Growth & Marketing Analytics

§1 The Growth Accounting Framework Foundational

Growth is not a rate. It's an identity. Before you can improve growth, you have to decompose it — and the most useful decomposition is the growth accounting equation: net new users = new users + reactivated users − churned users. This seems obvious until you actually split your growth metrics this way and realize that most "growth" conversations collapse all three into a single top-line number, which makes it impossible to know what's actually driving the headline.

A company can show 20% month-over-month user growth while its churn is quietly accelerating. If new acquisition is growing faster than churn, the top line still looks healthy. But the company is running harder just to stand still — and when acquisition slows (it always does eventually), the underlying retention problem becomes visible. Growth accounting forces the diagnosis early, when it's cheaper to fix.

The growth rate vs. absolute growth distinction matters just as much. A product growing from 100 to 200 users is growing 100% — but that's a different problem than growing from 100,000 to 110,000 at 10%. Absolute growth matters for things like infrastructure, support load, and revenue. Relative growth rate matters for understanding trajectory and whether you're accelerating or decelerating. Reporting one without the other creates blind spots in both directions.

The deeper shift the framework demands is a move from measuring inputs to measuring outputs. Marketing spend is an input. CAC is an output. Impressions are an input. Qualified pipeline is an output. Campaigns that optimize for input metrics (spend, clicks, impressions) without connecting to output metrics (pipeline, closed revenue, net new retained users) produce vanity — you can always buy more impressions; you can't always buy more customers at a cost that makes sense. Building a growth accounting practice means deciding, upfront, which outputs the business actually cares about — and then reverse-engineering which inputs drive them.

Related glossary: growth accounting, churn, CAC, reactivation

§2 Funnel Analytics — Measuring the Conversion Journey Building

The acquisition funnel is the oldest model in marketing, and it's still the right starting point — not because it's complete, but because it forces you to put a number on each stage of the customer journey and find where you're losing people. The five-stage model (Awareness → Interest → Consideration → Intent → Conversion) maps to measurable events: impressions or reach at the top, signups or trial starts near the bottom, paid conversions at the close.

Funnel drop-off analysis answers the most important question in acquisition: where are you losing the most? A company that converts 10,000 site visitors to 2,000 signups (20% step conversion) but only 400 to trials (20% again) and 80 to paid (20% again) has a different problem than a company that converts 10,000 to 500 signups (5%) but 400 to trials (80%) and 300 to paid (75%). Both end at roughly the same place — but one needs to fix top-of-funnel awareness, and the other needs to fix its landing page or value proposition messaging. Drop-off analysis tells you where to put the next dollar of effort.

The top-of-funnel vs. bottom-of-funnel distinction matters for how you measure. Top-of-funnel metrics — impressions, reach, click-through rate — are leading indicators that tell you about brand awareness and message resonance. They're proxies. Bottom-of-funnel metrics — trial starts, demo requests, closed deals — are the actual outcomes. A common error is reporting top-of-funnel metrics as if they're the goal; they're not. They're only useful insofar as they predict bottom-of-funnel outcomes, and that relationship needs to be empirically validated, not assumed.

Micro-conversions (email captured, content downloaded, webinar attended) vs. macro-conversions (paid subscription, contract signed) is a distinction worth maintaining explicitly in your tracking plan. Micro-conversions are signals of intent; macro-conversions are the actual goal. Optimizing micro-conversions at the expense of macro-conversions is a real failure mode — you can improve email capture rates dramatically while the quality of those leads (and their eventual conversion to paid) declines. Always model the full funnel, not just the stage you're currently working on.

Cohort-level vs. aggregate funnel analysis is the last nuance most teams miss early. Aggregate funnel metrics tell you the average across all users. Cohort-level funnel analysis tracks users acquired in the same period through the funnel together — which lets you see whether conversion rates are improving or declining over time, and which acquisition sources produce the highest-quality leads.

Related glossary: acquisition funnel, conversion rate, micro-conversion, top-of-funnel

§3 Attribution Modeling — Figuring Out What Drove the Conversion Building

The attribution problem is simple to state and genuinely hard to solve: a customer touched your paid search ad, opened three emails, read a blog post, saw a retargeting banner, visited your site directly, and then got a call from a sales rep — and then bought. Which of those seven touchpoints caused the conversion? The answer shapes where you invest next, so getting it wrong has real compounding consequences.

The standard attribution models represent different answers to that question. First-touch gives 100% credit to the first interaction — useful for understanding what creates awareness, but ignores everything that closed the deal. Last-touch gives 100% credit to the last interaction before conversion — overstates the value of branded search and direct, which often capture demand already created elsewhere. Linear splits credit evenly across all touchpoints — defensible but mathematically arbitrary. Time-decay weights touchpoints closer to conversion more heavily — makes intuitive sense for short cycles, less so for long enterprise deals where an early discovery call may have been the real decision-maker. Data-driven (or algorithmic) attribution uses historical conversion patterns to assign credit based on actual predictive lift — the most accurate when you have enough data, but a black box that's hard to explain to stakeholders.

B2B attribution is structurally harder than B2C for two reasons. First, the sales cycle is longer — weeks or months, during which a prospect might be touched by dozens of campaigns across multiple channels. Second, there are multiple stakeholders: the economic buyer, the champion, the technical evaluator, the legal reviewer. Any of them might have a touchpoint that matters. Standard person-level attribution models assume a single buyer journey; B2B reality is a committee.

Multi-touch attribution and media mix modeling (MMM) are complementary, not competitive. Multi-touch attribution is bottom-up — it traces individual customer journeys. MMM is top-down — it uses aggregate spend and revenue data to estimate channel contribution via regression. MMM handles offline channels and dark social better; multi-touch attribution handles individual journey granularity better. Sophisticated growth teams use both.

The limits of attribution are real and worth acknowledging upfront. Dark social — content shared in private channels (Slack, WhatsApp, email forwards) — is completely invisible to tracking. Word of mouth leaves no digital fingerprint. Brand advertising has long, diffuse effects that no person-level model captures. The honest position is that attribution models are approximations that help allocate marginal spend, not ground truth about what caused a conversion.

§4 Cohort Analysis — Understanding Retention by Acquisition Vintage Building

A cohort is a group of users who share a defining characteristic — typically, the period in which they were acquired. Cohort analysis tracks what happens to those users over time, together, so you can see how behavior evolves without the noise of new users constantly mixing in. It's the closest thing growth analytics has to a controlled experiment when you don't have one.

The reason cohorts reveal what averages hide is product improvement. If your product was bad in January and significantly better by March, your average Day-30 retention across all users conflates those two realities. Your aggregate retention curve might look flat or even declining — masking the fact that newer cohorts are retaining at twice the rate of older ones. Cohort analysis separates those signals. Conversely, it also catches the failure mode where a new acquisition channel looks great on volume but brings in low-quality users who churn faster — a problem invisible in aggregate until it becomes a revenue crisis.

Retention curve shape is diagnostic. A cliff-churn pattern (most users leave in the first 1-2 periods, then stabilize) usually indicates an onboarding problem or a mismatch between what acquisition promised and what the product delivers. Gradual churn (linear decay over time) often indicates low switching costs or competitive erosion. Negative churn — the cohort is worth more over time because of expansion revenue — is the holy grail for SaaS businesses; it means existing customers more than offset any losses. The shape tells you where to look.

DAU/WAU/MAU ratios — daily active users divided by monthly active users — measure engagement depth, not just presence. A DAU/MAU of 50% means the average user uses the product every other day. A DAU/MAU of 10% means they use it roughly three times a month. Whether 10% is good or terrible depends entirely on the product's intended use case: a tax-filing app should have low DAU/MAU; a messaging app with low DAU/MAU is in trouble. The ratio is only meaningful in context, and the right context is "what's the natural use cadence of this product?"

LTV curves by cohort extend the retention analysis to revenue. If newer cohorts are retaining better but at lower price points (because of a freemium expansion, say), the retention improvement might not translate to LTV improvement. Track both together to get the full picture of acquisition quality.

Related glossary: cohort, retention curve, DAU/WAU/MAU, LTV, negative churn

§5 CAC, LTV, and the Growth Efficiency Metrics Building

CAC — customer acquisition cost — is total sales and marketing spend divided by the number of new customers acquired in the same period. That's the formula; the discipline is in what goes into "sales and marketing spend." Full-loaded CAC includes not just ad spend but sales team compensation, marketing team compensation, tools, events, and agency fees. Blended CAC includes all channels; paid CAC isolates only paid acquisition. Companies that report "CAC" without specifying which definition are usually reporting the number that looks best, not the number that's most useful.

LTV — lifetime value — is the net revenue a customer generates over their entire relationship with the business. The simple formula for SaaS: LTV = ARPU × gross margin / monthly churn rate. If average revenue per user is $100/month, gross margin is 70%, and monthly churn is 2%, LTV is $3,500. Note what this formula actually captures: gross margin (not revenue), because you're measuring economic value to the business, not just revenue; and churn rate as the denominator, which means a business with half the churn rate has double the LTV even at identical revenue and margins.

The LTV:CAC ratio is the canonical growth efficiency metric. A ratio above 3:1 is generally considered healthy — you're getting $3 of lifetime value for every $1 spent to acquire. Above 5:1 often signals under-investment in growth: you could be spending more aggressively to acquire more customers and still be economic. Below 1:1 means you're destroying value with every acquisition. The ratio should be interpreted in context: early-stage companies often run below 3 deliberately while they refine the engine; mature companies with strong retention should sit higher.

CAC payback period — the number of months to recover CAC from gross margin — is often more operationally useful than LTV:CAC because it doesn't require a long-range LTV estimate. If CAC is $1,200 and monthly gross margin contribution is $70, payback is 17 months. Below 12 months is generally considered strong for SMB SaaS; enterprise with longer sales cycles tolerates longer payback periods because of lower churn. The magic number (net new ARR × gross margin / sales and marketing spend in the prior period) captures overall GTM efficiency at the company level — above 0.75 is healthy, above 1.0 is exceptional.

Related glossary: CAC, LTV, LTV:CAC ratio, CAC payback period, magic number

§6 Growth Loops — The Mechanics of Compounding Growth Strategic

The funnel model is useful, and it's also incomplete in a specific way: it's linear. A user enters the top and exits at the bottom. The funnel doesn't model what happens next — whether that user creates value that attracts more users. Growth loops model that feedback. They're compounding where funnels are additive.

The key distinction: in a funnel, you acquire 100 users and that generates 100 units of value. In a loop, acquiring 100 users generates activity that attracts 40 more users, who generate activity that attracts 16 more, and so on. The loop multiplier — the fraction of users each cohort produces organically — determines how much of your growth is self-funded versus requiring continued investment. A loop with a multiplier of 0.4 means 40% of your growth comes for free from the previous period's users. A multiplier above 1 means growth is truly viral and will accelerate without additional investment (extremely rare; Slack and WhatsApp early growth approached this).

There are four canonical loop types. Viral loops work when users invite users — either through sharing features, network effects, or referral programs. The product is intrinsically better with more people in it (communication tools, marketplaces). Content loops work when users generate content that attracts new users through search or social — Yelp, Reddit, Stack Overflow, TikTok. Product-led loops work when product usage creates outcomes that pull in new users — Slack spreading team-by-team, Figma spreading designer-by-designer, Dropbox spreading through shared files. Paid loops close the circle when revenue funds acquisition that generates more revenue — only stable when LTV:CAC is high enough.

Identifying the loop native to your product is more diagnostic work than strategy. Ask: when your happiest users use your product, what do they produce that's visible outside the product? If the answer is nothing — the product is purely private — you don't have a native viral or content loop, and you'll need to engineer one deliberately (through referral mechanics, public profiles, share features) or rely on paid or sales loops. Neither is wrong, but the cost structure and defensibility are very different.

Related glossary: growth loop, viral coefficient, product-led growth, network effects

§7 Experimentation and Channel Strategy at Scale Strategic

A growth experimentation function is not a testing culture. A testing culture runs A/B tests on button colors. A growth experimentation function has a hypothesis backlog, a prioritization framework, a consistent statistical methodology, and a process for translating findings into scaled changes. The difference is discipline: most companies can run an experiment; few can run a program.

The experiment lifecycle is straightforward: hypothesis → design → test → learn → scale. The failure modes are in the details. Hypotheses without a stated mechanism ("we think X will improve Y because Z") are almost always wrong in ways that teach nothing. Designs without sufficient sample size and runtime produce false positives. Tests measured on the wrong metric (CTR instead of downstream conversion) optimize the wrong thing. The learn step — which most companies skip or rush — is where the hypothesis is updated, not just confirmed or rejected. And the scale step requires explicit criteria: at what effect size and confidence level does this become a default?

ICE scoring (Impact × Confidence × Ease) is a lightweight framework for prioritizing channel experiments. Score each axis 1-10, multiply. Impact captures how large the upside is if the experiment works. Confidence captures how certain you are the experiment will work, based on analogous evidence. Ease captures the effort and cost to run the test. ICE doesn't produce objectively correct prioritization, but it forces explicit scoring of the variables that actually determine priority — and it creates a defensible record of why you ran what you ran.

Every channel has an S-curve: early adopters, rapid growth, saturation. The channels that drove your first 1,000 customers rarely drive your first 100,000 — not because they stopped working, but because you saturated them. Content SEO saturates when you've covered the high-value keyword clusters. Paid social saturates when you've exhausted the addressable audience. The strategic challenge is identifying when a channel is approaching saturation and beginning to test the next one before you need it. Companies that wait until a channel has demonstrably saturated before exploring the next one are always late.

The most important principle governing growth investment is sequencing relative to product-market fit. Before PMF, spend on growth is mostly waste — you're filling a leaky bucket. After PMF, the constraint on growth is capital and execution, not product. The signal that PMF exists strong enough to justify accelerated growth investment: organic retention is healthy without heroic intervention, customers can explain why they use the product in a way that predicts who else will buy, and the product spreads within customer organizations or networks without explicit push.

Open this guide in BizTech Primer →