03 · Product-Led Growth at Scale¶
Product-led growth is easy to describe and hard to sustain. A product sells itself, users onboard without a human, and the sales team — if there is one — joins conversations the product has already started. It works beautifully until roughly $20M ARR, at which point three things happen at once: the cheap acquisition channels saturate, the largest customers start asking for things a self-serve product cannot do, and someone proposes hiring a sales team, which sets off a year-long argument about whether the company is "still PLG".
That argument is a category error. At scale, PLG is not a business model you either have or abandon. It is a claim about where qualification happens: in the product, from usage, rather than in a conversation, from stated intent. Everything else — sales, contracts, enterprise features — can be layered on top without contradiction, provided the layering is driven by what the product observed.
This module covers the PQL model, the arithmetic that decides who gets a human, packaging at scale, and the specific ways PLG companies break as they grow.
Three motions, and the hybrid nobody names properly¶
| Product-led | Sales-led | Hybrid (PLG + assist) | |
|---|---|---|---|
| Qualification | Product usage | Conversation | Usage triggers the conversation |
| First value | Minutes, self-serve | After a demo and a pilot | Self-serve, then assisted expansion |
| CAC | Low, mostly fixed | High, mostly variable | Mixed — the discipline is knowing which accounts get the variable cost |
| Deal size | Low | High | Bimodal, which breaks averages |
| Sales role | None, or expansion only | Owns the whole cycle | Joins where the product signals value |
| Fails by | Ceiling on deal size | Cost per deal | Touching everyone, or nobody |
ListUp is the third. The hard question is not whether to have salespeople. It is which accounts a salesperson should ever contact — and that is an arithmetic question with a defensible answer.
The product-qualified lead¶
A PQL is an account whose behaviour indicates readiness. Not a form fill, not a job title, not a firmographic guess — an observed pattern.
A workable PQL model has three parts, and most companies build only the first:
| Part | Question | Failure if missing |
|---|---|---|
| Fit | Do they look like a customer who succeeds here? | You chase accounts that will never expand |
| Intent / readiness | Have they done the things that precede buying? | You contact people at the wrong moment |
| Value at stake | How much is this account worth if it converts? | You spend the same effort on a $1,900 and a $14,000 outcome |
ListUp's PQL score, built by correlating each signal with 90-day expansion among 6,150 direct accounts:
| Signal | Part | Weight | Rationale |
|---|---|---|---|
| Catalogues managed ≥ 3 | Fit | 25 | Strongest single predictor of Console attach |
| SKUs under management ≥ 500 | Fit | 15 | Proxy for the pain being expensive |
| Channels connected ≥ 4 | Fit | 10 | Core product fit |
| Opened the Decide paywall ≥ 2× in 14 days | Intent | 20 | Direct expansion intent |
| Invited a second user | Intent | 15 | Multi-user accounts expand at 2.3× |
| Published ≥ 200 changes in 30 days | Readiness | 10 | Product is load-bearing in their operation |
| Support contact about limits | Readiness | 5 | Hitting a ceiling |
Scores above 55 are "qualified". But qualification is only half the decision.
Who gets a human: the arithmetic¶
An account executive fully loaded at $190,000 can run roughly 340 meaningful assisted sequences a year — so each touched account costs about $559. That cost is the same whether the account is worth $1,920 or $14,000, which is why a single PQL threshold across all accounts is always wrong.
Segment the qualified population by value at stake, then compare the conversion lift from assistance against the cost:
| Segment | Accounts | ACV uplift if converted | Assisted conv. | Self-serve conv. | Lift | EV per touch | ROI |
|---|---|---|---|---|---|---|---|
| A — agency-shaped (3+ catalogues, 500+ SKUs) | 430 | $14,000 | 14% | 5.0% | 9.0pt | $1,260 | 2.25× |
| B — large single seller (Publish-only, >2k SKUs) | 890 | $3,900 | 11% | 4.0% | 7.0pt | $273 | 0.49× |
| C — standard single seller | 2,740 | $1,920 | 9% | 3.5% | 5.5pt | $106 | 0.19× |
Touch segment A only:
| Line | Value |
|---|---|
| Accounts touched | 430 |
| AE capacity required | 1.26 AEs |
| Annual cost | $240,294 |
| Incremental ARR | 430 × 9.0pt × $14,000 = $541,800 |
| Net | +$301,506 (2.25× ROI) |
Touch everyone qualified (A + B + C):
| Line | Value |
|---|---|
| Accounts touched | 4,060 |
| Annual cost | $2,268,824 |
| Incremental ARR | $1,074,114 |
| Net | −$1,194,710 |
Assisting every qualified account destroys $1.19M a year while generating real, attributable, celebrated incremental revenue. This is how PLG companies quietly become unprofitable sales-led companies: every individual touch produces a win, and nobody computes the aggregate.
What B and C get instead: the product. An in-product expansion prompt at the moment the Decide paywall is hit costs about 5 engineer-weeks ($19,020) to build and roughly nothing to run:
| Prompt lift on the 3,630 B and C accounts | Incremental ARR |
|---|---|
| +1.0 point | $87,318 |
| +1.5 points | $130,977 |
| +2.0 points | $174,636 |
At a 1.5-point lift the prompt returns $130,977 a year for a one-off $19,020 — and unlike AE capacity, it does not need to be re-bought next year, and it scales to 10,000 accounts at the same cost. This is the core economic claim of PLG at scale, stated precisely: the product is a fixed cost and the sales team is a variable one, so people should be reserved for the segment where the variable cost clears its own bar.
Packaging and the free tier at scale¶
| Model | Mechanism | Best when | Risk at scale |
|---|---|---|---|
| Free trial (time) | Full product, 14–30 days | Value is obvious quickly | Users evaluate instead of using |
| Freemium (capacity) | Free forever below a limit | Network or data effects; low marginal cost | Serving free users becomes a real cost line |
| Reverse trial | Start on the paid tier, drop to free at day 14 | Premium value needs to be experienced | Perceived as a downgrade if handled clumsily |
| Free + paid seats | Viewers free, editors paid | Multiplayer products | Users game the role boundary |
| Usage-based with a free allowance | Pay above a threshold | Value scales with volume | Revenue is hard to forecast |
Two rules that matter more at scale than at launch:
- The free tier's job is to produce qualified accounts, not users. Measure it on PQLs generated per 1,000 free accounts and on cost to serve, not on signup count.
- The paywall must sit at the value boundary, not the cost boundary. Gating on what is expensive for you to serve produces a free tier that feels arbitrary; gating on what is valuable to them produces one that converts.
Metrics that matter at this stage¶
| Layer | Metric | ListUp | Healthy range |
|---|---|---|---|
| Acquisition | Visitors → signup | 4.1% (8,610/mo from 210,000) | 2–5% |
| Activation | Signup → activated | 52% (4,477/mo) | 30–60% |
| Monetisation | Activated → paid | 31% (1,388/mo) | 15–35% |
| Expansion | Net revenue retention, direct | 116.7% | 105–125% |
| Expansion | Multi-product attach | 47.8% of direct accounts | — |
| Efficiency | % of new ARR self-serve, no human | 71% | Above 60% for a PLG claim to be true |
| Efficiency | CAC payback, blended | 9.4 months | Under 12 |
| Assist quality | ROI on assisted touches | 2.25× (segment A only) | Above 2× |
The row that keeps you honest is % of new ARR closed with no human involvement. When that number drifts below about 50%, you are a sales-led company with a good trial experience, and you should say so — and change your hiring, forecasting and packaging to match — rather than defending a label.
How PLG breaks at scale¶
| Failure | What it looks like | Fix |
|---|---|---|
| Assist creep | Sales touches every qualified account because every touch "works" | Segment by value at stake; publish the ROI per segment quarterly |
| Two products in one | Self-serve UX degraded by enterprise controls | Separate the surfaces: admin and governance in a console, not in the main flow |
| Channel saturation | Organic and paid plateau; CAC rises quarter on quarter | Invest in loops (Level 3, Module 4), not in more channels |
| Enterprise gravity | The roadmap becomes the top 10 accounts' requests | Cap deal-driven work as a standing allocation (Level 3, Module 6) |
| Free tier cost drift | Cost to serve free users grows faster than conversion | Measure cost per free account; re-fence on value |
| Attribution wars | Sales claims self-serve expansions | Agree the rule before compensation is designed: no touch in 30 days = self-serve |
| Activation rot | New features raise activation friction one point at a time | Activation is a guardrail on every launch, not a growth-team metric |
That last one deserves its own sentence. Every team adding a step to onboarding is doing something individually reasonable, and the aggregate effect over two years is an onboarding flow nobody would have designed. Make activation a mandatory guardrail metric on every launch, at every group, and review the trend annually against a holdout.
How It Actually Works: the expected-value math behind who gets a human¶
EV per touch is a decomposition, not a lookup. The formula behind every row in the segment table is:
EV per touch = (assisted_conv - self_serve_conv) × ACV_uplift - cost_per_touch
ROI = (EV per touch + cost_per_touch) / cost_per_touch
= incremental_revenue_per_touch / cost_per_touch
Segment A: (0.14 - 0.05) × $14,000 = $1,260 incremental revenue per touch,
against a $559 cost, giving $1,260 / $559 = 2.25×. Segment C:
(0.09 - 0.035) × $1,920 = $105.60 against the same $559 cost, giving
0.19×. The lift (percentage points of conversion) barely differs between
segments A and C — 9.0pt vs 5.5pt — but ROI differs by more than 10x,
because ROI is driven by the product of lift and deal size, and deal
size varies by 7.3x across segments while lift only varies by 1.6x. This is
the concrete reason a single PQL score threshold fails at scale: score
alone captures readiness (the lift term) but says nothing about the value
term, and a low-value, high-readiness account can still have ROI below 1.0.
Why summing individual "wins" produces an aggregate loss. Every touched
account in segment C shows positive incremental ARR when measured in
isolation ($105.60 > $0 — some sales are attributable to the touch), so a
rep and their manager can honestly report a win. The aggregate math is
different because cost is linear in touches while the per-touch revenue is
fixed by segment: summing 4,060 × $559 = $2,268,824 in cost against
430×$1,260 + 890×$273×(dilution) + 2,740×$106×(dilution)-style blended
revenue nets negative once B and C's sub-$559 per-touch revenue drags the
average below cost. The fallacy is treating "this touch had positive
attributable revenue" (true, locally) as evidence that "this program has
positive ROI" (false, globally) — the correct test is always
marginal revenue per touch vs. marginal cost per touch, applied
segment-by-segment, never averaged across a population with 7x internal
variance in deal size.
The product-substitute comparison is a fixed-cost-vs-variable-cost
crossover. An AE's $559/touch is a marginal cost that recurs every time
(and every year) an account is touched. An in-product prompt's $19,020 is a
one-time fixed cost whose per-account marginal cost is ~$0. The crossover
point — the touch volume at which the AE's cumulative cost exceeds the
prompt's build cost — is $19,020 / $559 ≈ 34 touches. Any segment
expected to need more than 34 touches over the asset's lifetime is
structurally better served by a product mechanism than by headcount, which
is why B and C (3,630 accounts) go to a paywall prompt while A (430
accounts, each worth 25x more per conversion) goes to a human: the
crossover math and the ROI math point the same direction for different
reasons — cost amortization for the former, deal-size economics for the
latter.
Exercise¶
- Build your PQL model with all three parts — fit, intent, value at stake. Derive weights by correlating each signal with 90-day expansion, and report the correlation, not your intuition.
- Compute your cost per assisted touch: fully loaded AE cost divided by realistic annual sequence capacity. Get the capacity number from sales ops, not from a plan.
- Segment your qualified accounts by value at stake into three tiers, and for each estimate assisted and self-serve conversion. If you have no estimate, run a hold-out for one quarter before deciding.
- Compute EV per touch and ROI per segment. Draw the line where ROI crosses 1.0× and state how many accounts sit above it.
- Compute the aggregate of touching everyone qualified. If it is negative, take that number to your sales leader with the per-segment table, not as a criticism but as a capacity plan.
- Design the product substitute for the segments below the line, and compare its build cost against the AE cost it replaces, over three years.
- Audit your free tier: PQLs generated per 1,000 free accounts, cost to serve, and whether the paywall sits on a value boundary or a cost boundary.
- Measure the share of new ARR closed with no human involvement. If it is below 50%, write one page on what should change — the label, the packaging, or the motion.
- Check activation for rot: plot your activation rate over 24 months and list every onboarding step added in that period. Propose two to remove.