Introduction: Why RTB Win Rate and CPA Pull in Opposite Directions (and How to Improve Both in 14 Days)
If your RTB campaign is winning auctions but missing CPA, you’re probably buying impressions efficiently—just not the right ones. If you’re hitting CPA but can’t spend, you’re likely priced out, over-filtered, or limiting delivery with controls like caps and pacing. In both cases, the usual culprit is bid strategy + signal quality + supply path—not your banner.
Quick definitions (these get blurred in dashboards):
- RTB (Real-Time Bidding): Each impression triggers a bid request → your DSP decides whether/what to bid → the exchange runs an auction → you win/lose → the ad renders → conversions may happen later.
- Bid rate: bids submitted ÷ bid requests received. Low bid rate often means you’re filtering yourself out.
- Win rate: wins ÷ bids submitted. Low win rate usually means you’re not competitive or you’re entering tough auctions.
- CPA: cost per acquisition (your primary efficiency metric).
In the next 14 days, you’ll:
- Stabilize delivery (less “spend at 9am, stall by noon”)
- Improve win rate without letting CPA spike
- Reduce CPA without collapsing scale
- Build a repeatable optimization loop you can run weekly
Guardrails for the sprint
- Change one major lever every 24 hours (unless delivery collapses).
- Treat results as directional. Inventory quality, vertical, geo, and attribution rules can swing outcomes.
- Plan for a 2026 measurement reality: fewer identifiers, more modeling, more conversion lag. First job: don’t “optimize the tracker.”
Problem Statement: Common RTB Failure Patterns (Win Rate vs CPA)
Most RTB problems fall into three patterns. Identify the pattern first, then pull the smallest lever that fixes it.
Pattern A: High win rate, poor CPA (you’re buying the wrong impressions efficiently)
What it looks like:
- Win rate is strong, CPM is low-to-moderate
- Spend is stable
- CTR may look fine
- CVR is weak and CPA is high
- You may see low viewability and/or suspicious traffic quality
Likely causes:
- Low-quality supply (spammy apps, MFA-like pages, “clicky” placements with no intent)
- Weak contextual fit (ads show in places that don’t convert)
- Excess frequency on the same users
- Attribution/measurement issues (e.g., inflated view-through, double-firing events)
Pattern B: Low win rate, good CPA but no scale (you’re priced out or over-restricting)
What it looks like:
- Lots of bid requests available
- Bid rate may be low (you’re not bidding on much)
- Win rate is low (you bid but don’t win)
- CPA looks good on small volume, but spend is inconsistent
Likely causes:
- Bid caps too low / bids too conservative
- Targeting too narrow (too many constraints stacked together)
- Supply path too restrictive (SPO cuts too deep too fast)
- Frequency caps choking delivery early in the day
Pattern C: Volatile delivery (pacing/frequency/forecast mismatch)
What it looks like:
- Some days spend, other days nothing
- Big intraday swings
- CPA swings because the system never settles
- Changes don’t “stick”
Likely causes:
- Pacing fighting frequency caps (front-load then stall)
- Too many line items with tiny budgets (fragmentation)
- Too many changes at once (learning resets)
- Conversion lag (24–72 hours+) leading to premature decisions
Why teams get stuck
- Noisy conversion signals (dedupe issues, delayed conversions, modeled conversions)
- Too much cheap inventory (easy to win, hard to convert)
- Too many simultaneous changes (no way to attribute what worked)
Step-by-Step Solution
The 14-Day RTB Optimization Sprint (Day-by-Day)
This is the executable plan. Each day ends with a checklist so you can run it like an ops playbook.
Internal linking opportunities (Whalefeed placeholders):
Day 1–2: Baseline and measurement sanity check (so changes are attributable)
What to do
- Lock a baseline window (typically last 7–14 days, or shorter if spend is volatile).
- Verify conversion tracking and deduplication.
- Quantify conversion lag (median and 75th percentile).
- Create a change log (date/time, lever, expected impact, rollback plan).
Why it matters
If the pixel double-fires—or your conversion lag is 48 hours and you judge changes same-day—you’ll optimize in the wrong direction and call it “improvement.”
How to do it (practical)
- In your analytics/CRM: confirm the primary conversion is unambiguous (e.g., purchase, qualified lead) and separate from micro-events (add-to-cart, page view).
- In the DSP: confirm you’re optimizing to the correct event, and document view-through attribution settings and windows.
- Pull a lag view: conversions by days since click (or impression if you rely on view-through). If you can’t export this, approximate by comparing DSP conversions vs backend outcomes across days.
Example
A subscription advertiser excludes an app bundle and sees CPA “spike.” Two days later, CPA “recovers.” The real issue: trials convert same-day, paid subscriptions convert 48–72 hours later. They judged too early.
Day 1–2 checklist
- Baseline snapshot saved (date range, spend, conversions, CPA, win rate, bid rate)
- Pixel/event fires once (no duplicates)
- Primary conversion defined (not mixed with micro-events)
- Attribution windows documented (click + view)
- Conversion lag measured (median + 75th percentile)
- Change log created and used going forward
Day 3–4: Supply path audit (SPO) and inventory hygiene
What to do
- Rank sites/apps/placements by spend, conversions, CPA.
- Add quality columns where available: IVT (invalid traffic), viewability, plus simple engagement proxies (bounce rate/time on site if you can connect).
- Identify “high spend, zero conversion” pockets and low-quality pockets.
- Prefer more transparent paths (direct/authorized sellers) where possible.
Why it matters
Win rate can look great when you’re winning cheap, low-quality auctions. SPO is how you stop paying for “easy wins” that don’t produce outcomes.
How to do it (practical)
- Export a placement report (domain/app, app bundle, placement ID if available, exchange/SSP, deal ID).
- Create three buckets:
- Keep / scale: acceptable CPA and quality
- Watch: low volume or unclear signal
- Exclude / restrict: high spend with no results, suspicious quality, obvious mismatch
Using ads.txt / app-ads.txt / sellers.json
These IAB Tech Lab standards help verify authorized selling paths:
- ads.txt / app-ads.txt: publisher lists authorized sellers
- sellers.json: seller identity and relationships
Practical use: when a source performs poorly and the selling path looks indirect or unclear, deprioritize it—or test a more direct path.
References:
Day 3–4 checklist
- Top 50 sites/apps by spend reviewed
- “High spend, zero conversion” list created
- IVT/viewability flags applied where available
- Exclusion list v1 implemented (small, high-confidence cuts first)
- SPO test plan defined (avoid deep cuts in a single move)
Day 5–6: Tighten or rebuild targeting signals (context, device, geo, audiences, placements)
What to do
- Reduce stacked targeting that kills bid rate.
- Rebuild around signals that still hold up in 2026: contextual, geo, device/OS, publisher-provided signals, and first-party audiences (if available).
- Remove segments that look good on paper but don’t lift conversions.
Why it matters
If bid rate is low, you may not be competing. If you’re too broad, you may be inviting low-quality supply. The goal is eligible enough to scale but filtered enough to protect CPA.
How to do it (practical)
- Start from outcomes: pull conversions/CPA by geo, device/OS, time-of-day, environment (in-app vs web), and context category (if supported).
- Use a simple rule:
- Volume + bad CPA: restrict it.
- Good CPA + low volume: expand adjacent segments cautiously (not all at once).
Example
A DTC brand runs “all devices, all apps, all geos” and gets high win rate plus high CPA. Breakdown shows:
- iOS in-app: high spend, low CVR
- Android in-app: moderate spend, decent CVR
- Desktop web: low spend, strong CVR
They rebuild:
- Separate line items by environment (in-app vs web)
- Add a desktop-only contextual segment for shopping/comparison content
- Exclude worst app bundles from the Day 3–4 audit
Day 5–6 checklist
- Bid rate checked (if low, reduce stacked filters)
- Targeting rebuilt around 2–4 strong signals (not a long list of weak ones)
- Worst geo/device/environment segments restricted
- Line items simplified (avoid fragmentation)
Day 7: Bid ceilings and bid shading strategy (protect CPA without killing win rate)
What to do
- Set bid caps/ceilings to prevent runaway CPMs during learning.
- Adjust bids in small increments (e.g., +5–15%), not step-changes.
- If your DSP offers bid shading, test carefully and track win rate plus clearing CPM.
Why it matters
Win rate is partly a price problem. CPA is partly a cost-control problem. Bid boundaries let you compete without letting the system overpay when signals are noisy.
How to do it (practical)
- Decide what you’re solving:
- Priced out: raise bids or broaden targeting (pick one).
- Buying junk: tighten supply/signals first, then consider bids.
- Use caps to keep learning stable:
- If CPMs jump after a bid change, caps prevent a week of overspend while you diagnose.
Bid floors (what you can and can’t do)
- A floor is a minimum price set by the seller/exchange. Buyers generally can’t change open auction floors.
- You can respond by shifting supply paths, using deals with known pricing, or avoiding inventory where floors force inefficient CPMs.
Day 7 checklist
- Bid cap/ceiling set (aligned to CPA tolerance)
- Bid changes made in small increments
- Only one of: bids ↑ or targeting broadened (not both)
- If bid shading tested, impact tracked on win rate + CPM + CPA (confirm behavior in DSP docs)
Day 8: Pacing setup (daily + intraday) to stabilize learning and spend
What to do
- Match pacing decisions to your measurement reality:
- If conversion lag is meaningful, avoid same-day calls on incomplete data.
- Reduce line-item fragmentation so each line item has enough budget to learn.
Why it matters
A great CPA on $20/day isn’t a reliable system. Pacing is what turns a “good pocket” into steady delivery.
How to do it (practical)
- Use fewer, larger line items during the sprint:
- Prospecting (broad but controlled)
- Retargeting (separate budgets/bids)
- Watch for pacing + frequency collisions:
- Front-load then stall often means frequency caps or small audiences are binding.
Day 8 checklist
- Pacing mode chosen intentionally (daily vs lifetime, even vs ASAP)
- Line items consolidated (avoid fragmented learning)
- Intraday spend curve reviewed (front-load vs smooth)
- Alerting set for under-delivery and CPA spikes
Day 9: Frequency and recency controls (avoid overbuying the same users)
What to do
- Set frequency caps by funnel stage.
- Add recency rules (time since last visit/exposure), especially for retargeting.
Why it matters
Frequency is a quiet CPA killer. You can keep winning auctions against the same person and wonder why CPA rises while win rate stays healthy.
How to do it (starting points—adjust to your context)
- Prospecting
- Lower frequency (protect reach)
- Broader recency (you’re not chasing immediate action)
- Retargeting
- Moderate frequency can work, but add recency controls to avoid hammering users minutes after a visit
- Split by recency windows if you have volume (e.g., 1–3 days vs 4–14 days)
If delivery collapses after caps:
- widen audience or bids first,
- then relax frequency—not the other way around.
Day 9 checklist
- Frequency caps set separately for prospecting vs retargeting
- Recency rule applied for retargeting
- Frequency and reach monitored daily (diminishing returns check)
Day 10: Creative is a second-order lever—validate quickly, don’t thrash
What to do
- Run a minimal creative validation:
- 1–2 new variations max
- Same landing page
- Don’t rotate lots of creatives while also changing bids/supply
Why it matters
Creative matters, but it won’t fix broken supply or weak signals. Also, constant creative churn muddies attribution and can disrupt learning.
How to do it (practical)
- Use creative to test message-market fit, not to “rescue” low-quality inventory.
- Evaluate on CVR and post-click quality, not CTR alone.
Day 10 checklist
- Creative test limited (max 1–2 new variants)
- No other major lever changed the same day
- Decision criteria defined (CVR/CPA with lag accounted for)
Day 11: Retargeting vs prospecting budget split and bid separation
What to do
- Separate prospecting and retargeting:
- different budgets
- different bids/caps
- different frequency rules
- Prevent one from crowding out the other.
Why it matters
Retargeting often looks efficient and can absorb budget—until it saturates a small pool and growth stalls.
How to do it (practical)
- Prospecting goal: stable qualified acquisition volume at target CPA (or slightly above while learning)
- Retargeting goal: efficient conversions with tight frequency/recency and controlled spend
Fast cannibalization fix:
- Cap retargeting budget and/or tighten recency windows.
- Ensure prospecting has enough daily budget to compete and learn.
Day 11 checklist
- Prospecting and retargeting split into separate line items
- Separate bid caps and frequency/recency rules applied
- Retargeting budget capped to prevent crowd-out
Day 12: Conversion lag and attribution tuning (delayed signals without misleading the DSP)
What to do
- Align evaluation windows to your lag distribution.
- Adjust attribution settings if they’re clearly over- or under-crediting (carefully).
- If your DSP reports modeled conversions, confirm how they’re calculated and separated in reporting (DSP-specific—check docs).
Why it matters
If you cut sources before lagged conversions arrive, you bias toward “fast but low-quality” pockets and starve inventory that converts later.
How to do it (practical)
- Define a hold period for major changes: often 48–72 hours is a reasonable starting point if your data shows lag is material. Use your 75th percentile lag as a guide.
- Compare performance pre vs post while holding other levers constant.
Day 12 checklist
- Hold period defined based on lag (not guesswork)
- Attribution windows documented and reviewed
- Decision-making uses lag-aware comparisons
Day 13: Consolidate winners and cut losers (rules for pruning)
What to do
- Double down on the best combinations of:
- supply path (exchange/deal)
- context
- geo/device
- Prune losers with clear rules.
Why it matters
Accounts rarely fail because of one giant mistake. They fail from accumulated inefficiency. Consolidation improves learning, transparency, and control.
How to do it (practical pruning rules)
- Cut when:
- High spend + zero conversions beyond your lag window
- Consistently poor CPA with no quality upside signals
- High IVT / extremely low viewability (where measurable)
- Be cautious with low-volume segments: don’t “optimize” based on tiny samples.
Day 13 checklist
- Winners consolidated (budget shifted, not just noted)
- Losers cut using documented rules
- No mass changes without a rollback plan
Day 14: Lock the new steady state and set a weekly operating cadence
What to do
- Freeze the sprint setup as your new baseline.
- Create a weekly routine: SPO review, signal pruning, bid boundary checks, pacing/frequency review.
Why it matters
The sprint is a reset. The weekly cadence is how improvements compound instead of drifting.
Day 14 checklist
- New baseline snapshot saved
- Weekly cadence scheduled (who does what, when)
- Alerts/guardrails configured
- Change log kept as a permanent habit
Step-by-Step Solution Details (What to Do, Why It Matters, How to Do It)
This section explains the mechanics behind the sprint levers.
Step 1: Establish a baseline dashboard (bid rate, win rate, CPM, CVR, CPA, viewability, IVT)
What to do
Build one daily table with a primary metric and a few guardrails.
Why it matters
You need to know whether you’re fixing an eligibility problem (bid rate), a price/competition problem (win rate/CPM), or a quality problem (CVR/IVT/viewability).
How to do it (copy this table)
| KPI |
What it diagnoses |
How to use it in decisions |
| Bid requests |
Available opportunity |
If huge but spend low → you’re filtering or underbidding |
| Bid rate |
Eligibility + filters |
Low bid rate → loosen stacked targeting/brand safety (carefully) |
| Win rate |
Competitiveness + auction dynamics |
Low win rate with normal bid rate → bids/caps/supply path issue |
| CPM |
Cost pressure |
CPM up with CPA up → tighten supply/signals or adjust caps |
| CTR |
Placement/creative fit |
CTR up but CVR down → clicky junk or mismatch |
| CVR |
Post-click quality |
CVR down across the board → landing page/offer/context |
| CPA |
Primary outcome |
Optimize to this with guardrails |
| Spend (vs plan) |
Delivery stability |
Under-delivery → check bid rate, win rate, caps, pacing |
| Frequency |
Waste risk |
Rising frequency with flat conversions → cap or widen audience |
| Viewability (if available) |
Attention proxy |
Very low viewability often correlates with poor outcomes |
| IVT (if available) |
Fraud/invalid traffic proxy |
High IVT → exclude sources/supply paths faster |
Benchmarks are context-dependent. Use trends and relative ranks (best vs worst sources) more than universal “good” numbers.
Step 2: Run a supply path optimization (SPO) sweep
What to do
- Reduce exposure to opaque reselling loops.
- Prioritize transparent, consistently performing paths.
- Use curated deals when they demonstrably improve quality.
Why it matters
Supply path is often the fastest way to improve CPA without touching bids.
How to do it
- Break down by exchange/SSP + domain/app + deal ID.
- Run a controlled test:
- 80–90% of budget stays stable
- 10–20% tests a cleaner path (e.g., curated PMP)
- Evaluate with lag awareness.
Step 3: Improve signal quality (and remove misleading signals)
What to do
- Prefer fewer, stronger signals over many weak ones.
- Remove segments that look “good” due to attribution artifacts.
Why it matters
In privacy-limited environments, weak identity signals and modeled conversions can mislead optimization. Your job is to create stable patterns: context, placement quality, and first-party data when available.
How to do it
- Separate reporting views:
- Prospecting: context, geo, device, environment
- Retargeting: recency, frequency, audience size
- If a segment only “works” under view-through but fails under click-through, treat it cautiously (not always wrong, but often fragile).
Step 4: Set bid boundaries (floors, caps, bid multipliers) with a testing protocol
What to do
- Use caps/ceilings to prevent overpaying.
- Use small bid increments and log every change.
Why it matters
Bid changes are high-impact and can hide supply quality problems.
How to do it
- If spend is low:
- check bid rate (eligibility)
- then check win rate (competitiveness)
- then adjust bids/caps
- Keep a rollback plan:
- “If CPM rises >X% and CPA worsens after the lag window, revert.”
Step 5: Implement pacing and budget guardrails
What to do
- Ensure each line item has enough budget to generate signal.
- Smooth intraday delivery where possible.
Why it matters
Too many tiny line items create unstable learning and volatile delivery.
How to do it
- Consolidate.
- Set a spend plan (expected daily spend band).
- Add alerts (under-delivery, CPA spikes, sudden win-rate drops).
Step 6: Apply frequency/recency controls by funnel stage
What to do
- Prospecting: cap frequency to protect reach.
- Retargeting: manage recency to avoid immediate saturation.
Why it matters
Frequency waste often hides behind “good win rate.”
How to do it
- Monitor frequency vs incremental conversions.
- If frequency rises and conversions don’t, you’re paying more for the same outcomes.
Step 7: Use structured experiments (one variable at a time) to avoid false learnings
What to do
- Change one major lever per day.
- Keep a stable control when possible.
Why it matters
RTB systems adapt. If you change bids, targeting, supply, and creative together, you won’t know what worked—and you’ll repeat the same mistakes.
How to do it (minimum viable experiment)
If you can’t run a clean A/B:
- Use time-based testing with strict logging:
- Day 1–2 baseline
- Day 3 change supply only
- Day 4–5 evaluate (with lag)
- Or split by geo (if volumes allow) as a rough control.
Practical Checklists and Templates (Copy/Paste)
RTB audit checklist (measurement, supply, signals, bids, controls)
Measurement
- Primary conversion event defined and verified (no double-fire)
- Deduplication rules documented (DSP vs analytics vs backend)
- Attribution windows documented (click + view)
- Conversion lag measured (median + 75th percentile)
Supply
- Top domains/apps by spend reviewed
- Top domains/apps by CPA reviewed
- Exclusion list includes high-spend/no-conversion sources (lag-aware)
- Transparency checks started (ads.txt/app-ads.txt, sellers.json where relevant)
Signals
- Bid rate checked (eligibility not accidentally crushed)
- Targeting reduced to a few strong signals
- Prospecting and retargeting separated
Bids
- Bid caps/ceilings set
- Bid changes logged and incremental
- Only one of: bids up OR targeting broadened
Controls
- Pacing mode chosen intentionally
- Frequency and recency set by funnel stage
- Alerts configured for under-delivery + CPA spikes
SPO checklist (direct paths, domain/app transparency, curated deals)
- Export report: exchange/SSP, deal ID, domain/app, placement (if available)
- Identify redundant paths (same domain via many resellers)
- Prefer authorized/direct paths when performance supports it
- Test curated PMP vs open auction with a controlled budget split
- Exclude sources with persistent quality issues (IVT/viewability/zero engagement)
Bid adjustment worksheet (when to raise bids vs broaden targeting)
Use this decision tree:
- Spend low
- Check bid rate
- If bid rate is low → loosen filters/stacked targeting first
- If bid rate is normal → check win rate
- Win rate low
- If CPM is stable and you’re losing auctions → raise bids slightly (or relax caps)
- If CPM is already high → fix supply/signals before bidding more
- Win rate high but CPA high
- Don’t raise bids.
- Tighten supply (exclusions/SPO), fix frequency, and improve signals.
Pacing + frequency starter settings (prospecting vs retargeting)
These are starting points, not universal rules. Adjust based on audience size, conversion cycle, and inventory.
Prospecting
- Goal: reach + qualified traffic at target CPA
- Frequency: low-to-moderate cap
- Pacing: smoother delivery (avoid front-loading)
- Controls: stronger exclusions, contextual focus
Retargeting
- Goal: efficient conversions without saturation
- Frequency: moderate cap (watch diminishing returns)
- Recency: enforce cool-down windows (avoid immediate repeat exposure)
- Pacing: stable, with a budget cap to prevent cannibalization
How to Connect Traffics.io Data to DSP Bidding Decisions (Cleaner Optimization Loops)
Many teams lose efficiency because the DSP optimizes on what it can observe, while acquisition analytics often tells you which traffic is actually engaged and converting.
Note on accuracy: Traffics.io capabilities can change. Confirm current integrations and exports in Traffics.io documentation and your account. The workflow below is vendor-agnostic and focuses on the loop, not a specific feature.
What “clean” acquisition data looks like
You want data that is:
- Deduped (one conversion counted once)
- Consistent events (stable naming and triggers)
- Bot/suspicious patterns flagged (at minimum: abnormal engagement, extreme bounce, impossible session patterns)
- Mapped to campaign metadata (UTMs, campaign IDs, placement identifiers where possible)
Map Traffics.io acquisition signals to DSP dimensions
Align Traffics.io reporting with DSP breakouts such as:
- Site/domain, app bundle
- Placement ID (if available)
- Exchange/SSP, deal ID
- Geo, device/OS, environment (in-app vs web)
- Time-of-day/day-of-week
Practical mapping tip
- Standardize naming: campaign names in the DSP should match UTMs (or use a simple lookup table).
- Keep a consistent “source key,” for example:
exchange|deal|domain/app|placement|geo|device|hour
Build a feedback loop: daily insights → supply/bid adjustments
A simple daily loop:
- Pull yesterday’s top spend sources in the DSP
- Pull yesterday’s quality outcomes in Traffics.io (engagement + conversion quality indicators)
- Classify sources:
- Scale: good quality + acceptable CPA
- Watch: mixed signals
- Suppress: high volume + low quality
Then translate that into DSP actions:
- Add exclusions (domain/app/placement)
- Shift budget toward better paths (deal/open/exchange)
- Apply bid multipliers (carefully) by geo/device/time when patterns repeat
- Tighten frequency where engagement drops
Example workflow: suppress high-volume, low-quality sources
Scenario:
- DSP shows an app bundle has great win rate and cheap CPM.
- Traffics.io shows sessions from that source have high bounce, near-zero meaningful engagement, and conversions that don’t match backend quality (or none).
Action:
- Add the app bundle (and similar bundles) to the DSP exclusion list.
- Reallocate budget to the best-performing sources from your “Keep / scale” list.
- Wait through your lag window before final judgment.
Governance: naming conventions and change logs so learning doesn’t get reset
To reduce “learning resets”:
- Keep a change log with:
- timestamp
- exact change (what was excluded, what bids changed)
- why (data evidence)
- expected effect
- evaluation date (after lag)
- Avoid renaming campaigns mid-sprint unless necessary.
- Avoid changing targeting, bids, and supply all at once.
If you want shorter feedback cycles, Traffics.io can work as an acquisition analytics layer to spot anomalies and quality gaps faster—then you push those learnings back into the DSP through exclusions, deal preferences, and bid boundaries. (Confirm which dimensions you can export in your Traffics.io setup.)
FAQ
What’s the difference between bid rate and win rate in RTB?
Bid rate is the share of bid requests where your DSP submits a bid (bids ÷ bid requests). Win rate is the share of those bids that win the auction (wins ÷ bids). If spend is low, check bid rate first (eligibility/filters), then win rate (competitiveness/supply path).
How do I improve win rate without blowing up CPA?
Compete with guardrails: (1) clean up supply (SPO + exclusions), (2) tighten signals so you bid on better contexts, (3) raise bids in small increments or broaden targeting (not both), (4) use bid caps/ceilings, and (5) control pacing and frequency to reduce waste.
Why can a high win rate still produce a high CPA?
Because you can win lots of cheap impressions that don’t convert (low-quality inventory, low viewability, higher invalid traffic, poor contextual fit) or because attribution is mis-crediting conversions. Win rate measures auction success, not outcome quality.
My CPA is good but delivery is inconsistent. What should I fix first?
Start with common spend chokepoints: overly strict targeting, low bid caps, tight frequency caps, and pacing mode conflicts (front-load then stall). Fix one major lever per 24 hours so you can attribute results.
What are the most useful guardrail KPIs during a sprint?
Use CPA as the primary metric, and guardrail with spend stability, win rate, CPM, CVR, frequency, viewability (if available), and IVT/fraud signals. Trend direction and relative rankings matter more than universal benchmarks.
What is Supply Path Optimization (SPO) in plain English?
SPO is buying impressions through fewer, more transparent, often more efficient routes (direct paths or curated deals) to reduce fees, improve transparency, and avoid low-quality reselling loops—while managing the scale trade-off.
How do ads.txt/app-ads.txt and sellers.json help buyers?
They improve transparency: ads.txt/app-ads.txt lists authorized sellers, and sellers.json provides seller identity/relationships. Use them in SPO work to prefer authorized/direct paths and to investigate suspicious selling chains.
Do bid floors hurt performance—and can buyers do anything about them?
Floors can reduce win rate if you’re below them and can push CPM higher if you must bid up. Buyers typically can’t change open auction floors, but you can shift supply paths, use deals with known pricing, adjust caps/targeting to protect CPA, and avoid inventories where floors force inefficient clearing prices.
What is bid shading, and when can it help?
Bid shading is a DSP feature (implementation varies) that may reduce what you pay in first-price auctions by bidding closer to the expected clearing price. Confirm behavior in your DSP documentation and monitor impact on win rate and delivery.
How should I set frequency and recency for prospecting vs retargeting?
Prospecting usually needs lower frequency to protect reach; retargeting can tolerate more frequency but benefits from recency controls to avoid immediate overexposure. If delivery collapses after caps, widen audience or increase bids before raising frequency aggressively.
How long should I wait before judging changes when conversions lag?
If lag is material (often 24–72 hours, but account-specific), don’t judge same-day. Use a hold period aligned to your lag distribution (e.g., through the 75th percentile lag) and compare against the pre-change baseline while keeping other levers stable.
How can Traffics.io data improve DSP decisions?
By tightening the loop between traffic quality and bidding. Use acquisition analytics to spot high-volume/low-quality sources (site/app/placement, geo, device, time), then translate that into exclusions, supply preferences, bid multipliers, and budget shifts. Confirm Traffics.io dimensions and exports against current documentation.
Key Takeaways: Your 14-Day Action Plan to Raise Win Rate and Lower CPA
What to do today (first 60 minutes)
- Build your baseline dashboard (bid rate, win rate, CPM, CVR, CPA, spend, frequency; plus viewability/IVT if available).
- Verify conversion tracking and dedupe.
- Measure conversion lag (at least a rough estimate).
- Export top sites/apps/placements by spend and CPA.
- Start a change log and commit to one major lever per 24 hours.
What to monitor daily vs weekly
Daily (fast checks)
- Spend vs plan (under-delivery is a symptom)
- Bid rate and win rate (eligibility vs competitiveness)
- CPA and CVR (efficiency and quality)
- Frequency (waste creep)
- IVT/viewability flags (if available)
Weekly (deeper work)
- SPO sweep: top supply paths and transparency checks
- Signal pruning: remove weak/misleading segments
- Structure review: consolidate line items to keep learning stable
- Lag-aware evaluation: judge changes after your hold period
The minimum viable optimization loop
- Diagnose: is it eligibility (bid rate), competitiveness (win rate), or quality (CVR/IVT/viewability)?
- Change one lever: supply or signals or bids or controls.
- Wait for lag: evaluate after your defined hold period.
- Scale winners / cut losers: move budget, don’t just observe.
- Feed better data back: use acquisition analytics (e.g., Traffics.io) to spot low-quality sources faster and translate them into exclusions and supply preferences.
Run the sprint cleanly and you’ll end with something more useful than a one-time “optimization”: a system you can repeat, measure, and defend with clear logs and baselines.
real-time bidding, rtb optimization, programmatic advertising, dsp optimization, improve win rate, reduce cpa, supply path optimization, bid strategy, pacing and budgeting, frequency capping, conversion tracking, attribution