Timeline.azMarketing, HR and business services
Marketing · 28.10.2026 · 5 min read

How Digital Advertising Works: Formats, Buying Models, Targeting and Measurement

Asiman Abdullayev

From the outside, digital advertising looks simple: you pay a platform and get views or clicks. In practice, four decisions determine whether the money works: the format, the buying model, the targeting and the measurement. Get one of them wrong and the other three cannot save the campaign.

Formats: how much attention are you buying?

Each format buys a different kind of attention:

  • Search ads: text and shopping ads shown against a query. High intent, limited reach.
  • Feed and stories ads: image, carousel and video ads in social feeds. Flexible and quick to test.
  • Short vertical video: TikTok, Reels, Shorts. Strong for reach and storytelling, if the first seconds work.
  • In-stream video: ads before or during videos, skippable or not. Good for demonstration and brand recall.
  • Display and native: banners and content-style ads on websites and apps. Cheap reach, low attention.
  • Audio: streaming and podcast ads, useful for frequency.
  • Sponsored products: ads inside marketplaces and retailer sites, next to the buy button.
  • Lead and messaging ads: instant forms, or click-to-WhatsApp and Messenger ads that start a conversation.

Choose the format by the job. Demonstration needs video, intent needs search, and a sale that happens in conversation needs messaging ads.

Buying models: what you pay for and who carries the risk

Ad buying models compared
ModelYou pay forRisk is carried byBest when
CPMEvery 1,000 impressionsAdvertiserYou need reach and frequency
CPCEach clickSharedYou need traffic to a site or store
CPVEach video view, as the platform defines itSharedYou need the video to be watched
CPA / CPL / CPIEach action, lead or app installMostly the sellerThe action is clearly defined and verifiable
Fixed feeA placement for a set periodAdvertiserContext matters: local media, sponsorships
Table and CSV
  • CPM (cost per mille): you pay per thousand impressions. CPM = cost ÷ impressions × 1,000. You carry the risk that nobody reacts; it is the standard model for reach.
  • CPC (cost per click): you pay only when someone clicks. CPC = cost ÷ clicks. Good for traffic, but a click is not a customer.
  • CPV (cost per view): common for video. What counts as a "view" differs by platform, so read the definition before you compare.
  • CPA, CPL, CPI: you pay per action, lead or app install. Your risk is lowest, but the partner optimizes for whatever counts as an action. If a "lead" is just a phone number, you may get many cheap and useless leads.
  • Fixed fee or sponsorship: common with local news sites, influencers and placements sold by the week or month.

These models are connected: CPC = CPM ÷ (1,000 × CTR). For example, a CPM of 5 and a CTR of 1% give a CPC of 0.50. That is why a "cheap" CPM can still be expensive if nobody clicks, and a "high" CPM can be fine if the audience responds.

One more point. On large platforms you are almost always in an auction, and the platform's system bids toward the objective you choose, even when you are billed per impression. The campaign objective you select matters more than the billing unit. Pick the objective closest to your real business result, not the one that produces the prettiest metric.

Targeting: who sees the ad

  • Contextual: keywords, topics, the content of the page or video.
  • Audience: age, location, interests and behaviour, as modelled by the platform.
  • Intent: people who are researching or "in market" for a category.
  • First-party: your own customer lists, website visitors (retargeting) and lookalike audiences built from them.
  • Geo and time: city, radius, day of the week and hour.

Two shifts matter today. First, privacy changes, such as Apple's App Tracking Transparency and browser limits on third-party cookies, have made detailed third-party tracking less reliable. Platforms have answered with automation and broad targeting, which means your creative now does much of the targeting: the message itself selects the audience. Second, first-party data collected with consent has become more valuable than any purchased segment.

In small markets, be careful with narrow targeting. A small audience saturates quickly, frequency rises and costs follow. With multilingual audiences, run separate ads per language rather than one mixed version.

Measurement: did it work?

Look at four layers:

  1. Delivery: impressions, reach, frequency, viewability. Did the ad reach real people, on screen, the right number of times?
  2. Response: CTR, video view rate, engagement. Did anyone pay attention?
  3. Outcome: conversions, CPA, ROAS. Did it produce business?
  4. Incrementality: would those sales have happened anyway?

The fourth layer is the one most reports skip. Every platform attributes conversions to itself, so if you add up the conversions each platform reports, the total is often higher than your real sales. To get closer to the truth, use holdout tests (a group that does not see the ads), geo tests (advertise in some cities and not in others) or the lift studies that some platforms offer.

Technically, measurement starts with a clean setup: platform pixels plus server-side connections such as conversion APIs, consistent UTM tags on every link, and consent handling that meets local rules.

Checklist before launch

  1. The campaign objective matches the business result.
  2. Conversion tracking is tested with a real test purchase or lead.
  3. Every link carries UTM tags that follow an agreed naming convention.
  4. A frequency cap or reach plan is set.
  5. Creatives are made for each format: vertical, subtitled, brand in the first seconds.
  6. Placement exclusions and brand safety settings are switched on.
  7. The report template and review dates are agreed before money is spent.

The takeaway

The next time someone shows you a cheap CPM or CPC, ask one question: "What did it cost us per customer, and would we have won that customer anyway?" Formats, buying models and targeting are tools. That question is the measure.

Related articles