Distribution Channels

What are the main channel archetypes?#

Real systems are usually hybrids, but these archetypes clarify who performs which function.

How do you choose and test a channel?#

1. Define the channel–customer–job unit#

Do not evaluate "paid search" or "partners" across the whole company. State the unit:

For this customer segment, in this buying situation, for this offer and price, this channel is expected to create this path to value.

For example: US accounting firms with 10–50 employees that have outgrown email-based document collection will discover the workflow through practitioner search, start a guided trial, buy a $300 monthly plan without a sales call, and activate after completing one client intake. Use Ideal Customer Profile, Customer Segments, and Jobs to Be Done to keep the unit precise.

2. Map the entire route, including handoffs#

For each step, name the owner, evidence, elapsed time, cost, and failure mode.

A channel that generates leads but fails at procurement, implementation, or renewal is not suffering a "sales problem." The route is incomplete.

3. Write a channel contract sheet#

Before signing an intermediary, document target accounts and territories; exclusivity and conflicts; lead ownership, deal registration, and attribution rules; price authority, discount limits, and rebates; marketplace or reseller economics; invoicing, refunds, chargebacks, taxes, and cash timing; customer-data access and permitted communications; implementation, support, and service-level responsibilities; use of trademarks, claims, and case studies; renewal and termination rights; and post-termination access to customers, data, and integrations.

Accounting can also change. Under IFRS 15's principal-versus-agent framework, whether you recognize gross customer consideration or a net fee depends on whether you control the specified good or service before transfer. Involve a qualified accountant before presenting GMV, billings, or gross revenue as company revenue.

4. Define one falsifiable hypothesis and guardrails#

A useful channel hypothesis states the segment and offer, the channel and mechanism, expected stage conversions and lag, minimum activation and retention quality, maximum spend and downside, the decision date, and the rule for continue, revise, or stop.

Example: A $12,000 referral-partner test will produce at least 12 paid target accounts within 90 days, at least 75% will activate within 14 days, and projected contribution-aware payback will be under six months. No partner will exceed 20% of the test pipeline.

Isolate the uncertain mechanism. Google's Ads experiment guidance recommends a clear business-linked hypothesis, split budget between base and experiment, stable settings, and enough time and volume to interpret results. Changing audience, offer, price, sales process, and channel at once produces an uninterpretable result.

5. Instrument funnel, economics, quality, and capacity together#

Funnel: eligible reach, engagement, qualification, evaluation, paid conversion, activation, renewal, expansion — reported as conditional rates so a strong top-of-funnel number cannot hide a broken downstream step.

Economics: fully loaded CAC, channel fees, discounts, onboarding and support, contribution margin, payback, refunds, bad debt, cash lag.

Quality: time to first value, retained behavior, logo and revenue retention, support burden, referenceability, customer fit.

Capacity and risk: spend available at the current marginal return, seller or partner throughput, inventory or implementation capacity, platform dependence, concentration.

Match acquisition cost and customer timing, count paying customers rather than leads, and segment CAC by channel, customer, or motion. Acquisition spend in one period may convert in another.

6. Read attribution as a model, not ground truth#

A customer may hear about you from a peer, read content, attend an event, click a paid ad, receive outbound, and transact through a marketplace. Source fields cannot make each touch independently causal. Google Analytics describes attribution as a rule or algorithm for assigning credit: last-click assigns all credit to the final eligible channel, while data-driven models estimate contributions from observed converting and non-converting paths. Even a sophisticated output remains model-dependent.

Use holdouts, geographic tests, randomized splits, or matched cohorts when material spending decisions require incremental evidence. At minimum, compare models and publish the rule used.

7. Scale marginally, then diversify deliberately#

Average CAC from the first customers is not the price of the next customer.

Incremental CAC = change in channel acquisition cost ÷ change in new customers

Increase spend or capacity in bounded steps, re-estimating conversion, retention, contribution, and lag after each step. A channel can saturate because the highest-intent audience is exhausted, media prices rise, partner attention shifts, or operational capacity binds. Diversification should reduce a named dependency or reach a distinct customer — not create a long list of underpowered experiments.

What are the core channel formulas?#

Conditional funnel conversion

Stage conversion = entities completing the next stage ÷ entities eligible at the current stage

End-to-end conversion = product of conditional stage rates

Fully loaded channel CAC

Channel CAC = channel-attributable acquisition cost ÷ new paying customers acquired through the channel

Include media, content, events, sales labor and commissions, partner enablement, tools, agencies, and reasonable shared-cost allocations. Publish the policy — a CAC number is not comparable if one channel includes seller compensation and another includes only media spend.

Contribution per active customer

Monthly channel contribution = monthly revenue × gross margin rate − channel fee or revenue share − other variable service cost

If gross margin already includes a cost, do not subtract it twice.

Contribution-aware payback

Payback months = upfront acquisition and activation cost per customer ÷ monthly contribution per active customer

For uneven usage, churn, implementation, or annual prepayment, use cumulative cohort cash instead: payback occurs at the first month where cumulative contribution ≥ cumulative acquisition and activation cost.

Allowable CAC

Allowable CAC = target payback months × expected monthly contribution − upfront activation cost

This converts a cash constraint into a spending guardrail. It is not permission to spend up to estimated lifetime value; early-stage LTV is usually too sensitive to immature retention.

Channel concentration

Channel concentration = revenue, customers, pipeline, or receivables linked to the largest channel partner ÷ total for that measure

Measure several denominators. As GitLab's disclosure shows, a company can have no end customer above 10% of receivables while two channel partners hold 18% and 12%.

Worked example: paid self-serve versus referral partners#

A B2B workflow startup runs two 90-day tests for the same target segment. Observed test data is separated from the assumptions used for a 12-month projection.

Step 1 — conversion. Paid self-serve: 144 ÷ 1,200 = 12.0% visit-to-trial, 36 ÷ 144 = 25.0% trial-to-paid, 36 ÷ 1,200 = 3.0% visit-to-paid. Partners: 18 ÷ 45 = 40.0% opportunity-to-paid. The 40% partner rate is not comparable with 3% visit-to-paid — the denominators enter at different levels of intent.

Step 2 — CAC and contribution.

Paid CAC = $24,000 ÷ 36 = $666.67 · Monthly contribution = $300 × 82% = $246

Partner upfront cost = $16,500 ÷ 18 = $916.67 · Monthly contribution = $750 × (78% − 20%) = $435

Step 3 — contribution-aware payback.

Paid = ($666.67 + $200) ÷ $246 = 3.5 months

Partner = ($916.67 + $600) ÷ $435 = 3.5 months

Headline CAC makes paid self-serve look far cheaper. After revenue, margin, channel share, and onboarding are included, the two routes have almost identical projected payback.

Step 4 — first-year cohort contribution. With constant monthly churn, expected active customer-months in the first 12 months are (1 − (1 − churn)^12) ÷ churn:

Paid = (1 − 0.96^12) ÷ 0.04 = 9.68 months · Partner = (1 − 0.985^12) ÷ 0.015 = 11.06 months

Paid cohort = 36 × $246 × 9.68 − $24,000 − (36 × $200) = $54,546

Partner cohort = 18 × $435 × 11.06 − $16,500 − (18 × $600) = $59,283

The partner cohort produces slightly more projected first-year contribution despite half as many customers — conditional on the churn assumptions, partner-reported opportunity quality, consistent implementation, and the availability of more partners with similar performance.

Step 5 — test the next dollar, not the old average. Suppose paid spend doubles from $24,000 to $48,000 but paid customers rise only from 36 to 54:

Average CAC at scale = $48,000 ÷ 54 = $888.89

Incremental CAC = $24,000 ÷ 18 = $1,333.33

Incremental payback = ($1,333.33 + $200) ÷ $246 = 6.2 months

The blended result still looks respectable, but the next tranche is far less efficient. The decision should use the 6.2-month marginal payback and roughly $15,273 of projected incremental first-year contribution, not the original $666.67 CAC.

Caveats. Eighteen customers cannot distinguish a 40% conversion rate from a 28% one. The constant-churn shortcut breaks when churn varies by tenure, contracts renew annually, or expansion is material — use an actual cohort schedule then. And the 78–82% margin assumption quietly holds onboarding and support labor inside the margin, which is exactly what breaks first when a founder-led motion scales.

What to conclude. Not that one channel wins universally. Paid self-serve offers more direct learning and volume but is already saturating. Partners produce larger, stickier accounts in this small test but create enablement, concentration, and attribution risk. The next experiment should isolate the largest uncertainty: whether another partner cohort reproduces the quality, and whether a narrower paid audience restores marginal efficiency.

What are the common mistakes?#

• Treating a source label as a channel. "Organic," "direct," and "partner" are analytics buckets, not operating definitions. Specify entry, ownership, transaction, delivery, and renewal.
• Choosing the lowest attributed CAC. Low CAC can bring low-revenue, low-retention, high-support customers. Compare contribution-aware payback and mature cohorts, not acquisition cost alone.
• Ignoring lag, then treating attribution as causality. This month's spend may create next quarter's revenue, so same-month division makes a growing channel look inefficient. And last-click credit overvalues the closing touch while self-reported attribution overvalues memorable ones. Use cohort matching and run experiments where the decision is material.
• Double counting sourced, influenced, and transacted revenue. A partner can introduce an account, a seller can close it, and a marketplace can process the order. Define mutually exclusive sourced revenue and keep influenced and transacted as separate measures.
• Assuming partners are free sales capacity. Partners require recruitment, training, certification, account mapping, content, technical support, and economic motivation. Track active producing partners, not signed partners — and set territory, lead-ownership, and discount rules before direct sellers and resellers collide over the same deal.
• Presenting gross transaction value as revenue. When another party provides the good or service, accounting may require net commission revenue. GMV, billings, cash collected, and recognized revenue are different metrics.

When does channel analysis break?#

• Long enterprise cycles make early CAC unstable. A six- or twelve-month cycle leaves few closed cohorts, and one deal can dominate. Use pipeline aging, buying-committee progress, and pilot outcomes as leading evidence — labeled as such — rather than manufacturing a precise CAC from an immature denominator.
• Power-law channels violate average assumptions. One reseller, creator, or marketplace listing may produce most outcomes. An average across partners can describe no actual partner. Show the distribution, the top concentration, and whether recruiting another high performer is repeatable.
• Product-led loops are not free. Invites, shared artifacts, templates, and open source can distribute a product, but engineering, moderation, support, abuse prevention, and free usage still cost money. Measure the incremental customer and contribution the loop produces.
• Physical goods and high variable-cost products break software-style payback. CAC and gross margin omit working capital, stockouts, markdowns, returns, warranty, and retailer payment terms. AI, payments, logistics, and managed services can generate revenue while contribution deteriorates. See Managed Services, Hardware-as-a-Service, and API-as-a-Product.
• Platforms can change the bargain. Marketplaces and app stores improve discovery while controlling ranking, fees, data, and access. Treat platform rules as an external dependency and keep a diversification plan. See Marketplace Business Model and Two-Sided Markets.
• A channel can outlive its positioning. The route may keep producing customers after the product, buyer, or value proposition changes, mixing old and new positions in historical conversion. Re-baseline the channel–customer–job unit whenever positioning, price, packaging, or segment changes.

Founder checklist#

Before increasing fixed spend or signing an exclusive channel agreement:

• Which exact customer and buying situation does this route serve?
• What does the channel contribute: access, trust, qualification, transaction, delivery, or renewal?
• Which stages and costs are observed, and which are still assumptions?
• Are activation and retention at least as strong as the comparison cohort?
• What is fully loaded CAC, monthly contribution, cash lag, and payback — and what happens to marginal CAC as spend or headcount rises?
• Who owns price, discounting, customer data, support, renewal, and expansion?
• Are sourced, influenced, and transacted revenue defined without double counting?
• What is the largest concentration or platform dependency, and what contractual, accounting, privacy, or advertising obligations apply?
• What result will cause the team to continue, revise, or stop?