Programmatic Display Ads:
What to Buy and Why

A campaign report can say that a million display ads were served, the average price was low, and hundreds of people converted after seeing an ad. Those figures still leave a buyer with the decision that matters: which impressions were worth paying for, and how many of the reported conversions would have happened anyway? Programmatic display advertising makes that question harder because the same campaign can combine several publishers, sellers, data providers, fees, and ways of counting results.

The sensible way to buy it is to start with the people and outcome you need, choose a deal that gives you suitable placements at an acceptable total cost, and judge delivery separately from business impact. Automation is valuable because it lets a buyer make and adjust many placement decisions at scale. It does not make every available impression useful or turn an attributed sale into a sale caused by the ad.

Separate the display ad from the way it was bought#

Programmatic describes an automated way to buy and sell digital advertising. Display describes the kind of ad being placed, such as a banner or a richer visual unit on a site or app. A display ad can be bought through an automated platform or through a direct, manual agreement; programmatic buying can also be used for formats beyond display. The IAB and PwC revenue report explicitly treats programmatic as a way of selling rather than a separate format.

That distinction prevents a surprisingly common planning error. For the United States in 2025, the same IAB and PwC report records $81.6 billion in display revenue, including banners, rich media, sponsorship, and native advertising. It separately records $162.4 billion in programmatic revenue excluding search. Those figures describe overlapping dimensions, so adding them together would double count some activity. The programmatic figure is also broader than programmatic display: the report notes automated buying in channels such as connected television.

For a campaign brief, specify both dimensions. “Programmatic display” tells the team that the creative will occupy display placements and that buying will use automated transaction tools. It does not tell them which sites or apps, how the audience will be selected, whether access is open or negotiated, what fees sit above the inventory price, or how success will be measured. Those are the choices that determine whether a low quoted cost per thousand impressions, or CPM, is a good purchase.

The format line also matters when comparing reports. A graphic that happens to contain motion can still fall under a display definition, while a commercial in a video player belongs to digital video in the IAB and PwC survey definitions. A buyer should agree on the creative and placement definitions before comparing a display campaign with a video campaign or reading a supplier’s “programmatic” total as though it described one type of ad.

Follow one impression from the publisher to the ad#

Imagine a reader opening a publisher’s article with space for a display ad. The publisher’s systems make that opportunity available to buyers, often through a supply-side platform, or SSP, and an exchange. A demand-side platform, or DSP, receives information about the opportunity, checks whether a buyer’s campaign wants it, and may return a bid. The publisher’s ad server ultimately selects and renders an ad. The IAB UK’s programmatic guide describes this flow, including the roles of SSPs, exchanges, DSPs, and the publisher ad server.

The opportunity is an impression, a chance to serve an ad in a particular placement. It is not a promise that a person noticed the creative. A DSP can use campaign settings and available information about the page, placement, and audience to decide whether to bid, but the decision can only be as good as the information and rules available to it. That is why “automated” describes the transaction, not the quality of the decision.

In a real-time auction, the decision happens while the page or app loads. The publisher may offer an impression to more than one exchange or SSP, which can return bids through a publisher wrapper before the ad server makes the final selection. The IAB UK guide notes that this chain can add latency. Speed is therefore part of the buying and publishing trade-off: extra demand can create more bidding opportunities, while a heavy path can slow delivery for the person trying to use the page.

Real-time bidding, or RTB, is only one way a programmatic transaction can happen. The IAB’s explanation of programmatic deals also describes private auctions, preferred deals, and automated guaranteed transactions. A buyer who equates “programmatic” with an open auction may reject useful reserved inventory; a buyer who hears “automated” and assumes every impression is auctioned may misunderstand a fixed-price agreement.

The practical question is what the automation is being asked to select. It might be allowed to buy broadly across available display inventory, restricted to a set of sellers, or tied to a negotiated placement. Each version changes the buyer’s reach, certainty, and ability to choose the context around an ad. The transaction path is a means to that choice, not a campaign objective in itself.

Choose the deal by the access and certainty you need#

The four familiar deal routes differ on two points: who may compete for the inventory, and whether price or volume has been agreed in advance. The IAB and PwC definitions distinguish a private marketplace, where selected advertisers bid; a preferred deal, where price is fixed but volume is not; and automated guaranteed buying, where price and inventory are agreed. The IAB’s deal description sets those alongside an open auction.

Use an open auction when reach and flexibility matter more than a named placement, provided the buyer can set acceptable inventory and seller rules. Use a private marketplace when a specific publisher group or environment matters enough to justify narrower access. Neither route makes a vague audience definition or weak creative effective. The private route buys a different selection of opportunities; its value still depends on the actual placements, total price, and outcome.

A preferred deal suits a buyer who wants to know the unit price before an impression becomes available but can tolerate uncertain volume. Automated guaranteed buying fits a brief that needs a committed amount of inventory in a named environment, such as a planned launch where placement certainty is central. The cost of that certainty is reduced flexibility: if the message, audience, or page context proves wrong, a reserved commitment may be harder to redirect than auction spend. The contract and delivery terms matter more than the label, so check exactly which placements and volumes are included.

Do not choose a deal by assuming the most exclusive route is always best. If the goal is learning which publishers and messages work, a broad but controlled auction can be useful. If the campaign needs to appear beside a particular publisher’s content on a known date, paying for a committed deal may be rational. What reverses the call is the value of that placement certainty relative to the price and the ability to change course after seeing results.

Set the audience, placement, and creative before the bid#

The buying platform needs a clear job. A campaign intended to introduce a product to new buyers needs a different audience and message from one meant to bring back people who have already shown interest. The IAB UK guide describes first-party customer information, third-party audience data, contextual information, and creative that can change with audience or context. Those tools offer choices; they do not justify piling every available targeting signal onto one campaign.

Start with the outcome the advertiser can recognize. If the aim is qualified enquiries, decide what makes an enquiry qualified and where it will be recorded. If the aim is reach, decide which people should have had the opportunity to see the ad and what repetition is acceptable. Only then choose the audience rule. A first-party group built from the advertiser’s own customer activity may be useful for a relevant follow-up message. A contextual rule, based on the content around the ad, can be useful when the page topic itself is the reason to advertise there. A purchased audience segment needs a clear account of how it was built and whether its description is accurate enough for the decision being made.

That distinction matters because an audience label can sound more precise than the data behind it. The IAB UK guide explains that audience profiles can come from different sources and that more data can complicate the buyer’s task. Before paying for an extra segment, ask what new decision it enables. If the segment only gives a familiar name to the same broad inventory, its fee may be easier to measure than its value. If it lets the buyer deliver a genuinely different message to a well-defined group, the case is stronger.

Privacy and access restrictions also change which audience tactics are practical. The IAB and PwC report describes fragmented data and signal loss across systems. A campaign should therefore have a usable plan when a preferred identifier or measurement link is unavailable, instead of assuming every impression can be matched to a known individual. What data can be used, and on what terms, depends on the market, publisher, platform, and current rules; check those terms when the campaign is built.

Placement is equally deliberate. A buyer may want the reach of many sites, the context of a few relevant publishers, or a fixed position on one property. These are different purchases even if every creative file is a display ad. Where the ad appears also shapes whether its message makes sense. For a specialised product, a smaller set of relevant pages might be a better starting point than the largest available impression pool, even if the quoted CPM rises. That is a judgment about likely usefulness, not a claim that one deal type has a universal performance advantage.

Creative has to earn the opportunity the media purchase creates. A display unit may have only a brief moment to identify the product and give a reason to act. The IAB UK guide warns against treating programmatic as a way to buy cheaper space while neglecting the ad itself. If a campaign has several audiences, vary the message only when the distinction is meaningful: a prospect may need the basic offer, while someone familiar with it may need a specific reason to return. Changing a headline automatically is not the same as having a better idea.

Compare the full cost, then the useful impressions#

The auction or deal price is not necessarily the buyer’s full media cost. The IAB’s programmatic fee calculator announcement lists layers such as DSP technology, data and targeting, ad serving, campaign management, pre-bid and post-bid evaluation, and verification. How those charges are assessed differs by supplier and contract. Ask for the effective cost of the delivered campaign, with each layer named, rather than treating the inventory CPM as the complete price.

An illustrative calculation shows why. Suppose a campaign buys one million impressions at an inventory CPM of $8. Assume, only for this example, that DSP, data, verification, and ad-serving charges add $4 per thousand delivered impressions and that every charge uses the same denominator. Inventory costs $8,000; the added layers cost $4,000; the buyer spends $12,000, or a $12 effective CPM. The example is arithmetic, not a market rate. In a real proposal, a percentage fee, fixed charge, minimum, or different billing denominator could change the result, so the line items need to be reconciled to the invoice.

The cheaper route by inventory price can become dearer by useful exposure. Suppose, again only as an illustration, that two routes each deliver 100,000 measured impressions. Route A costs $1,200 in total and 50,000 impressions meet the chosen display viewability standard; route B costs $1,300 and 75,000 do. Route A’s cost per thousand viewable impressions is $24, while route B’s is about $17.33. Route B costs more in total but less per qualifying viewable impression. That still does not establish which route creates more sales: it only reveals a different denominator for the media purchase.

This is the budget discipline I would use. Compare inventory cost, technology and data charges, delivered impressions, viewable impressions, and the business outcome on the same campaign scope. If a fee pays for a capability that materially improves placements or decisions, it may be worth paying. If two paths reach the same placements and people with similar outcomes, the one with avoidable extra layers needs a strong explanation. The IAB fee calculator was designed to show how such layers contribute to effective CPM; it does not declare every fee wasteful.

Check the seller and the opportunity to see the ad#

A buyer should know who is authorised to sell a publisher’s inventory. In an ads.txt file, the publisher lists seller accounts that may sell its web inventory, including whether an account is direct or a reseller. Google’s Display & Video 360 guidance explains that the declaration is public and that its platform checks seller authorisation; it also describes the related app-ads.txt mechanism for apps. The practical reason to check is straightforward: an attractive domain name in a bid request is less reassuring if the selling path cannot be connected to the publisher’s declared sellers.

Seller authorisation answers only one question. It does not tell you whether the page is appropriate for the brand, whether a real person had a chance to see the ad, or whether the ad changed behaviour. Use the seller record as a supply check, then ask for placement information and for an explanation of how unsuitable or invalid inventory is handled. The IAB UK guide describes verification services that assess suspect inventory and invalid traffic. The useful conversation is about what the campaign actually excluded and reported, not whether a vendor has a reassuring product name.

Viewability gives another, narrower answer. Under the display standard cited in the IAB and MRC attention guidelines, at least 50% of the ad’s pixels must be in view for one continuous second. A qualifying impression indicates an opportunity to see the ad. It does not say that the person looked at it, understood it, remembered it, or bought because of it. The guidelines distinguish viewability from attention and distinguish attention from campaign outcomes.

That difference should shape reporting. Ask how many impressions were served, how many could be measured for viewability, and what share of those met the standard. Keep the denominator visible; a high viewability rate among a small measurable subset is a weaker description of the whole campaign than the same rate with broad coverage. If attention measures are added, ask what signals they use and what part of the inventory they cover. The IAB and MRC guidelines set out several attention methods and state that attention alone is not a business-result measure.

Read conversions as attribution before claiming lift#

After delivery, most buyers want to know whether the campaign brought in customers. The first report they see is often an attribution report: it assigns a conversion to an ad interaction using configured rules. In Display & Video 360’s conversion examples, a post-view conversion can be credited to an impression and a post-click conversion to a click within the example windows; clicks take precedence over impressions when both qualify. The help page uses 10-day post-view and 15-day post-click windows for its examples. Those are example settings, not a universal programmatic standard.

This is useful accounting, but it answers a limited question: which ad interaction receives credit under the chosen window and precedence rule? If a person was already likely to purchase, showing a display ad before the purchase can create a credited post-view conversion without creating an extra purchase. Changing the window can change the count even when the underlying customer behaviour stays the same. Read attributed conversions alongside the settings that produced them, and be wary of comparing two campaigns if their windows, conversion definitions, or eligible impressions differ.

To estimate what the advertising added, compare outcomes for people who had the opportunity to receive the campaign with outcomes for a comparable group held back from it. Google’s explanation of user-based Conversion Lift describes treatment and control groups and reports the difference in conversions as incremental conversions. If the groups are properly assigned and measured, their difference addresses a question attribution cannot: what changed because the ads ran? The feasibility and precision of a lift study depend on its design, scale, and available data, so a weak or inconclusive result needs to be reported as such.

The distinction is more than a technical preference. In 15 US advertising experiments at Facebook, researchers compared randomized results with several observational methods and found that the observational methods often failed to reproduce the experimental effects, despite extensive demographic and behavioural variables. Those experiments do not give a universal lift estimate for display advertising. They show why a large attribution dataset, by itself, cannot settle the causal question.

An illustrative result makes the decision clear. Suppose a campaign reports 600 attributed conversions and a properly designed holdout estimates 120 incremental conversions from the same campaign. Those figures are not contradictory: one counts conversions assigned to ads by rules, while the other estimates additional conversions. If the campaign costs $12,000, the cost per attributed conversion is $20; the cost per incremental conversion is $100. The numbers are invented for illustration, and a real study would also need an uncertainty range. The right buying decision depends on whether the advertiser can profitably acquire an additional customer at the incremental cost, not on which report has the more flattering denominator.

For a small campaign that cannot support a useful lift estimate, do not pretend a post-view total solves the problem. Use delivery, placement, cost, and attributed outcomes to make cautious changes, while stating that the additional business effect is unknown. If the spend becomes large enough for a holdout, agree on the outcome and study design before the campaign starts. Measurement choices made after seeing a result are less helpful for deciding what to buy next.

Run the next campaign as a sequence of decisions#

Begin with a brief that names the product, the audience, the action sought, and the reason a display placement is suited to that job. Pick the buying route that serves those needs: flexible auction access for discovery, a private pool for particular environments, a fixed-price preferred deal where rate certainty matters, or reserved inventory where a known placement and volume are essential. Then specify the acceptable sellers and placements, the creative each audience will see, and the full cost the buyer is prepared to pay.

Before launch, make the reporting definitions agree with the decision. A “cheap” CPM is only useful with its fee basis; a viewability rate needs its measured denominator; a conversion total needs its action definition and attribution windows. If the campaign’s purpose is incremental sales, reserve a way to estimate incrementality when scale permits. These choices are easier to make while the buyer can still change the brief than after the budget has been spent.

During delivery, look first for practical failures: spending on an unwanted seller, a placement that does not match the message, an unexpected fee, or creative that fails to communicate the offer. Change the condition that caused the failure rather than treating the DSP as a machine that will repair an unclear strategy. The automated path is worth using when it makes good choices repeatable across many impressions and gives the buyer enough information to improve the next purchase. The strongest programmatic display campaign is therefore the one whose placement, total cost, and measured outcome all make sense together.