Product-Led Growth

How do you build a product-led funnel? A six-step framework#

1. Define the eligible account and the entry path#

State who can reach value without bespoke data, security review, integration work, or professional services. That set is your eligible population.

Do not use aggregate signups as the addressable funnel. If most signups cannot become successful customers, every rate you compute downstream is diluted by an audience you never wanted, and improvements to the product will look like they did nothing. Use the Ideal Customer Profile to define eligibility before you define conversion.

2. Define activation behaviourally#

An activation event is the earliest observable behaviour strongly associated with later retention: a deployed project, a completed workflow, an invited teammate, an imported dataset, an accepted output. It is not account creation, and it is not a tutorial completion.

activation rate = activated eligible accounts ÷ eligible new accounts

time to value = median time from eligible entry to first verified value event

Validate the choice: take a past cohort, split it on the candidate event, and compare 90-day retention. If retention is similar on both sides, the event is a milestone, not an activation.

3. Define product-qualified signals — then test them#

A product-qualified lead (PQL) or account is one whose usage suggests both realised value and commercial potential: repeated value events, multiple active users, approaching a usage limit, requesting security controls, or adopting a high-value integration.

PQL rate = product-qualified accounts ÷ activated accounts

The definition must predict paid conversion or expansion in a holdout period. A score fitted to match past sales opinions is circular: it will reproduce the team's existing biases and look accurate doing it. Hold out a cohort, score it, then check what actually happened.

4. Measure the complete loop#

visitor-to-paid conversion = eligible signup rate × activation rate × activated-to-paid rate

Because these multiply, a funnel can degrade badly while every individual stage looks "fine." Track alongside it: retained activation, gross and net revenue retention, invite or template loops, support cost per account, and expansion by entry cohort. See Cohort Analysis.

5. Attach the economics — including the free users#

PLG CAC = (product-acquisition spend + free-user delivery cost + attributable lifecycle and sales-assist cost) ÷ new paid customers

CAC payback (months) = CAC ÷ monthly contribution per new customer

Free infrastructure and free support are acquisition costs for managerial analysis, even when accounting classifies them inside cost of revenue. Excluding them makes PLG look structurally cheaper than it is, and hides the exact failure mode — a free tier that grows faster than conversion — that kills PLG companies. Use contribution after variable delivery cost, not gross revenue; see Contribution Margin.

6. Route assistance deliberately#

Send a human only when expected incremental gross profit exceeds the cost and the delay of the assistance. Enterprise security review, data migration, multi-team rollout, and procurement usually justify it. A user who has been active for three days usually does not.

Worked example: a product-led funnel, and how it breaks#

Fictional company, one month of data.

Base case#

A workflow product receives 10,000 eligible signups. 55% complete setup; 40% of those reach the verified value event.

activated accounts = 10,000 × 55% × 40% = 2,200

18% of activated accounts convert to a $90 monthly plan:

new paid customers = 2,200 × 18% = 396

new MRR = 396 × $90 = $35,640

Spend: $120,000 on acquisition and lifecycle programmes, $45,000 serving free users, $15,000 on attributable sales assistance.

PLG CAC = ($120,000 + $45,000 + $15,000) ÷ 396 = $454.55

At 82% contribution margin:

monthly contribution per customer = $90 × 82% = $73.80

CAC payback = $454.55 ÷ $73.80 = 6.2 months

The volume trap#

A campaign doubles signups to 20,000 but brings lower-fit traffic. Setup falls to 35%, value-event completion to 30%; paid conversion holds at 18%:

20,000 × 35% × 30% × 18% = 378 paid customers

Twice the signups produced fewer customers. And free-user delivery cost roughly doubled, so CAC deteriorates on both sides of the ratio. Signups are an input; retained value is the result.

Where sales assistance wins#

Among activated accounts, 600 show enterprise signals. A sales-assisted route reaches 120 and wins 36 at $900 MRR. The assist programme costs $90,000.

assisted CAC = $90,000 ÷ 36 = $2,500

monthly contribution = $900 × 82% = $738

CAC payback = $2,500 ÷ $738 = 3.4 months

Higher-touch selling is economically superior for that segment — 3.4 months versus 6.2 — even though the company remains product-led. The product created the signal; the human captured the value.

What this example does not prove#

The payback figures assume the 82% contribution margin absorbs onboarding and support labour, which is the assumption that fails first as volume grows. They also ignore retention: a 6.2-month payback is meaningless if the median self-serve account churns at month four. Recompute payback with realised retention by cohort before treating either number as a plan. And the 600 "enterprise signals" are a scored population — if the score has not been validated on a holdout, the 36 wins may reflect selection rather than prediction.

What are the common mistakes?#

• Calling freemium PLG. A price tier does not create a value-and-conversion loop. The question is not whether the product is free; it is whether a user reaches provable value alone.
• Defining activation as signup. Registration proves intent to look, not receipt of value. Everything measured against a signup denominator is optimistic by construction.
• Optimising signup volume. Low-fit traffic consumes free-tier cost, dilutes every conversion rate, and makes a healthy product look broken.
• Shipping an unvalidated PQL score. If it has not predicted conversion in a holdout period, it is a ranking of the team's assumptions.
• Treating free-user cost as someone else's line item. Infrastructure and support for non-paying accounts are the price of the acquisition channel. Put them in CAC for decision-making, whatever the income statement says.

When does PLG break?#

• Value requires what a stranger cannot supply. Proprietary customer data, deep integration, irreversible change, professional judgement, field installation, or a lengthy security and regulatory review all block independent value. A sandbox may still support the sale, but it is not the purchase.
• The user and the economic buyer want different things. Users may love a tool while procurement, security, or management rejects it. Build admin controls, audit trails, and an organisational value case rather than treating usage as authority. See Land-and-Expand.
• The free tier outgrows the conversion path. If free cost per activated account rises faster than activated-to-paid conversion, growth is a widening loss. Check the ratio monthly, not annually.
• Small numbers. Early PLG funnels look precise and are not. A 40% activation rate from 50 eligible accounts is a range, not a rate. Publish denominators.
• Product signals become surveillance. Collect only necessary telemetry, explain its use, protect customer data, and do not trigger outreach from sensitive behaviour. A more precise funnel is not worth the trust it costs — and in regulated segments it can be a compliance problem, not a taste problem.

Above all: PLG cannot manufacture pull. If eligible users reach the value event and still do not convert or retain, the problem is the value, the segment, or the price — not the onboarding flow. See Product-Market Fit.