Marketing Analytics

First-touch credits all value to the initial touchpoint (awareness heavy); last-touch credits the final interaction (conversion heavy). Multi-touch distributes credit across the journey: linear (equal weight), time-decay (recent touches weighted higher), U-shaped (40% first, 40% last, 20% middle), or algorithmic (ML-based on historical data). No single 'correct' model, choice depends on your campaign structure and business goals. Start with time-decay if unsure, validate against incrementality testing before investing heavily.

LTV = (Annual revenue per customer) × (Gross margin %) ÷ (Annual churn rate). More precise: LTV = Σ(revenue in month t × (1 - churn_rate)^t) discounted back. Common mistake: using raw annual revenue without margin. If a customer generates $1000 annually but your margin is 40%, LTV ≠ $1000. Cohort-based LTV (track a cohort month-by-month, sum total revenue) is more reliable than formula-based for early-stage companies with changing unit economics.

MTA uses clickstream data (what touchpoints the user saw/clicked) to assign credit; works best for digital/trackable channels, struggles with dark social and offline. MMM uses aggregate sales data and external variables (seasonality, competitor spend, weather) to estimate channel contribution; works for all channels (TV, radio, offline, paid search) but can't attribute individual conversions. Best practice: use both. MTA for daily optimization (which campaign did this customer respond to?), MMM for strategic planning (should we shift budget from TV to digital?). MTA favors short-term, MMM favors long-term planning.

CAC payback = (Total CAC) ÷ (Monthly contribution margin per customer). If CAC is $100 and monthly margin is $50, payback is 2 months. Rule of thumb: payback <12 months is healthy, <6 months is strong, >12 months signals unit economics problems. Payback period + LTV together determine business viability: CAC payback 3 months + LTV 24 months = 8x LTV:CAC ratio (excellent, implies $7 margin per $1 spent). Most VCs expect LTV:CAC ≥ 3:1; enterprise SaaS targets 3-5 year payback, e-commerce targets <3 months.

Automate anomaly detection and baseline metrics on dashboards (CAC, LTV, payback, cohort retention curves, channel attribution tables). Investigate manually when: (1) anomaly hits a guardrail, (2) a campaign unexpectedly underperformed, (3) you're trying to understand 'why' not just 'what'. A dashboard showing 'CAC up 20%' is not useful without deeper analysis, did conversion rate drop, did CPM rise, did audience quality shift? Manual investigation answers 'why'. Modern tools (Mixpanel, Amplitude, GrowthBook) blend both: automated dashboards + drill-down exploration interfaces.

iOS 14.5+ lets users opt out of tracking; ~85% do. Result: Facebook/iOS apps can't see the full conversion path, attribution breaks, ROAS becomes noisy. Solutions: (1) Implement first-party data collection (pixel firings on web, server-to-server events); (2) Use conversion API / Conversions API (Facebook, Google) to send conversions server-side, bypassing the SDKs; (3) Run incrementality tests (hold-out groups) to estimate true ROAS independent of attribution; (4) Adopt MMM for budget planning; (5) Rely on LTV cohort analysis (long-term user value) not short-term ROAS. Expect 20-40% loss of data visibility and don't panic, everyone is in the same boat, competitors are blind too.

In-house: slow ramp (3-6 months to productive), expensive ($100k+ headcount), but develops deep institutional knowledge, can integrate with product/eng, owns long-term roadmap. Agency: fast execution, specialized skills, lower commitment, but lacks context, hard to iterate, expensive ($3-10k/mo), and turns over analysts. Best practice: hire in-house analyst (L2-L3 seniority) to own strategy + dashboard, use agency for one-off deep dives (cohort analysis, MMM modeling, incrementality testing). For startups <$5M ARR with limited data complexity, agencies are faster. For >$10M ARR, hire in-house.