Plan A · Buy a publication's audience
Negotiate a fixed price directly with a publisher for banner impressions on its health pages. Choose the publication and section in advance.
How does an ad get onto this page?
A reader opens an article. Several brands want the same ad space. Follow the sale, then ask whether buying ads this way helps the advertiser.
Programmatic means software automates buying and selling digital ads. Here we explore one type: an auction for an individual ad opportunity, called real-time bidding (RTB). Programmatic also includes negotiated, fixed-price deals.
Start with one reader and one space
Who gets the top banner, and what do they pay?
Our example: GlowRoute sells skincare. The Metro Chronicle publishes an article about dry skin. Maya opens that article. All three names are fictional; the bids are teaching inputs.
Step 1 of 5
CPM is dollars per 1,000 impressions. For one impression, divide by 1,000. Try a $4.50 bid.
1× keeps the top-banner minimum at $2.50 CPM. Raising it can leave the space unsold.
| Advertiser | Submitted bid | Decision |
|---|
A simplified cap for this example. The slots are processed top banner → article → sidebar. A brand that already won is excluded from later auctions. Other brands can still fill them.
One more comparison: ask buyers together or in order?
Why compare? A publisher may miss a higher offer if it accepts an earlier offer before asking everyone else. Both examples use the same top-banner bids and floor, with no previous winner.
Together: consider all five offers, then accept the highest eligible one. This illustrates the competition that parallel approaches such as header bidding can enable.
In order: ask GlowRoute, LuxeDerma, QuickGlow, EcoPure, then FinBank; stop at the first eligible offer. This is a simplified sequential example.
Real header bidding connects multiple selling platforms, which may each run an auction, before the publisher's final selection. This example omits those layers and timing. It does not estimate page speed or guarantee a revenue increase.
What did we simplify in this auction?
Every brand is represented by one comparable dollar bid. All ads are approved, arrive on time, and have equal fees and delivery probabilities. The highest eligible bid wins; the winner pays its submitted bid. Ties go to the first brand in the displayed list. No buyer sees rivals' bids before submitting.
Actual display systems can compare net expected publisher value, apply creative and policy filters, and involve several auctions. Advertiser relevance can influence how much a buyer bids. A universal “bid × Quality Score” rule does not govern these display auctions.
Buying method → price and audience → possible outcomes
Does automated buying give GlowRoute better value?
The auction decided who bought one placement. It did not tell us whether anyone clicked or bought skincare. To plan a campaign, we need assumptions about those later actions.
Plan B · Evaluate individual opportunities
Use buying software to bid for banner impressions that meet the campaign's criteria across eligible sites. Set an average purchase price for the plan.
What is comparable? Two online display plans with the same budget, banner format, campaign period, and purchase definition. This is not a TV-versus-digital comparison. Direct buying can also use automation or targeting. The buying method alone does not specify price or performance.
1. State the assumptions before reading the result
Start with identical assumptions. Then change one number. These are hypothetical planning values, not industry benchmarks or results from the auction above.
Media spend only. Creative production, service fees, taxes, refunds, and product margin are excluded from both plans.
| What we assume | Plan A · Direct | Plan B · Auction buying |
|---|---|---|
| Average price per 1,000 impressions ($ CPM) | ||
| Clicks per 100 served impressions (% CTR) | ||
| Purchases per 100 clicks (% post-click conversion) |
2. Follow the same calculation for both plans
| Calculation | Plan A · Direct | Plan B · Auction buying |
|---|
Fractions are expected averages across repeated campaigns, not partial people. We calculate with unrounded values. “Purchase” means a purchase credited after a click under the same tracking rule in both plans.
Reproduce these two plans in Python:
3. Decide what this example can tell us
A different result follows from a different price, click rate, or purchase rate. The calculator does not establish that the buying method caused those differences.
How could we test whether ads caused additional purchases? For example, randomly assign comparable eligible audience members to receive the campaign or to a holdout group that does not receive it. Compare purchase rates over the same period. Random assignment helps separate the campaign's effect from people's existing tendency to buy. A credible test also needs consistent delivery and measurement, enough observations, and an assessment of uncertainty.
What about the publisher and the reader?
Publisher: advertiser spending is not the publisher's net revenue. For an illustrative $5 CPM sale with a 20% selling fee, the publisher receives $4 per 1,000 billable impressions: $5 × (1 − 0.20). That fee is an assumption, not a market average. The publisher's other advertisers, unsold spaces, and costs are absent from this campaign model.
Reader: this model has no measure of satisfaction, attention, privacy comfort, or brand recall. More targeted ads could be useful, intrusive, or repetitive. We need reader evidence to judge the experience.
A successful auction can still buy an unhelpful ad
Where can a campaign go wrong?
Buying software follows the rules and signals it receives. Winning a cheap impression does not establish that a person saw it, that it reached someone new, or that a purchase followed.
Compare two setups. Change one assumption.
| What we compare | Baseline setup | Changed setup |
|---|
Why does the outcome change?
How would we check this in a real campaign?
Where do fraud and made-for-advertising sites fit?
Invalid traffic includes activity that does not represent legitimate ad engagement, such as automated traffic. Made-for-advertising (MFA) describes sites built around generating ad revenue, often with heavy ad loads or low-value content. MFA classification and fraud detection answer different questions; an MFA label does not itself prove bot traffic.
Check site-placement reports, traffic verification, ad density, attention or outcome measures, and the publisher's content. Low CPM or high reported viewability alone cannot establish that an impression was useful. No fixed percentage of campaign spend is automatically “wasted” on these sites.
References
The mechanisms below are documented by industry and platform sources. Every bid, response rate, fee, and exposure count in this simulator is an invented teaching input. These sources do not validate our numerical examples as forecasts.
- Glaser, B. (2014, April 23). Standardizing programmatic terminology with the IAB. Interactive Advertising Bureau. Supports: the distinction between auction-based and fixed-price programmatic transactions.
- Cox, S. (2019, March 6). Simplifying programmatic: first price auctions for Google Ad Manager. Google. Supports: the first-price payment rule and keeping buyers' bids private before bidding. This announcement describes a specific platform's transition, not every advertising auction.
- Google. (n.d.). Auction model. Google Ad Manager Help. Supports: comparing net bids and expected publisher value, and the distinction between buyer charges and publisher receipts.
- Prebid.org. (n.d.). Introduction to Prebid. Supports: gathering demand-partner bids and passing results to the publisher's ad server. The sequential-versus-parallel example omits the implementation's layers and timing.
- Google. (n.d.). Manage frequency with frequency caps. Display & Video 360 Help. Supports: frequency limits and the role of identifiers and modeling.
- Google. (n.d.). Active View reporting. Google Ad Manager Help. Supports: the display visibility threshold and viewable/measurable denominator.
- Zhou, T., et al. (2021). An efficient deep distribution network for bid shading in first-price auctions. Proceedings of the 27th ACM SIGKDD Conference on Knowledge Discovery & Data Mining. Open-access manuscript. Supports: estimating the tradeoff between bid price and probability of winning. Its empirical results apply to the authors' implementation and tests.
- IAB UK. (n.d.). Made for Advertising (MFA) website. Supports: the description of sites designed to maximize ad exposures.
- Google. (n.d.). About Quality Score for Search campaigns. Google Ads Help. Supports: treating the displayed keyword score as a diagnostic rather than an auction input, and distinguishing it from this display-auction example.
- Gordon, B. R., Zettelmeyer, F., Bhargava, N., & Chapsky, D. (2019). A comparison of approaches to advertising measurement: Evidence from big field experiments at Facebook. Marketing Science. Author's publication page. Supports: distinguishing observed purchase associations from effects measured in randomized advertising experiments. This research does not establish a universal advantage of auction buying over direct buying.
Sources checked October 20, 2026. No model is connected to a live campaign or ad exchange.