Customer Acquisition

Customer Acquisition

Workflow

1. CAC Calculation

Blended CAC (company-level):

Blended CAC = (Total Sales + Marketing spend) / New customers acquired

Per-channel CAC (more actionable):

Channel CAC = Channel spend (ads + tools + headcount allocation) / Customers from that channel

Fully-loaded CAC (most accurate):

Fully-loaded CAC = (Ad spend + Sales salaries + Marketing salaries + Tools + Agency fees + Content production) / New customers

What to include:

| Include | Don't include | |---------|---------------| | Ad spend (all platforms) | Product development costs | | Sales team compensation (base + commission) | Customer success costs | | Marketing team compensation | Infrastructure/hosting | | Marketing tools (HubSpot, analytics, etc.) | General overhead (rent, legal) | | Content production costs | | | Agency/contractor fees | | | Event/sponsorship costs | |

2. Channel Evaluation

> There is no universal channel CAC. A "$150 Google CAC" is meaningless without industry, ACV, country, the funnel stage you count as a "customer" (lead vs trial vs paid vs net-of-refund), gross margin, and the measurement window. Benchmark against your own history and unit economics, not a generic table. Derive each channel's CAC from the formulas in §1, then score relative scalability/time/quality.

Scoring matrix — fill CAC from YOUR data (§1), score the rest 1–5:

| Channel | Your CAC (compute) | Scalability | Time to first result | Acquired-cohort LTV/quality | Score | |---------|--------------------|-------------|---------------------|-----------------------------|-------| | Organic search / SEO | $___ | High | 6–12 mo | Often high intent | | | Paid search (Google) | $___ | High | Immediate | High intent, capped by query volume | | | Paid social (Meta Advantage+) | $___ | High | 1–2 wk | Varies by creative/offer | | | LinkedIn ads | $___ | Medium | 1–2 wk | High for B2B/high-ACV | | | Content / thought leadership | $___ | High | 3–6 mo | Compounding, high quality | | | Referral program | $___ | Medium | 1–3 mo | Usually highest LTV, lowest CAC | | | Outbound (SDR/cold) | $___ | Medium | 2–4 wk | High if ICP-targeted | | | Partnerships / co-marketing | $___ | Low–Med | 3–6 mo | High, trust-transferred | | | Events / field | $___ | Low | 1–3 mo | High-touch enterprise | | | Product-led (PLG/viral) | $___ | Very high | Varies | Varies; watch self-serve→paid rate | |

**Order-of-magnitude CAC ranges by motion (illustrative / synthetic — calibrate to your market, not gospel):**

| Sales motion | Typical CAC range* | First-year ACV it must support | Notes | |--------------|--------------------|--------------------------------|-------| | B2C consumer app/subscription | $5–80 | $30–300 | Margin-thin; payback must be fast | | B2C high-AOV ecommerce | $20–200 | $100–1,000+ | Watch first-order vs repeat LTV | | PLG / self-serve SaaS (SMB) | $50–600 | $300–3,000 | Mostly automated; CAC = ads + a little ops | | Inside-sales / mid-market SaaS | $1,000–8,000 | $8k–60k | Sales comp dominates CAC | | Enterprise field sales | $15k–100k+ | $60k–500k+ | Long cycle; allocate by close date |

\*Ranges are synthetic illustrations, vary 5–10× by geography (US/UK CPMs ≫ LATAM/SEA), niche competitiveness, and what you count as a conversion. Never quote them externally as benchmarks.

3. Attribution Models

| Model | How it works | Best for | Bias | |-------|-------------|----------|------| | First touch | 100% credit to first interaction | Understanding discovery | Over-credits awareness channels | | Last touch | 100% credit to last interaction | Understanding conversion | Over-credits bottom-funnel | | Linear | Equal credit to all touchpoints | Simple multi-touch | Treats all touches equally (unrealistic) | | Time decay | More credit to recent touchpoints | Long sales cycles | Under-credits awareness | | Position-based (U-shape) | 40% first, 40% last, 20% middle | Balanced view | Arbitrary weights | | Data-driven (DDA) | ML/Shapley-style weights from observed paths | High-volume, well-instrumented funnels | Black box; trained on modeled + consented data only — biased by consent loss and platform self-attribution |

> The old "DDA needs 1,000+ conversions" rule is obsolete. Platform DDA (GA4, Google Ads) now runs on far less but fills gaps with modeled/estimated conversions and only sees consented users. Ad platforms also self-attribute (each claims the same conversion), so summed platform-reported conversions routinely exceed real total sales by 20–60%.

**Correlational attribution (first/last/DDA) tells you paths, not causation. Use it for directional reads and budget pacing, but settle real channel value with incrementality (§3a).**

How to actually use attribution:

1. Pick one last-non-direct multi-touch model (or platform DDA) as your day-to-day pacing view — consistency beats theoretical purity.
2. Run first-touch in parallel to credit awareness/demand-gen channels the conversion model starves. Disagreement = a candidate for an incrementality test, not a verdict.
3. Reconcile every model against the CRM/finance source of truth (closed-won, net of refunds). De-dupe platform-reported conversions; never sum them.
4. For anything you spend real money to scale, validate with §3a incrementality before reallocating budget.
5. Capture self-reported attribution ("How did you hear about us?") at signup — it's the single best counterweight to dark-social and view-through blind spots that no pixel sees.

3a. Incrementality (the truth layer above attribution)

Attribution answers "which touchpoints were on the path?" Incrementality answers "what would have happened anyway?" — the only number that justifies scaling spend. A channel can win in last-touch reporting yet be ~0% incremental (e.g., brand search, retargeting already-converting users).

| Method | How it works | Best for | Gotchas | |--------|--------------|----------|---------| | Geo split / matched-market test | Hold out spend in matched regions; compare conversions vs control geos | Mid/large budgets, any channel | Needs enough geos + spend contrast; use a matched-market or synthetic-control design | | Conversion lift (platform) | Platform randomizes exposed vs holdout, reports lift | Meta/Google/LinkedIn at sufficient spend | You trust the platform's own holdout; spend minimums apply | | PSA / ghost ads | Control group sees a placebo (PSA) or "would-have-served" ghost ad | Display/video where supported | Limited availability; ghost-ad support varies by platform | | Holdout cohort | Withhold a channel/audience % entirely for a period | Email, push, retargeting, lifecycle | Discipline to keep holdout untouched; measure on net revenue | | MMM (media mix modeling) | Regression/Bayesian model of spend→outcome across all channels | Privacy-durable, top-down, cross-channel | Needs 1.5–2+ yr of data; can't see individuals — pairs with geo tests for calibration | | Scaled-spend (CAC-curve) test | Step spend up/down, watch marginal CAC | Pacing a single proven channel | Confounded by seasonality/auctions; change one thing at a time |

Workflow to reconcile with reporting:

1. Run an experiment (geo split or platform lift) on the channel in question.
2. Compute incremental CAC = test spend ÷ incremental conversions (not platform-reported).
3. Derive an attribution multiplier = incremental / last-touch-reported conversions. Apply it to deflate that channel's reported numbers between tests.
4. MMM for the top-down allocation; experiments to calibrate the MMM; attribution for daily pacing — this triangulation is the 2026 best-practice stack. Re-test quarterly or when CAC drifts >20%.

3b. 2026 Measurement Stack (privacy-era reality)

Browser/device privacy has broken naive pixel tracking. Plan acquisition measurement around these constraints — they directly inflate reported CAC and shrink observable conversions:

• iOS ATT + SKAN (SKAdNetwork/AdAttributionKit): Most iOS users are opted out of IDFA. App-install/in-app conversions arrive aggregated, delayed, and with coarse conversion values via SKAN postbacks — no user-level join. Optimize on SKAN conversion-value schemas, expect a 24–72h+ reporting lag, and never compare iOS SKAN CAC apples-to-apples with web CAC.
• Consent Mode / cookie loss: With third-party cookies degraded and consent banners required (EU/EEA/UK + expanding US state laws), a large share of conversions are unobserved. Google Consent Mode v2 (required for EEA personalized ads/measurement) lets platforms model the conversions consent-denied users would have generated.
• Modeled / estimated conversions: GA4 and the ad platforms now report a blend of observed + modeled conversions. Treat platform conversion counts as estimates with error bars, not ground truth — reconcile to CRM/finance (§3, §5).
• Server-side tagging + first-party data: Move tagging server-side (Google Tag Manager Server-Side, or a CDP) to improve match rates, control PII, and reduce reliance on browser cookies. Send first-party events server-to-server.
• Conversions APIs (server-to-server): Send offline/closed-won and web conversions back to platforms via Meta Conversions API (CAPI), Google Enhanced Conversions / offline conversion import, and LinkedIn Conversions API — with hashed first-party identifiers (email/phone) for matching. This is now table-stakes for B2B (upload closed-won, not just form fills) and for recovering signal post-cookie.
• Data clean rooms: For deduplicated cross-platform reach/conversion analysis without sharing raw user data (e.g., Google Ads Data Hub, Amazon Marketing Cloud, Meta Advanced Analytics). Use when you need cross-channel overlap/dedup at scale.
• Deduplication: A single sale is claimed by multiple platforms. Use a deterministic event ID across pixel + CAPI to dedup, and always net platform totals back to one CRM source of truth.

> Practical default for a new program: GA4 + server-side GTM + CAPI/Enhanced Conversions on every paid channel + Consent Mode v2 + self-reported attribution at signup + CRM closed-won as the arbiter. Exact setup steps and quotas change frequently — verify against the official docs (Google Ads/GA4, Meta Business, LinkedIn Marketing) as of Jun 2026.

4. LTV:CAC Analysis

Benchmarks by stage:

| Metric | Seed/Early | Series A | Series B+ | |--------|-----------|----------|-----------| | LTV:CAC ratio | > 2:1 | > 3:1 | > 4:1 | | CAC payback | 5:1 | | Inside sales mid-market | $500-2,000 | $5,000-30,000 | > 3:1 | | Enterprise field sales | $5,000-50,000 | $50,000-500,000 | > 3:1 |

Payback period (always gross-margin-adjusted):

Payback (months) = CAC / (Monthly ARPU × Gross margin %)

Using revenue instead of gross-margin-adjusted contribution overstates payback speed — a 70%-margin SaaS and a 25%-margin marketplace with identical ARPU have very different real paybacks.

CAC cohorting — never trust blended/period CAC alone:

• Cohort by acquisition month, not reporting month. Spend in March acquires customers who pay back over later months; matching this-month spend to this-month revenue distorts both ratios (lethal when spend is growing fast).
• LTV by channel × segment, not company-wide. Referral and SEO cohorts usually retain far better than paid-social cohorts at the same headline CAC — blended LTV hides this.
• Adjust LTV for refunds, chargebacks, and early churn. Use net revenue retention and survival curves; a 14-day-refund-heavy cohort has lower realized LTV than gross bookings imply.
• Split sales-assisted vs self-serve even within one channel. Fully-loaded CAC must carry the SDR/AE/CS-onboarding cost onto the assisted cohort, or you'll over-credit self-serve.
• Pick and document an attribution lookback window (e.g., 30/60/90-day click; view-through separately) and hold it constant — changing windows silently re-prices every channel's CAC.

5. Channel Saturation Signals

When to diversify (channel is saturating):

• Rising marginal CAC: incremental CAC (§3a) climbs as you add spend — 2x budget ≠ 2x conversions. This is the real saturation signal; the rest are proxies.
• CAC up >20% over 3 months with no strategy/auction-mix change
• Search impression share ceiling (Google Ads "lost IS (budget)" → 0 while "lost IS (rank)" high = you've maxed quality, not budget)
• Creative/audience fatigue on paid social — there is no universal frequency threshold. Read it from the data: rising frequency and falling CTR/CVR and rising CPA together. Tolerable frequency depends on audience size, creative-refresh cadence, buying cycle, and objective (prospecting vs retargeting); a 7-day frequency of 1.5 can fatigue a tiny retargeting pool while 4+ is fine for a broad prospecting audience with fresh creative.
• Organic/SEO plateau despite continued investment — and check whether AI Overviews / AI search answers are absorbing clicks (see §8).

Response (derive the test budget; don't hardcode it):

1. Optimize the existing channel before abandoning (offer, creative, landing page, bid strategy).
2. Size the new-channel test from learning requirements, not a flat %. You need enough budget to detect a CAC at-or-below your bar within an acceptable time:

`` Min conversions to learn ≈ derived from your MDE (minimum detectable effect) & baseline CVR Min test spend ≈ Min conversions × expected CAC × safety factor (1.5–2×) Min test duration ≥ sales cycle + conversion lag (don't read a 60-day-cycle channel at week 2) `` For most paid channels the platform also needs a minimum events/week to exit the learning phase (~50 optimization events/week is the common rule of thumb) — fund at least that or the algorithm never optimizes. 10–15% of budget is a starting sanity check, not the rule.

1. Run for at least one full sales cycle + conversion lag (often 60–90 days; longer for enterprise) before judging.
2. Compare the new channel on incremental CAC and cohort LTV (§3a, §4), not platform-reported CAC.
3. Scale when incremental CAC is competitive with your best channel and the marginal-CAC curve still has headroom.

6. Budget Allocation Framework

Portfolio approach:

| Category | % of budget | Purpose | |----------|------------|---------| | Proven channels | 60-70% | Channels with known, acceptable incremental CAC | | Scaling channels | 20-25% | Channels showing promise, increasing spend | | Experimental | 10-15% | New channels, testing hypotheses |

> Treat these splits as a default heuristic, not a constraint. The experimental slice must still clear each test's minimum learning spend (§5) — if 15% can't fund one channel past its learning phase, run one test properly rather than three underfunded ones. Mature, capital-efficient programs often run leaner experimentation (~5–10%); early-stage discovery may justify 20%+.

Rebalance quarterly:

• Move budget from declining-ROI channels to improving ones
• Kill experiments that haven't shown promise in 90 days
• Double down on channels where LTV:CAC is improving

7. Acquisition Dashboard

| Metric | Cadence | View | |--------|---------|------| | Blended CAC | Monthly | Trend line, 6-month rolling | | Channel CAC | Monthly | Per-channel bar chart | | LTV:CAC by channel | Quarterly | Stacked comparison | | Payback period | Monthly | Trend vs target | | New customer count by source | Weekly | Stacked area chart | | CAC efficiency (CAC / ARPU) | Monthly | Track improvement | | Pipeline contribution by channel | Weekly | Marketing → Sales attribution | | Incremental CAC (last test) | Per test / quarterly | Channel vs last-touch multiplier (§3a) | | Self-reported source mix | Monthly | "How did you hear about us?" vs pixel attribution | | SKAN vs web CAC (if mobile) | Monthly | Tracked

…

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: san-npm
• Source: san-npm/skills-ws
• License: MIT
• Homepage: https://skills-ws.vercel.app

Install and usage instructions live in the source repository linked above.