# Paid Search (Google Ads Quality Score, Bidding, Match Types, PMax), Meta Ads, CRO, A/B Testing Statistics & SaaS Unit Economics

> Ad Rank formula, Broad/Phrase/Exact match, Smart Bidding, Meta CAPI & EMQ, frequentist vs Bayesian A/B testing, SRM, and CAC/LTV/ROAS/MER/NRR

- **Canonical URL:** https://flashcards.world/flashcards/sets/ad2d8843-3369-4063-a9d9-84ffccfc20d2/
- **Markdown URL:** https://flashcards.world/flashcards/sets/ad2d8843-3369-4063-a9d9-84ffccfc20d2.md
- **Level:** advanced
- **Languages:** en → en
- **Total Cards:** 32

## Flashcards (32 Cards)

| # | Front (Term / Prompt) | Back (Definition / Answer) | Example | Multiple-Choice Options |
| :--- | :--- | :--- | :--- | :--- |
| 1 | What is the 'Peeking Problem' in fixed-horizon frequentist A/B testing, and how does Sequential Testing solve it? | In a standard frequentist z-test/t-test, checking the p-value daily and stopping the test as soon as p < 0.05 inflates the actual false positive rate (Type I error) from 5% to 25–30%+. Sequential Testing (using alpha-spending functions like O'Brien-Fleming or Always-Valid p-values / mSPRT) adjusts significance boundaries across continuous monitoring so false positive rates stay bounded below α. |  |  |
| 2 | How do Baseline Conversion Rate (p), Minimum Detectable Effect (MDE), Significance Level (α), and Statistical Power (1 − β) affect required A/B test sample size? | Required sample size per variation increases when: 1. You try to detect a smaller Minimum Detectable Effect (sample size scales inversely with the square of MDE: n ∝ 1 / MDE²); 2. Baseline conversion rate is smaller (for relative MDE); 3. You demand higher confidence (lower α, e.g., 99% vs 95%) or higher power (1 − β, e.g., 90% vs 80%). |  |  |
| 3 | In Conversion Rate Optimization (CRO), how does a 'Multi-Armed Bandit' (e.g., Thompson Sampling or Upper Confidence Bound) differ from a classic A/B test? | A classic A/B test keeps traffic allocation fixed (e.g., 50/50) for the full test duration to maximize statistical inference on all variants (pure exploration). A Multi-Armed Bandit dynamically shifts traffic toward better-performing variants in real time (balancing exploration and exploitation), making it ideal for short-lived seasonal promotions or headline tests where minimizing regret outweighs causal precision. |  |  |
| 4 | Compare Google Ads Keyword Match Types: Exact Match ('[keyword]'), Phrase Match ('"keyword"'), and Broad Match ('keyword') in modern semantic matching. | Exact Match ('[running shoes]') matches queries with the same exact meaning or intent as the keyword; Phrase Match ('"running shoes"') matches queries that include the meaning of the keyword in a more specific form (preserving word-order dependency only when order alters meaning); Broad Match ('running shoes') matches any query semantically related to the keyword and leverages additional auction-time signals (user recent searches, landing page content) when paired with Smart Bidding. |  |  |
| 5 | In Meta Ads (Facebook/Instagram), how do you prevent double-counting conversions when firing both the browser Meta Pixel and Server-Side Conversions API (CAPI)? | Send both the browser Pixel event and the server CAPI event with the exact same 'event_name' (e.g., 'Purchase') and the exact same deterministic 'event_id' string (e.g., the order ID or a UUID generated in the dataLayer). Meta's ingestion engine deduplicates events sharing the same 'event_name' and 'event_id' within a 48-hour window. |  |  |
| 6 | Compare the four automated Google Ads Smart Bidding strategies: 'Maximize Conversions', 'Target CPA (tCPA)', 'Maximize Conversion Value', and 'Target ROAS (tROAS)'. | 'Maximize Conversions' spends the full daily budget to get the highest volume of conversions regardless of individual unit cost; 'tCPA' bids to acquire as many conversions as possible at or below a target average cost per action; 'Maximize Conversion Value' spends the budget to maximize total revenue/points; 'tROAS' adjusts bids based on predicted conversion value per auction to hit a target Return on Ad Spend ratio. |  |  |
| 7 | What are the three components of Google Ads 'Quality Score' (1–10 scale), and how does the 1–10 number in the UI differ from auction-time quality? | The three components (rated Below Average, Average, or Above Average) are: 1. Expected Click-Through Rate (CTR); 2. Ad Relevance; 3. Landing Page Experience. The 1–10 Quality Score column in the UI is a diagnostic historical summary rolled up at the keyword level, whereas the live auction calculates high-precision real-time quality scores for the exact query, device, and user context. |  |  |
| 8 | In Meta Ads auction mechanics, what are the three components of the 'Total Value' formula that determines which ad wins the impression? | Total Value = (Advertiser Bid × Estimated Action Rate) + Ad Quality (User Value). Here, Estimated Action Rate is Meta's ML prediction of the probability that the user will take your optimization event action, and Ad Quality reflects positive/negative user feedback, engagement bait penalties, and post-click experience. |  |  |
| 9 | What is 'Event Match Quality' (EMQ, scored out of 10) in Meta Conversions API (CAPI), and which customer parameters increase it most? | EMQ measures how effectively the customer information parameters sent with a server CAPI event match a Meta user profile for attribution and audience optimization. The highest-impact parameters are 'fbc' (Click ID cookie from 'fbclid'), 'fbp' (Browser ID cookie), SHA-256 hashed 'em' (Email) and 'ph' (Phone Number), External ID, Client IP Address, and User-Agent. |  |  |
| 10 | In Google Ads and Paid Search, how is 'Search Impression Share' broken down into 'Search Lost IS (Rank)' vs 'Search Lost IS (Budget)'? | Search Impression Share = Impressions Received / Total Eligible Impressions. 'Search Lost IS (Budget)' is the percentage of eligible impressions your ads missed because your daily campaign budget was exhausted early. 'Search Lost IS (Rank)' is the percentage of eligible impressions missed because your Ad Rank (Bid × Quality Score) was too low to win a slot in the auction. |  |  |
| 11 | Distinguish ROAS (Return on Ad Spend), POAS (Profit on Ad Spend), and MER (Marketing Efficiency Ratio / Blended ROAS). | ROAS = Platform-Attributed Revenue / Ad Spend; POAS = Gross Profit Attributable to Ads (Revenue − COGS − Shipping/Fulfillment) / Ad Spend (preventing high-ROAS, negative-margin campaigns from scaling); MER (Blended ROAS) = Total Company Revenue across all channels / Total Marketing Ad Spend (measuring holistic top-line efficiency independent of platform attribution overlap). |  |  |
| 12 | How do you calculate the Breakeven ROAS and Maximum Allowable CPA for an e-commerce product with a $120 Average Order Value (AOV), $42 COGS, and $6 payment/fulfillment fee? | Contribution Margin before Ad Spend = $120 − $42 − $6 = $72 (which is a 60% Contribution Margin ratio: $72 / $120 = 0.60). The Maximum Allowable CPA to break even on the first order is $72.00, and the Breakeven ROAS = 1 / Contribution Margin % = $120 / $72 = 1 / 0.60 = 1.67x. |  |  |
| 13 | How do Google Ads Negative Keyword match types differ critically from positive keyword match types? | Unlike positive match types, Negative Keywords do NOT match close semantic variants, synonyms, or implied intent! Negative Exact ('-[running shoes]') blocks only that exact query with no extra words; Negative Phrase ('-"running shoes"') blocks queries containing those exact words in that exact consecutive order; Negative Broad ('-running shoes') blocks any query containing all of those words in any order—and you must explicitly add plurals, misspellings, and synonyms. |  |  |
| 14 | What is the 'SaaS Magic Number' and the 'Burn Multiple', and how do investors use them to evaluate capital efficiency? | SaaS Magic Number = [(Current Quarter GAAP Revenue − Previous Quarter GAAP Revenue) × 4] / (Previous Quarter Total Sales & Marketing Expense), where >= 0.75–1.0 indicates efficient S&M spend. Burn Multiple = Net Cash Burned in Period / Net New ARR Added in Period, where < 1.5x is considered strong capital efficiency. |  |  |
| 15 | Compare Frequentist A/B testing with Bayesian A/B testing in how they report results to decision-makers. | Frequentist testing assumes fixed unknown parameters and outputs a p-value (P(Data \| H_0 is true)) and confidence interval around the difference. Bayesian testing combines a Prior distribution (e.g., Beta(α, β)) with observed Binomial likelihood data to compute a Posterior distribution, directly reporting the 'Probability that Variant B beats Control A' and the 'Expected Loss' (risk) of choosing B. |  |  |
| 16 | What is 'Sample Ratio Mismatch' (SRM) in an A/B test, how is it detected statistically, and why does it invalidate test results? | SRM occurs when the observed split of users across Control and Variant deviates significantly from the configured allocation ratio (e.g., a 50/50 test yields 50,800 Control vs 49,200 Variant users). It is detected using a Chi-Square Goodness-of-Fit test (typically flagged at p < 0.001) and indicates a critical bug (redirect drops, bot filtering bias, or slow variant script timeouts) that invalidates any observed conversion lift. |  |  |
| 17 | Compare Gross Revenue Retention (GRR) and Net Revenue Retention (NRR / NDR) in SaaS unit economics, including their formulas and upper bounds. | Starting from a cohort's Beginning MRR: GRR = (Beginning MRR − Churn MRR − Contraction/Downgrade MRR) / Beginning MRR (strictly capped at <= 100%). NRR = (Beginning MRR + Expansion/Upsell MRR − Contraction MRR − Churn MRR) / Beginning MRR (can exceed 100%, with top enterprise SaaS companies achieving 115–130%+ NRR). |  |  |
| 18 | How is 'Fully Loaded Customer Acquisition Cost' (CAC) calculated in B2B SaaS, and how does 'Paid CAC' differ from 'Blended CAC'? | Fully Loaded CAC = (Total Sales & Marketing Expenses including ad spend, salaries, commissions, software tools, and agency fees in period t) / (Number of New Customers Acquired in period t, often cohort-lagged by sales cycle length). Paid CAC divides paid acquisition spend by customers acquired solely through paid channels, whereas Blended CAC divides total S&M spend by all new customers including organic and word-of-mouth. |  |  |
| 19 | Calculate the LTV, LTV:CAC Ratio, and Gross-Margin-Adjusted CAC Payback Period for a SaaS company with ARPA = $200/month, Gross Margin = 80%, Monthly Churn = 2%, and CAC = $1,600. | Monthly Gross Profit per Customer = $200 × 0.80 = $160. Expected Lifetime = 1 / 0.02 = 50 months. LTV = $160 × 50 = $8,000. LTV:CAC Ratio = $8,000 / $1,600 = 5.0:1 (above the healthy >= 3.0:1 SaaS benchmark). Gross-Margin-Adjusted CAC Payback Period = $1,600 / $160 = 10 months. |  |  |
| 20 | How does Google Performance Max (PMax) interact with Standard Search campaigns in the same Google Ads account? | If a user's search query identically matches an eligible Exact Match keyword in a Standard Search campaign, the Standard Search campaign takes priority over PMax. For non-identical queries (Phrase, Broad, or semantic matches), whichever campaign (Standard Search vs PMax) has the highest Ad Rank wins the internal auction. |  |  |
| 21 | In e-commerce cohort analysis, why can a linear '1 / Churn Rate' formula fail for non-contractual retail brands, and what probabilistic model is used instead? | In non-contractual retail (e.g., apparel or supplements without subscriptions), customers never formally announce 'churn'—they simply stop buying quietly. Analysts use the BG/NBD (Beta-Geometric / Negative Binomial Distribution) model paired with the Gamma-Gamma monetary value model to estimate each customer's probability of still being 'alive' (P(Alive)) and predicted future transaction count from their Recency and Frequency (RFM) history. |  |  |
| 22 | What is 'Simpson's Paradox' in digital marketing and A/B testing, and how can a Variant lose in every single device segment yet appear to win overall? | Simpson's Paradox occurs when a confounding variable (such as Desktop vs Mobile traffic mix) is unequally distributed between Control and Variant. If Desktop converts at 6% and Mobile converts at 1%, and Variant B accidentally receives a higher proportion of Desktop traffic (or traffic shifts during the test), Variant B can show a higher aggregate conversion rate despite having a lower conversion rate than Control A on both Desktop and Mobile individually. |  |  |
| 23 | In Conversion Rate Optimization (CRO), compare the 'ICE', 'PIE', and 'PXL' prioritization frameworks for ranking experiment hypotheses. | ICE (Impact, Confidence, Ease) and PIE (Potential, Importance, Ease) use subjective 1–10 scores that are vulnerable to HiPPO (Highest Paid Person's Opinion) bias. PXL (created by CXL) replaces subjective guesses with objective binary criteria: Is the change above the fold? Noticeable within 5 seconds? Backed by qualitative user testing, heatmaps, or analytics drop-offs? Running on high-traffic pages? |  |  |
| 24 | State the formula for Customer Lifetime Value (LTV) in a subscription SaaS business using ARPA, Gross Margin %, and Monthly Logo Churn Rate. | LTV = (ARPA × Gross Margin %) / Monthly Customer Churn Rate — where ARPA is Average Revenue Per Account per month, Gross Margin % = (Revenue − COGS) / Revenue (always include Gross Margin so LTV reflects gross profit rather than raw top-line revenue), and 1 / Churn Rate is the expected customer lifetime in months. |  |  |
| 25 | How is Google Ads 'Ad Rank' calculated in each real-time auction, and how does it determine the Actual CPC paid by the winning advertiser? | Ad Rank = Max CPC Bid × Quality Score (Expected CTR + Ad Relevance + Landing Page Experience) + Context Signals (device, location, time, search query intent) + Expected Impact of Ad Assets (extensions). Under generalized second-price auction mechanics, Actual CPC = (Ad Rank of the advertiser directly below you / Your Quality Score) + $0.01 (subject to Ad Rank thresholds). |  |  |
| 26 | In A/B Testing (Conversion Rate Optimization), define Type I Error (α), Type II Error (β), and Statistical Power (1 − β). | Type I Error (α, false positive rate, typically set at 0.05 for 95% confidence) is rejecting the null hypothesis when there is actually zero real difference between Control and Variant. Type II Error (β, false negative rate, typically set at 0.20) is failing to detect a true lift when one genuinely exists. Statistical Power (1 − β, typically 80%) is the probability of correctly detecting a real effect of at least the Minimum Detectable Effect (MDE). |  |  |
| 27 | How does CUPED (Controlled-experiment Using Pre-Experiment Data) reduce variance and shorten A/B test runtimes? | CUPED uses a pre-experiment covariate X (such as a user's purchase revenue or activity in the 14 days prior to entering the test) that correlates with the experiment metric Y, computing the adjusted metric Y_cv = Y − θ(X − E[X]), where θ = Cov(Y, X) / Var(X). This removes pre-existing user variance and shrinks required sample size by 1 − r² (where r is the correlation between X and Y). |  |  |
| 28 | What is the 'SaaS Quick Ratio' and the 'Rule of 40', and what benchmarks indicate healthy SaaS growth efficiency? | The SaaS Quick Ratio = (New MRR + Expansion MRR) / (Churned MRR + Contraction MRR), where a ratio >= 4.0 indicates rapid, healthy net MRR growth. The Rule of 40 states that a SaaS company's (YoY Revenue Growth Rate %) + (Free Cash Flow Margin % or EBITDA Margin %) should equal or exceed 40%. |  |  |
| 29 | Why must A/B tests run in full 7-day weekly increments (e.g., 7, 14, 21, or 28 days) even if the required sample size is reached in 3 days? | User intent, traffic sources, device mix (desktop workdays vs mobile weekends), and conversion rates fluctuate strongly by day of the week. Stopping an experiment after 3 days introduces day-of-week cyclical bias and fails to distinguish genuine conversion lift from a temporary Novelty Effect or Primacy Effect. |  |  |
| 30 | What is Google Ads 'Enhanced Conversions for Web' vs 'Enhanced Conversions for Leads'? | 'Enhanced Conversions for Web' captures first-party customer data (email, phone, address) entered on a checkout/lead page, normalizes and hashes it with SHA-256, and matches it against signed-in Google accounts to recover conversions lost to cookie restrictions. 'Enhanced Conversions for Leads' hashes first-party lead form data on the website and later joins it when the CRM uploads the offline closed-won sale without requiring 'GCLID' storage. |  |  |
| 31 | Calculate Incremental ROAS (iROAS) and Cost Per Incremental Acquisition (iCPA) given a Geo-Holdout test where $50,000 ad spend generates $250,000 revenue (1,000 orders) in the Treatment group vs $150,000 baseline revenue (600 orders) in the matched Control group. | Incremental Revenue = $250,000 − $150,000 = $100,000, so iROAS = $100,000 / $50,000 = 2.0x (compared to a naive 5.0x platform ROAS). Incremental Conversions = 1,000 − 600 = 400 orders, so iCPA = $50,000 / 400 = $125.00 (compared to a naive $50.00 CPA). |  |  |
| 32 | What is 'Incrementality Testing' (Geo-Lift / Holdout Testing) in performance marketing, and why can a campaign have an 8.0x platform ROAS but near-zero incremental lift? | Incrementality testing divides users or geographic regions (DMAs) into a Treatment group (shown ads) and a Control Holdout group (no ads) to measure true causal lift: Incremental Lift = Conversions_Treatment − Conversions_Scaled_Control. Retargeting and branded search campaigns often report an 8.0x+ platform ROAS simply by claiming credit for repeat customers who were already going to buy organically. |  |  |

---
Study this deck interactively with FSRS-6 spaced repetition, audio review, and multiple-choice mode at [Flashcards World](https://flashcards.world/flashcards/sets/ad2d8843-3369-4063-a9d9-84ffccfc20d2/).