How online advertising actually works — explained plainly.
AdFire Digital is an independent, no-jargon resource on paid media: how search, social and programmatic ads are bought in real-time auctions, what the metrics really mean, and how to plan, launch and measure a campaign from scratch.
The three channels that carry most of the internet's ads
Nearly every paid-media plan is some mix of these. Each has its own auction, formats and use cases — start with the one that matches your goal.
Search advertising
Text ads on search-engine results pages. You bid on keywords and reach people at the exact moment they look for something — the purest form of demand capture.
Read the guide →Paid social
Ads inside social feeds and stories. Targeting is built on demographics, interests and behaviour, so social excels at creating demand with strong visual creative.
Read the guide →Programmatic display
Banner, video and native ads bought automatically across thousands of sites and apps through real-time bidding, demand-side platforms and ad exchanges.
Read the guide →Every ad is sold in an auction — in about the time a page takes to load
The mechanics differ per platform, but the pattern is the same everywhere. Understanding it explains most of what campaigns do.
1. A slot opens
Someone loads a page, opens an app or types a query. An ad placement becomes available along with anonymous context: the content, the device, rough location and audience signals.
2. Bids are placed
Advertisers whose targeting matches enter an automated auction. Each bid reflects what the click, impression or conversion is predicted to be worth to them.
3. Quality decides
Platforms rarely award the slot to the highest bid alone. Expected relevance and ad quality act as multipliers, so a better ad can beat a richer competitor.
4. The winner pays
The ad renders and the advertiser is charged under that marketplace's pricing rules. Auctions may use first-price, second-price or platform-specific logic, while reporting expresses cost as CPM, CPC or CPA.
The six numbers every campaign is judged by
Dashboards overflow with acronyms; these are the ones that actually connect spend to results. Full formulas and pitfalls live on the metrics page.
| Metric | Formula | What it tells you |
|---|---|---|
| CTR — click-through rate | clicks ÷ impressions | Whether the creative and targeting make people respond. |
| CPC — cost per click | spend ÷ clicks | What you pay to bring one visitor to your site. |
| CPM — cost per mille | spend ÷ impressions × 1000 | The price of reach — one thousand ad views. |
| Conv. rate | conversions ÷ clicks | Whether the landing page turns visits into outcomes. |
| CPA — cost per action | spend ÷ conversions | The all-in cost of one sale, lead or signup. |
| ROAS — return on ad spend | revenue ÷ spend | How much money each unit of ad spend brings back. |
Launching a first campaign, in five grounded steps
The order matters. Most beginner losses come from skipping straight to step four.
Define one conversion
Decide the single action that means success — a purchase, a booked call, a signup — and roughly what it is worth to you. Every later decision traces back to this number.
Pick the channel that matches intent
Sell something people actively search for? Evaluate search. Visual product or new category? Test social. Need inspectable reach across open-web, app, audio or CTV inventory? Assess programmatic.
Set a test budget you can afford to lose
Document the maximum loss, the expected click and conversion range, and the evidence needed for a decision. Do not spread a small budget across more test cells than it can meaningfully observe.
Build tight, honest ads
Match the message to the audience segment, promise only what the landing page delivers, and make the call to action obvious. Relevance lowers your costs in every auction.
Measure, then change one thing at a time
Verify the primary conversion before spending. Review cost and qualified outcomes on a cadence that accounts for conversion delay, then adjust one major variable so the result remains attributable.
Choose the campaign objective before choosing the platform
A platform is a delivery system, not a strategy. Start with the business change you need, then select the audience, message, channel and measurement method that can produce evidence about that change.
| Business question | Useful campaign job | Evidence to collect | Common mistake |
|---|---|---|---|
| Do people already search for this solution? | Capture existing intent through tightly themed search campaigns. | Search terms, qualified clicks, conversion rate and accepted leads or orders. | Buying broad traffic before negative terms, location rules and landing-page relevance are controlled. |
| Can a visual demonstration create interest? | Test paid-social creative against defined audience situations. | Hook retention, landing-page visits, qualified actions and downstream economics. | Calling an ad successful because it receives cheap reactions that never reach the business outcome. |
| Is repeated reach across the open web valuable? | Assess programmatic display, video, audio or connected-TV inventory. | Viewable reach, frequency, placement quality, incremental response and brand-safety records. | Optimizing to inexpensive impressions without checking whether people could see them or whether the placement fits the brand. |
| Do previous visitors need another useful reason to return? | Build a remarketing sequence tied to the page or action already completed. | Audience eligibility, recency, frequency, assisted conversions and suppression accuracy. | Showing the same message indefinitely or continuing to target people who already converted. |
A sound objective is specific enough to reject a seductive but irrelevant result. “Increase awareness” is incomplete unless it names the audience, geography, time window and observable change. “Generate leads” is incomplete until a lead has an acceptance rule. A submitted form can be spam, a duplicate, an out-of-area enquiry or a person seeking a service you do not provide. Media platforms can optimize only toward the event they receive; the business must define whether that event represents value.
Use the objective to limit the test. One campaign does not need to discover the audience, invent the offer, compare every channel, redesign the landing page and prove long-term profit simultaneously. State the riskiest assumption, choose the smallest experiment that can challenge it and preserve the other conditions long enough to interpret the result. The campaign-planning guide turns that reasoning into a written brief.
Translate business economics into a break-even advertising limit
ROAS is useful only when the revenue in its numerator has economic meaning. A campaign can report revenue above spend and still lose money after product cost, fulfilment, returns, payment fees, sales labour or repeat-purchase assumptions.
Start with contribution, not gross revenue
Suppose an accepted order produces $120 in revenue. Product, payment and fulfilment costs consume $72, leaving $48 before advertising and fixed overhead. If the business is willing to spend no more than half of that contribution on acquisition, the planning CPA is $24. A campaign with a $20 CPA is inside that chosen limit; a campaign with a $32 CPA is outside it even though each order still reports six dollars of gross revenue for every advertising dollar.
The example is not a universal target. It shows the sequence: calculate contribution per accepted outcome, reserve the amount needed for overhead and profit, then derive the maximum acquisition cost. If repeat purchases matter, separate observed customer value from hoped-for lifetime value. Do not use an optimistic future-value estimate to excuse a current campaign unless retention data supports it.
Connect the limit to traffic assumptions
A $24 target CPA and a 3% click-to-conversion rate imply an approximate break-even CPC of $0.72: multiply the allowed CPA by the expected conversion rate. If traffic costs $1.20 per click, at least one assumption must change. The landing page may need a higher qualified conversion rate, the offer may need more contribution, the audience may need lower auction cost, or the business may need to accept a different acquisition limit.
Write the arithmetic before launch so the dashboard cannot redefine success after the spend occurs. Then replace assumptions with observed values. Recalculate using accepted orders or qualified leads rather than every tracked event, and show the range when refunds, delayed approvals or variable margins make one exact number misleading.
Worked decision: if CPC is $0.80 and the landing page converts 4% of qualified clicks, the observed media CPA is $20. That is acceptable against a $24 planning limit, but only if tracking does not double-count conversions and the business system confirms that the reported outcomes are genuine.
The detailed metrics and formulas guide separates CPM, CPC, conversion rate, CPA, ROAS, margin and customer value. Keep their units visible. A percentage, a currency amount and a revenue multiple answer different questions and should not be placed in one score without showing how the calculation works.
Design creative tests that can teach you something
Creative testing is not the production of endless variations. It is a controlled comparison of meaningful hypotheses about the audience, promise, proof, format or call to action.
Name the hypothesis
Write what you believe and why. For example: “First-time buyers hesitate because setup appears difficult; a short demonstration will produce more qualified visits than a product-only image.” This statement tells the team what must differ between variants and what result would challenge the belief.
Change one major idea
Keep the audience, offer, destination and measurement constant while comparing the demonstration with the product image. If the headline, format, audience and landing page all change, the winning ad does not reveal which decision mattered.
Judge the complete response path
An ad may win attention and lose qualified action. Review the sequence from impression to click, landing-page behaviour, conversion and accepted business outcome. A high CTR paired with poor conversion often signals message-to-page mismatch or curiosity that the offer cannot satisfy.
Record the learning, including a null result
If the variants perform similarly within the available evidence, record that the tested distinction did not produce a decision. Do not manufacture a winner from tiny differences. The next test should address a more meaningful uncertainty or collect enough observations to support action.
Platform delivery systems do not always split exposure evenly, especially when optimization predicts that one variant will perform better. That means a live ad set is not automatically a clean scientific experiment. For high-stakes conclusions, document delivery, audience overlap, spend, timing and conversion delay, and use a formal experiment feature when the platform provides one. For routine creative operations, the practical standard is simpler: preserve comparability, avoid simultaneous major changes and state the limits of the evidence.
Build one measurement chain from impression to accepted outcome
Advertising platforms, web analytics and business systems observe different parts of the same journey. Their numbers should be reconciled, not forced to match.
| Layer | What it observes well | What it cannot prove alone | Control to document |
|---|---|---|---|
| Ad platform | Delivery, auction cost, attributed clicks and platform-defined conversions. | Whether every attributed result became a valid order, lead or retained customer. | Attribution window, event used for bidding, audience exclusions and conversion-delay expectation. |
| Web or app analytics | Sessions, page paths, events, campaign parameters and cross-channel reporting under its attribution model. | Whether an offline sale was approved, refunded or economically valuable unless that data returns. | Event specification, source/medium naming, consent behaviour, identity limits and referral exclusions. |
| CRM, commerce or booking system | Accepted leads, paid orders, margin, status changes, refunds and fulfilment. | Every view or influence that occurred before the record was created. | Deduplication key, acceptance definition, revenue field, cancellation logic and update cadence. |
| Finance or operations record | Realized revenue, variable cost, capacity and cash consequences. | Which individual ad deserves all credit for a multi-touch journey. | Reporting period, contribution model and how delayed costs or returns are applied. |
Start with a release identity for measurement: the event name, trigger, parameters, destination, consent condition and expected count for one test action. Test it before the campaign launches. A thank-you page that reloads can fire twice; a form can record success before the server accepts the lead; a purchase event can send gross value when the economic analysis expects net value. Each difference changes the optimization signal.
Attribution is a rule for assigning credit, not a camera that records an objective single cause. Platform and analytics reports may use different lookback windows, identities and models. Compare them at a useful aggregation level, investigate material gaps and retain the business system as the authority for accepted outcomes. The measurement-plan guide includes an event contract, naming convention, QA matrix and reconciliation process.
Diagnose campaign problems in the order they occur
Start upstream. A later metric cannot be interpreted safely when the earlier stage that feeds it is broken.
| Observed symptom | Check first | Then inspect | Avoid this reflex |
|---|---|---|---|
| Little or no delivery | Eligibility, approval status, date, budget, bid strategy and audience size. | Auction competitiveness, search volume, placement restrictions and tracking-dependent optimization requirements. | Raising bids before confirming that the campaign is eligible and correctly scheduled. |
| Impressions but weak response | Message relevance, format, offer clarity and whether the ad reaches the intended situation. | Frequency, placement quality, search terms and the difference between viewable and served impressions. | Calling every low CTR a targeting failure; some channels and objectives are not click-first. |
| Clicks but few qualified actions | Message-to-page continuity, page function, speed, mobile usability and conversion-event integrity. | Audience intent, form friction, price/offer comprehension and downstream lead quality. | Optimizing the ad for more clicks when the destination or offer is the broken stage. |
| Tracked conversions but poor business results | Event definition, duplication and whether platform conversions map to accepted outcomes. | Lead routing, sales response, cancellations, margin, fraud and fulfilment capacity. | Using platform ROAS as a substitute for reconciled business economics. |
| Performance changes suddenly | Change log, tracking release, budget, creative, audience, landing page and site availability. | Seasonality, auction conditions, product availability and reporting delay. | Making several emergency edits before identifying whether the change is real or a data defect. |
Separate a diagnostic metric from a decision metric. CTR can help locate a creative or relevance issue, but CPA or contribution may determine whether the campaign continues. Conversion rate can reveal destination friction, but it does not show whether the resulting customers are profitable. Keep both levels: the diagnostic explains where to investigate; the decision metric connects the finding to the campaign objective.
Run a campaign on a documented learning cadence
A useful operating rhythm protects the campaign from both neglect and constant reaction. The cadence should reflect spend rate, conversion delay, business risk and how quickly the team can implement a reliable change.
Before launch
Approve the brief, destination, creative, audience, budget and measurement contract. Complete one end-to-end test conversion, verify its value and identifier in every required system, confirm exclusions, and record the campaign settings that matter. Assign an owner for delivery, measurement and downstream response.
During the first learning window
Check for operational defects: disapproval, zero delivery, broken URLs, unexpected search terms or placements, runaway spend, duplicate events and unhandled leads. Avoid declaring a commercial winner before enough outcomes have matured. Annotate every material intervention so the later report has a timeline.
At the decision review
Reconcile spend with accepted outcomes, compare the observed values with the assumptions, and choose one of four actions: stop because the premise failed; repair a diagnosed constraint; continue to collect the evidence already planned; or expand cautiously while preserving a control. Record why the action follows from the evidence.
Responsible optimization also includes privacy and user expectations. Collect only the data needed for the stated purpose, follow the consent and regional requirements that apply to the property, restrict access, document retention and avoid treating inferred audience membership as certainty about an individual. Measurement loss should be acknowledged in the analysis rather than “fixed” through unapproved collection. A campaign plan is stronger when it states what the team will not collect and which conclusions the available data cannot support.
The final campaign record should be reproducible by someone who did not operate it. Preserve the promoted creative, destination version, targeting, budget, conversion definition, attribution settings, key changes and business outcome extract. Without that release history, the next team can copy a result but cannot know which configuration produced it.
A campaign needs an operating plan, not only ads
Media buying becomes useful when every number can lead to a defined decision.
Write the campaign brief
The brief connects a business outcome to one primary conversion, an audience situation, a message hypothesis, a destination, a channel job, an approved budget and decision rules. It also records the assumptions behind expected CPC, conversion rate and CPA. When observed performance differs, the team can locate the failed assumption instead of changing every control at once.
Use the campaign planning framework →Specify the measurement contract
The measurement plan defines event triggers, identifiers, values, campaign parameters, attribution labels, consent behavior and QA tests. It distinguishes a platform-attributed conversion from an analytics event and from the accepted order or lead in the business system. That distinction prevents dashboards from assigning different meanings to the same word.
Build the measurement plan →Short answer: launch only after the destination completes the promised action, the primary event fires exactly once, targeting matches the brief, and a named owner knows which evidence will pause, continue or expand the test.
After launch, preserve the relationship between cause and result. Annotate material changes, account for conversion delay, reconcile platform reports with business outcomes and compare economics through CPA, ROAS, margin and customer value. A dashboard is an observation surface; the written plan determines what the observation means.
The operating plan also covers what happens after a response. Inventory, appointment capacity, lead routing, sales follow-up and customer support can all limit the value of additional traffic. Advertising cannot repair an unanswered phone, an unavailable product or a form that never reaches its owner. Verify the complete path from impression to fulfilled outcome before interpreting a low conversion rate as a media problem.
Frequently asked questions
What is digital advertising?
Digital advertising is paid promotion delivered through internet channels such as search engines, social platforms, websites, apps and streaming services. Advertisers bid in automated auctions to show ads to audiences they choose, and pay per impression, click or conversion.
How much does digital advertising cost?
There is no fixed price. Auction cost depends on competition, audience, placement, objective and ad quality. A test budget should state its maximum loss, expected cost range and the evidence required for a decision.
What is the difference between CPM, CPC and CPA?
They are ways of buying and measuring ads. CPM is the cost per thousand impressions (you pay for views), CPC is the cost per click (you pay when someone clicks), and CPA is the cost per action or acquisition (you pay when a defined conversion happens).
Which channel should a beginner start with?
Usually search or paid social. Search captures people already looking for what you offer, while paid social is better for creating demand with visual creative. The right choice depends on your goal, audience and budget.
How do I know if my ads are working?
Define one conversion that matters (a purchase, lead or signup), track it with the platform's tag or analytics, and judge the campaign on cost per conversion and return on ad spend rather than on clicks or impressions alone.
New to the vocabulary?
The glossary defines 40+ terms — from ad exchange to viewability — in one sentence each, with the formulas where they matter.