Reading programmatic advertising with a sceptical eye

What to take away

Programmatic advertising uses software and rules to transact digital media. It can include real-time bidding, private auctions, preferred arrangements, and automated guaranteed buying across display, video, audio, connected television, mobile, native, and other inventory. The label says little about the actual publisher, seller path, auction, audience, creative, price, measurement, or customer outcome.

A useful program starts with a business decision, then exposes the transaction in enough detail to govern it. The team should know what opportunity was eligible, which parties handled it, how a winner and price were determined, what was delivered, which fees were charged, what evidence survived, and who can stop or correct the system.

Define the assignment before the platform

Write a one-page decision brief and name the customer situation, market, offer, inventory boundary, creative claim.

Add primary outcome and maturity window. Also add budget, capacity, risk tolerance, learning question.

State why automated buying fits and which direct purchase, other channel, or nonmedia action competes for the same money, and set the paid media strategy governing how each channel earns its budget.

Map the commercial and technical chain

Draw the path from advertiser through agency, demand platform, exchange, reseller, supply platform, publisher, and measurement partners. Some parties may be absent, combined, or represented under another commercial entity. For each participant, record its role, legal name, account, identifier, fee basis, data access, subcontractors, logs, contractual status, and incident contact. Examples include Google Display & Video 360 and The Trade Desk as demand-side platforms (DSPs), Magnite as a supply-side platform (SSP), and Index Exchange as an exchange. Treat these as examples for identifying the roles of contracted parties, not as recommendations.

Then map the request path. Record the page or app, placement, format, device, seller identifiers, supply-chain nodes, auction and deal fields, bid, creative, win notice, price, delivery event, verification event, outcome event, and invoice line. The diagram should show unknown or aggregated fields rather than pretending they are available. Request a representative transaction record from each platform so you can match the documented path to what was actually bid on and delivered.

Understand what an auction does

OpenRTB, maintained by IAB Tech Lab, is a protocol used for bid requests and responses; it defines an exchange format, not proof that a particular buyer exposes every field. Ask whether each route is OpenRTB-based or uses a closed interface, and request the applicable version and field documentation. Use an auction transparency checklist to structure questions, then verify the actual rules in buyer and seller records.

Do not infer the price rule from a platform label. Ask how nominal bids become effective bids, whether quality or relevance affects rank, how floors and discounts work, which deal terms override the open auction, how multi-bid behavior is handled, and which loss reasons are reported. Preserve dated documentation because rules and fields change. For OpenRTB traffic, request sample fields such as imp.bidfloor, bid.price, and dealid, along with available win and loss notices. OpenRTB's lurl can carry a loss notice where supported, and its ${AUCTION_LOSS} macro can convey a loss reason; request the platform's codebook and confirm whether its reason codes are standardized or proprietary.

A first-price auction does not mean the advertiser's invoice equals the submitted bid. Currency conversion, demand-platform fees, supply fees, data, verification, creative serving, discounts, rebates, taxes, credits, and agency arrangements can alter the final economics. Define every denominator and calculation base.

Choose buying arrangements deliberately

A deal name is not evidence of quality or directness. Inspect the delivered supply: a private marketplace can still include several sellers, variable properties, undisclosed fees, or weak controls, and an open auction can contain valuable direct inventory.

Judge the actual record, not the sales category. That also applies to display advertising, where signal versus noise decides whether a deal name means anything.

Set supply eligibility and quality controls

Define allow, block, and manual-review rules for domains, apps, channels, content, placement, refresh, reselling, sourced traffic, device, geography, language, format, publisher ownership, and made-for-advertising indicators. Decide how missing identifiers are treated. Test whether settings apply before the bid, after delivery, or only in a report.

Seller declarations help investigation, but a published file does not prove that every request is truthful, the page is suitable, or the campaign is effective. Compare declarations with bidstream and delivery records. Sample actual pages and apps. Maintain evidence for exceptions, incidents, credits, and remediation. Check the publisher's ads.txt or app-ads.txt declarations and the relevant sellers.json file against the OpenRTB SupplyChain Object (schain) in bid requests where available. IAB Tech Lab publishes these specifications; treat them as evidence to compare, not proof that every transaction followed the declared path.

Control audiences and data

Translate customer eligibility into documented data rules. Record data origin, collection context, permitted use, consent or other applicable basis, fields, transformations, identity method, retention, geography, sensitivity, partner access, correction, deletion, suppression, and model version. Do not present probabilistic categories as verified facts about individuals. For California consumers, assess the California Consumer Privacy Act (CCPA), as amended by the California Privacy Rights Act (CPRA), and the role of the California Privacy Protection Agency (CPPA). For covered EU processing, assess the General Data Protection Regulation (GDPR) and the relevant supervisory authority.

Treat audience expansion and automated targeting as material changes. Define the seed, exclusions, eligible inventory, optimization event, minimum evidence, review cadence, customer risks, and stop condition. Preserve the original configuration and obtain approval before broadening the system.

Build creative for the transaction

Approve claims, disclosures, identity, rights, accessibility, formats, destinations, and combinations before delivery. Dynamic creative requires a controlled feed with stable product IDs, approved text and images, prices, inventory, dates, conditions, landing pages, validation, fallback, expiry, and an owner who can pause it.

Test files and tags in representative publisher contexts. Inspect crop, sound, motion, load, click area, disclosure, ad recognition, backup behavior, redirects, consent, accessibility, and destination accuracy. Automated assembly does not transfer responsibility for the resulting claim.

Measure technical states honestly

Publish counts beside rates. State the denominator, vendor, method, accreditation scope when relevant, device and format coverage, identity method, filtering, missingness, sampling, estimation, time zone, attribution window, outcome maturity, and restatement policy. Reconcile platforms, verification, owned events, and invoices without forcing them to match. Verification providers include Integral Ad Science and DoubleVerify; identify the exact product, measure, method, and coverage rather than treating a vendor name as a guarantee.

Judge economics at the business boundary

Calculate net contribution after qualification, cancellations, returns, fraud, discounts, fulfillment, service, sales labor, media, data, platform, verification, creative, agency, and internal operating cost. Separate new, existing, and ineligible customers. Compare the marginal dollar with the next best use of funds.

Attribution allocates credit under a model. Incrementality estimates the difference caused under a design. Neither automatically equals profit. Use a causal test when the decision requires causation, and prespecify assignment, exclusions, spillover, power, outcome maturity, guardrails, stopping, and analysis. The same distinction matters when comparing programmatic results with paid social advertising: platform-attributed conversions are not causal evidence without a suitable design. Retain assignment and exposure records, define contamination rules, and document the outcome window for each test.

Govern vendors, agencies, and access

The business should control core advertiser accounts, billing, domains, creative, audience sources, tags, contracts, measurement definitions, exports, and recovery methods. Grant least privilege. Review administrators, agencies, API credentials, linked partners, automations, change logs, alerts, and inactive users on a schedule.

Contracts should state agent or principal status, inventory and data ownership, fee basis, rebates, conflicts, subcontractors, reporting, records, audit, incident duties, credits, liability, intellectual property, retention, termination, and transition. Commercial terms need legal review for the applicable market. Operational owners must be able to execute them.

Run a controlled operating rhythm

Create thresholds that trigger investigation, pause, or escalation. Include unknown supply share, missing seller fields, invalid traffic, unsuitable placement, frequency, cost variance, event loss, outcome deterioration, complaint rate, and unpaid credits. A dashboard alert is not a response plan; name the person and deadline.

Maintain a programmatic advertising control file

Keep a dated inventory of accounts, contracts, partners, identifiers, settings, deals, supply rules, creatives, feeds, audiences, outcome definitions, reports, invoices, decisions, experiments, incidents, and source documentation. Assign an owner to verify current product and standards specifications and archive versioned copies before launch and after material changes.

A mature programmatic operation can reconstruct representative transactions and explain the limits of its evidence. It can show who participated, what was authorized, which rules applied, what arrived, how it was measured, what it cost, what customers did later, what changed, and who had authority to act.

Verify programmatic advertising before release

Before launch, request a representative OpenRTB bid request and response, plus a seller-path export. Where available, match publisher declarations in ads.txt or app-ads.txt and seller entries in sellers.json to the SupplyChain Object (schain) nodes. Record missing nodes as unknown rather than treating declaration files as proof of delivery.

Request the exact auction fields, price-rule documentation, and loss-reason codebook used by each contracted platform. Check whether the records show the bid floor, submitted bid, deal terms, and a loss notice such as OpenRTB's lurl and ${AUCTION_LOSS} macro where supported. Keep sample records and note which values are unavailable or platform-specific.

For measurement, name the verification vendor and product, such as Integral Ad Science or DoubleVerify, and document the measures, methodology, coverage, and limits relied on. For data use, record which privacy rules were assessed, including the CCPA as amended by the CPRA where applicable, or the GDPR for covered EU processing, and identify the responsible review owner.

Apply these checks to the actual programmatic advertising workflow. Record the tested data, roles, product versions, exceptions, approval date, and evidence that supports each decision. Repeat the review after a material source, model, access, contract, or decision change; a specification or vendor report does not certify the local implementation or guarantee a marketing result.

Common questions

Is programmatic advertising the same as real-time bidding?

No. Real-time bidding is one programmatic transaction method. Programmatic buying can also include automated private, preferred, and guaranteed arrangements.

Does programmatic guarantee lower prices?

No. Automation can improve workflow or access, but net cost depends on inventory, auction rules, fees, data, quality, creative, and outcomes.

Are private marketplaces always safer?

No. They can narrow participation or inventory, but the team still needs seller, placement, fee, quality, data, and outcome evidence.

What should a business audit first?

Start with ownership, contracts, supply paths, fees, administrator access, outcome definitions, and the ability to pause or exit.