PQL scoring for PLG: 12 signals that actually predict conversion
Most product-qualified lead scores are demographic models wearing a product-usage costume. The signals below are the ones that repeatedly survive validation across self-serve products.
Marcus Lee · Jul 28, 2026 · 6 min read
Signals that predict
The common thread is investment that is costly to fake and expensive to abandon. A user who has imported their own data, invited a colleague, and connected a system of record has made three decisions that are annoying to reverse.
- Real data imported, rather than sample data explored.
- A second user from the same email domain, active in the same workspace.
- An integration connected to a system the company already depends on.
- A scheduled or recurring action configured, not just run once.
- Return visits on three separate days within the first week.
- Depth: reaching a feature that only makes sense after the core workflow is understood.
- An export, share link, or report sent to someone outside the workspace.
- A billing page visit that is not immediately followed by a churn signal.
- A support question about limits or pricing tiers rather than about how to start.
- Workspace naming and settings customised away from defaults.
- Volume crossing a threshold that makes the free tier genuinely inconvenient.
- A second workspace or project created.
Signals to stop weighting
Total session count and total time-in-app correlate with confusion as often as with intent. Company size and job title are useful for routing and useless for timing. Email opens on onboarding sequences measure your subject lines, not the account.
Feature-flag exposure is the subtlest offender: if you score on features you selectively expose, your model will faithfully learn your rollout plan and tell you nothing about the user.
Scoring hygiene
Score at the account level, not the user level; in a PLG motion the buying unit is almost never one person. Decay scores over time so an account that was hot in March does not stay in the queue in July. And validate against closed-won, never against sales-accepted, or you are measuring how your own team triages.