Demand Side (Left)
Advertisers and agencies set goals and budgets; DSPs choose impressions and bid prices.
Brand / Agency
Trading Desk
DV360
The Trade Desk
Amazon DSP
Adtech / Programmatic Ecosystem — PM Visual Guide
Six diagrams plus a PM components playbook on auction quality, identity durability, supply trust, and attribution integrity — so this reads like an operating manual, not just a glossary.
Diagram 1
Programmatic Stack — Demand to Supply Relay
Horizontal topology: demand on the left, exchange intelligence in the middle, and supply on the right with directional flow.
Exchange Layer (Center)
Marketplace auctions match bid demand with inventory supply and pick winners in milliseconds.
Google AdX
OpenX
Index Exchange
Data & Identity Signals
Audience and contextual signals improve bid quality and targeting relevance.
DMP / CDP
Clean Rooms
Contextual Models
Supply Side (Right)
SSPs package publisher inventory, run yield optimization, and return winning creatives for rendering.
Google Ad Manager
Magnite
PubMatic
Publisher Site/App
Ad Slot + User
Verification & Measurement (Cross-Cutting)
Post-render systems verify viewability, brand safety, and invalid traffic across the whole chain.
IAS
DoubleVerify
MOAT
Nielsen / Comscore
PM lens: the middle layer (exchange + data + verification) is where optimization leverage and margin extraction both concentrate.
Diagram 2
Real-Time Bidding (RTB) Sequence
One impression clears in about 100–150ms: request fan-out, bidding, auction, render, then async verification.
User
Publisher
SSP
Exchange
DSP
Advertiser
0-15ms · Request fan-out
Loads page (0ms)
Ad tag fires, bid request (5ms)
Forwards to exchanges (10ms)
Distributes to DSPs + user signals (15ms)
20-60ms · Bid computation + auction
Collects bid responses, runs auction (55ms)
Evaluates user, computes bid, responds (20-50ms)
Campaign constraints + budgets pre-configured
60-100ms · Render + post-render verification
Sees ad by ~100ms
Renders winning creative (70ms)
Returns winner + creative URL (60ms)
Conversion may happen later (hours/days)
| Step | Actor | Action | Time |
|---|---|---|---|
| 1 | User | Loads page | 0ms |
| 2 | Publisher | Ad tag sends bid request to SSP | 5ms |
| 3 | SSP | Forwards bid request to exchanges | 10ms |
| 4 | Exchange | Sends requests to DSPs with user signals | 15ms |
| 5 | DSP | Scores user and computes bid | 20-50ms |
| 6 | Exchange | Runs auction, selects winner | 55ms |
| 7 | SSP/Publisher | Winner returned and creative rendered | 60-70ms |
| 8 | User | Ad visible | ~100ms |
| 9 | Verification | Viewability + brand safety checks | 200ms+ |
Hero constraint: each participant steals from the same latency budget. If your step is slow, someone else’s optimization work gets canceled by user abandonment.
Diagram 3
Cookie Death & Identity Alternatives
As third-party cookies fade, targeting shifts to first-party data, identity graphs, and contextual signals.
Third-Party Cookie (Legacy)
How: Browser cookie tracks behavior across sites for profile-based targeting.
Pros: Universal, simple, cheap.
Cons: Privacy backlash, browser blocks, unstable future.
Status: Dying.
First-Party Data
How: Logged-in publishers collect consented user data and activate through controlled environments.
Pros: High quality, durable, consent-based.
Cons: Hard for small publishers without authentication scale.
Winners: Amazon, NYT, large logged-in platforms.
Identity Solutions (UID2, RampID)
How: Hashed email identity passed across ecosystem as cookie alternative.
Pros: Cross-site addressability with consent pathways.
Cons: Fragmented standards and uneven adoption.
Players: TTD (UID2), LiveRamp, Google PAIR.
Contextual Targeting
How: Target by page/topic semantics instead of user-level tracking.
Pros: Privacy-safe, broad scale, no personal identifiers required.
Cons: Lower precision for niche intent targeting.
Revival: GumGum, Peer39, NLP-based contextual providers.
2020
Safari and Firefox enforce strict third-party cookie blocking.
2024
Chrome deprecation delayed again as ecosystem readiness lags.
2025
Further delay extends transition uncertainty for buyers and publishers.
2026?
Inevitable shift to hybrid identity + contextual stack regardless of exact sunset date.
Diagram 4
Revenue Flow — Who Takes What
For a $10 CPM buy, intermediaries often take half before the publisher sees revenue.
Agency Fee
$1.50 (15%)
DSP Fee
$1.00 (10%)
Exchange/SSP Fee
$1.50 (15%)
Verification Fee
$0.30 (3%)
Data Fee
$0.70 (7%)
Publisher Receives
$5.00 (50%)
Ad tech tax: in open auction flows, publishers often receive only ~50% of advertiser spend. The other 50% is fragmented across intermediaries and service layers.
Diagram 5
Advertiser vs Publisher Perspective
The same impression transaction looks efficient to one side and margin-destructive to the other.
Advertiser View
- Goal: Reach the right user at the lowest cost.
- Sees: Impressions, clicks, conversions, ROAS dashboard.
- Cares about: Targeting precision, viewability, brand safety, fraud filters.
- Pain points: "Am I reaching real humans?" "Where did spend leak?"
- North-star metric: ROAS = Revenue / Ad Spend.
Same $10 CPM transaction
two opposite value perceptions
two opposite value perceptions
Publisher View
- Goal: Maximize yield without hurting UX.
- Sees: Fill rate, CPM/eCPM, revenue per session, latency impact.
- Cares about: Yield optimization, ad quality, page speed, user trust.
- Pain points: "Why is CPM down?" "How much do intermediaries take?"
- North-star metric: eCPM = Revenue / Impressions × 1000.
Asymmetry: advertisers buy outcomes; publishers sell attention. Programmatic intermediaries arbitrate between those objectives and capture spread in the middle.
Diagram 6
Measurement & Attribution Models
Attribution choice changes budget allocation behavior as much as targeting model choice does.
Last-Click Attribution
How: Final clicked ad gets 100% conversion credit.
Pros: Simple and deterministic.
Cons: Ignores awareness assist touches.
Used by: Smaller advertisers, default channel reports.
Multi-Touch Attribution (MTA)
How: Credit is distributed across touchpoints (linear/time-decay/position).
Pros: Better journey visibility than single-touch models.
Cons: Data heavy, privacy-constrained, modeling fragile.
Used by: Sophisticated brands and agencies.
Media Mix Modeling (MMM)
How: Statistical model estimates channel contribution at aggregate spend level.
Pros: Privacy-safe and includes offline media.
Cons: Slow cadence; weak for intraday optimization.
Used by: Large enterprise advertisers.
Incrementality Testing
How: Holdout A/B design compares exposed vs control conversion outcomes.
Pros: Causal signal, not just correlation.
Cons: Operationally expensive and slower to run.
Used by: Advanced teams (often with platform-native experiments).
Practical stack: many mature teams run MMM for strategic budget setting, MTA for tactical optimization, and incrementality tests to validate both.
Reference
Adtech Components — PM Lens
Where product choices in adtech move revenue, margin, and trust: auction quality, identity strategy, and measurement credibility.
How this connects to the diagrams: Programmatic Stack defines control points, RTB flow shows real-time decisions, Revenue Flow quantifies take-rate impact, and Measurement reveals attribution risk. This section converts the map into a PM action model.
Auction & Bid Strategy
What it does: Decides which impressions to bid on and how aggressively to price each one.
PM metrics: Win rate, CPM efficiency, cost per outcome (CPA/CPI), spend pacing vs budget.
Pitfall: Optimizing only for cheap inventory often degrades downstream conversion quality.
Supply Quality & Curation
What it does: Filters inventory for fraud, viewability, and contextual fit before buyers pay for it.
PM metrics: Viewability %, IVT %, brand safety incidents, effective CPM lift for curated paths.
Pitfall: Overly broad supply creates apparent scale but weak business outcomes.
Measurement & Attribution
What it does: Connects impression exposure to conversion outcomes with model assumptions.
PM metrics: Attributed ROAS, incremental lift, modeled-vs-observed delta, reporting lag.
Pitfall: Last-click comfort can hide incrementality collapse.
Identity & Privacy Stack
What it does: Maintains targeting and frequency control as deterministic IDs become scarcer.
PM metrics: Match rate, reachable audience %, frequency control error, consented signal coverage.
Pitfall: Treating identity as a vendor checkbox instead of a product capability.
Interview shortcut: frame adtech tradeoffs as efficiency vs control vs trust. Buyers chase outcomes, sellers chase yield, and regulators/users demand privacy. Great PM narratives show how your product balances all three instead of maximizing only one.
Common PM Failure Patterns
Cheap CPM trap
Teams optimize media cost while conversion quality and incrementality collapse.
Supply quantity over quality
Inventory expansion without fraud/viewability controls inflates spend but weakens outcomes.
Attribution illusion
Last-touch reporting claims wins that were already likely; budget gets misallocated.
Identity vendor dependency
No first-party strategy means targeting performance drops with every privacy policy shift.
Decision Matrix — What to Optimize For
| Situation | Optimize For | Guardrail Metrics | Avoid |
|---|---|---|---|
| New campaign ramp | Learning speed + signal quality | Win rate, CVR trend, pacing error | Over-constraining bid strategy too early |
| Brand safety incidents rise | Inventory trust + controls | IVT %, unsafe placement rate | Blindly widening exchanges to recover scale |
| ROAS pressure from finance | Incremental efficiency | Lift tests, iROAS, CPA by cohort | Only tuning last-click rules |
| Cookie/signal degradation | Identity resilience | Match rate, addressable reach, frequency error | Treating privacy changes as one-off incidents |
Interview Scenarios + Strong Answer Angles
“ROAS is down 20% after privacy changes. What now?”
Angle: segment by channel/signal class, shift budget to high-confidence inventory, re-calibrate attribution with holdouts.
“Publisher says your SSP fees are too high.”
Angle: show net yield impact (not fee % alone), add transparency controls, and test curated paths that improve effective CPM.
“Marketing wants scale; trust team wants tighter filters.”
Angle: propose tiered inventory quality lanes with explicit spend caps and quality SLAs.
Adtech / Programmatic Ecosystem Visual Guide · PM Prep Papers Series · fullstackpm.tech · March 2026