is the fastest way to test whether a proposition survives contact with the market, and often the only realistic way to reach buyers who do not yet know your name. It is an argument for treating paid media as an operating cost measured against margin, rather than as a growth strategy measured against clicks.

That makes the first question arithmetic rather than tactical. What is your gross margin on the thing being sold, what share of enquiries becomes a customer, and therefore what can you afford to pay for an enquiry before the exercise stops making sense? Run those three numbers together and you get a target cost per acquisition, which is the only number that makes a click price meaningful. The same click price can be comfortable for a business selling a considered service and ruinous for one selling a low-margin consumable.

When the arithmetic does not work, our advice is to fix the offer, the margin or the conversion rate before renting traffic to expose them. We would rather say that than take a management fee to run something structurally doomed, and it is the reason our first conversation is usually about your close rate rather than your keywords.

What an audit of an existing account actually finds

We have never opened an inherited account and found nothing. The findings are boringly consistent, and almost none of them are exotic.

Conversions counted more than once. A form submission tracked in Google Ads, again through an imported Google Analytics 4 event, and again as a thank-you page view — so reported conversions run at double or triple the CRM figure, cost per acquisition looks excellent, and the bidding is optimising toward an inflated signal.

Brand search flattering the average. Campaigns bidding on your own company name mixed into the same reporting as prospecting campaigns. Brand terms convert at a rate no acquisition campaign can match, because those people were already looking for you. Blended together, they hide whichever campaign is actually losing money.

Settings nobody revisited. Location targeting left on presence-or-interest, so you pay for people merely reading about your city. Search partners and display expansion switched on inside a search campaign. Ad rotation set to optimise before there was any data to optimise against.

Automation pointed at the wrong signal. A target cost per acquisition strategy chasing a conversion action that includes newsletter signups, which the system dutifully buys in volume because they are cheap and plentiful.

A landing page nobody has looked at. Frequently the single largest available improvement, and almost always outside the scope the previous agency agreed to.

The search terms report is where the money goes

The search terms report shows the queries that actually triggered your ads, as opposed to the keywords you thought you were bidding on, and it is the most useful screen in the platform. It is also less complete than most advertisers assume. Google omits search terms without enough query activity in order to meet its own data privacy standards, grouping low-volume queries into subthemes or an aggregated bucket without exposing the queries themselves. You will therefore never see every term you paid for.

Negative keywords have limits worth knowing too. They do not match close variants, so excluding a plural does not exclude the singular, and each form has to be added deliberately. A negative can also be bypassed entirely when the offending word falls beyond the sixteenth word of a very long query. The practical consequence is that a negative list is containment, not elimination — and that the structural decisions, match type discipline and campaign separation, are doing more work than the list is.

Nothing launches until the measurement is trustworthy

Automated bidding is a system that learns from the signal you give it. Give it a good signal and it will find patterns no human could hold in their head. Give it a duplicated, over-generous or mistimed signal and it will learn the wrong lesson faster and more expensively than a manual account ever could. So the tracking comes first, every time, and if fixing it delays a launch then the launch is delayed.

Practically, that means every conversion action defined and deduplicated; a clear split between primary actions that bidding optimises toward and secondary actions that are merely observed; a value on each action, even a rough modelled one, so that a demonstration request is not weighted equally with a brochure download; and a test that follows a click all the way through to a record in your CRM before any budget is committed.

For businesses that close deals away from the website, this is where the account is genuinely won or lost. Enhanced conversions improve accuracy for consenting users by sending normalised, SHA-256 hashed first-party data such as an email address back to the platform. Offline conversion imports go further: the click identifier is captured when the visitor arrives, stored with the lead in your CRM, and returned to Google Ads when that lead becomes a customer, along with what it was worth. That single loop is what lets bidding chase revenue rather than form fills, and it is why our paid work sits so close to our analytics and measurement practice.

Structure, match types and audiences

Campaign structure is not an aesthetic preference; it is how you control where money can flow. Intents that behave differently need separating, because a shared budget and a shared bid strategy will always let the easiest conversions absorb the spend. Brand and non-brand, prospecting and remarketing, high-margin and low-margin product lines, and geographies with genuinely different economics all deserve their own containers.

Match types deserve the same deliberation. Exact match has not meant exact for years — close variant matching means the platform decides which queries share your keyword’s meaning — so the search terms report is not an optional review, it is the control mechanism. Broad match can work well, but it works well downstream of trustworthy conversion tracking and a mature negative list, and disastrously upstream of them.

Audiences are a lever most accounts underuse. Customer Match lets you target and, just as usefully, exclude your existing customers using your own data; the lists have real operating rules, including a maximum membership duration of 540 days and a requirement to keep at least a hundred members refreshed within that window, so an audience strategy that is not maintained quietly stops working. Layering audiences as observation before switching them to targeting is the cheap way to find out whether a segment behaves differently at all.

Ads, assets and the tests worth running

Ad Rank is not a bid. Google lists six inputs — your bid, the quality of your ads and landing page, Ad Rank thresholds, the competitiveness of the auction, the context of the search, and the expected impact of your assets — and describes minimum thresholds an ad must clear before it can show at all. That is the mechanism behind an unpopular truth: you cannot reliably bid your way past a weak landing page, because the threshold sits between you and the placement.

Quality Score decomposes part of that into three components — expected click-through rate, ad relevance and landing page experience — each graded against competitors on the same searches over the previous ninety days. It is worth reading Google’s own framing carefully here, because it contradicts a great deal of agency reporting: Quality Score is described explicitly as a diagnostic that should not be optimised or aggregated with performance data. A monthly report leading with average Quality Score is presenting a troubleshooting indicator as an outcome. We use the components to locate problems and report the outcomes separately.

Useful testing follows from that. Test the promise, not the punctuation: a different offer, a different objection answered, a different audience framing. Testing two headlines that differ by a comma consumes impressions and teaches nobody anything. Keep a written log of what was tested and retired, because without one the same experiment gets run again the following year by somebody new.

The post-click half of the account

Half of a paid media account is not in the account at all. It is the page the click lands on, and it is frequently the cheapest thing to improve because so few people are responsible for it.

Message match matters first: the page has to visibly continue the sentence the ad started. A visitor who searched for a specific service, clicked an ad naming that service and arrived on a generic homepage has to do the work of finding their way again, and a meaningful share of them will not. Speed matters next — Google’s guidance is that Largest Contentful Paint should be at or under 2.5 seconds for at least 75% of visits, and paid traffic skews mobile, which is the harder condition. Then form length, where every non-essential field is a reason to abandon, and mobile behaviour, where a form that works on a laptop can be unusable one-handed on a train.

We produce prioritised recommendations for these pages as part of the engagement, and where the pages need rebuilding rather than tuning, that is conversion-focused web design work rather than campaign management. Being clear about which of the two you need is part of the audit.

Shopping, Performance Max and retail

For retailers, the feed is the campaign. Google Shopping does not match on keywords you choose; it matches on the product data you supply, which makes the Merchant Center feed the primary optimisation surface. The product data specification sets out what is required and how — a unique identifier, a title of up to 150 characters that matches the landing page, an accurate description without promotional language, an image meeting the minimum size requirement, a price that agrees with both the landing page and the checkout, and a supported availability value. Feed work is unglamorous and it moves revenue more reliably than bid tinkering does.

Performance Max sits on top of that, reaching Search, Shopping, YouTube, Display, Discover, Gmail and Maps inventory from a single campaign, assembling ads from asset groups and using audience signals to accelerate its learning. It can perform very well. It also concentrates decisions inside the system, which is why we run it with guardrails: brand exclusions so it does not simply harvest people already searching for you and claim the credit, account-level negatives, feed segmentation so that margin tiers are not averaged together, and regular use of the asset, channel and placement reporting rather than accepting the headline figure.

Where you need to know precisely which query cost what, a conventional search campaign remains the better instrument, and we will say so. Retailers running both should also read our notes on organic product and category visibility, because paying for traffic to a page that could rank is a decision, not a default.

Paid social runs on a different mechanism

Paid search captures demand that already exists. Paid social creates it, interrupting someone who was not looking for you, which changes almost everything about how it should be judged. Creative carries the targeting load — the platforms’ delivery systems find the audience if the creative earns attention — so production capacity matters more than audience micro-segmentation, and creative fatigue is a real operating constraint rather than an excuse.

The channels differ sharply in character. Meta Ads Manager offers reach and cheap testing volume. LinkedIn Campaign Manager offers firmographic and seniority targeting available nowhere else, at costs per click that make sense only when a customer is worth a great deal, which is why it suits considered business-to-business purchases and almost nothing else. Attribution is harder across all of them, because the mechanism is influence rather than a click, and a last-click report will always undervalue the channel that created the demand another channel harvested. Where organic presence is also part of the plan, the social media marketing side runs alongside this rather than inside it.

Remarketing, exclusions and frequency

Remarketing is the easiest place in paid media to feel successful while adding very little. The people in your remarketing audience were already going to see you again in many cases, so the reported return flatters itself. It is still worth running, with discipline: exclude converters and existing customers, segment by depth of engagement so a pricing-page visitor is not treated like a bounce, cap frequency to a level you would tolerate yourself, and set a membership duration that reflects your actual buying cycle rather than the default.

Bidding automation, and the guardrails it needs

Manual bidding is not coming back at any meaningful scale, and pretending otherwise is not a strategy. What automation cannot do is decide what it should be optimising for. That remains the human work, and it is most of the job: defining and valuing conversions, maintaining exclusions, deciding which campaigns are separate and why, setting budgets that give a strategy enough conversion volume to learn from, and choosing the geography and the audience.

The commonest failure we see is not choosing the wrong strategy, it is interfering with the right one. Every substantial change to a target resets learning, so an account where somebody adjusts targets at the first sign of a bad few days never accumulates enough stable data to perform. Set the guardrails, then leave them alone long enough for the result to mean something.

Reporting: return on ad spend, cost per acquisition and incrementality

Platforms grade their own homework, and they grade generously. Each one claims a conversion it can plausibly attach itself to, which is why the sum of platform-reported conversions across four channels routinely exceeds the number of customers your finance system recorded. Our reporting reconciles against your CRM for that reason, and states which figures are counted and which are modelled.

The harder question underneath is incrementality: how many of those conversions would have happened anyway? Brand search and remarketing are the usual suspects, because both intercept demand created elsewhere. The only honest way to answer is to test — a geographic holdout, or a structured pause on a defined segment — and to accept that the answer may be uncomfortable. Most accounts never run such a test. The ones that do usually find that at least one campaign they had been proud of was mostly taking credit for demand that already existed, and that finding is worth more than a year of dashboards.

What actually drives the cost of managing paid media

We do not publish figures, because a number produced before the scope is understood is a guess somebody eventually absorbs. The variables are not secret, and knowing them makes it easier to brief and to compare proposals honestly.

The number of platforms and markets. Each additional platform brings its own creative specifications, reporting model and maintenance burden. Two countries in two languages is closer to two accounts than to one.

Whether measurement has to be rebuilt. An account with clean conversion tracking, working consent handling and a CRM connection starts from a very different place than one where the first job is establishing whether any number can be believed.

Account complexity, not budget size. A large-budget account with six well-structured campaigns can be lighter work than a small one with sixty campaigns, three feeds and a seasonal promotional calendar.

Feed and creative production volume. Retail accounts live and die on product data quality. Paid social lives and dies on the rate at which new creative can be produced and tested.

Sales cycle length. Where the outcome that matters happens long after the click, connecting offline results back into bidding is a genuine integration rather than a setting.

The auction itself also sets your budget, and it is set by your competitors rather than by us. Competitive markets cost more per click regardless of who manages the account; what management changes is how much of that spend reaches people who might actually buy.

How this connects to the rest of the work

Paid media rarely makes sense in isolation. It works best as one instrument inside a wider digital marketing programme where the channels are compared on the same measurement basis rather than each reporting its own version of success. It depends on search visibility being handled properly, because paying repeatedly for queries you could own is a choice worth making consciously. And for retailers, paid performance is often limited by the store itself, which puts it next to e-commerce build work rather than next to campaign management.

If you want an outside read on an existing account before deciding anything, that is a reasonable place to start — tell us what you are running and what it is supposed to be achieving, and we will tell you what we would change first.