Most SaaS teams still rank leads by form fills and firmographics. That works while a landing page is all the evidence you have. It stops working once people can sign up, explore and leave without ever talking to you. Product usage lead scoring ranks the users and accounts already inside your product by how close they are to buying or expanding, using events you already record, and it keeps the evidence visible so a person can act on it.
The weighted sum is the easy part. The work is choosing inputs you can defend, showing the reason behind every number, and getting hot accounts in front of a human before interest cools.
Start with four input families, not twenty
Long signal lists feel thorough and produce scores nobody trusts. Pick four families and make each one concrete.
Recency
Measure the last meaningful event — a project created, a report exported, an integration connected — not the last page load. Apply decay: full points inside 48 hours, fewer at seven days, almost none past thirty.
Frequency
Count distinct active days in the last 7 and 30, then add depth: teammates invited, integrations connected, projects created. A user who returns four days running tells you more than one who reloads a dashboard forty times.
Key pages and key events
Pricing, the upgrade screen, billing settings, the docs page for a paid-only feature, the screen that says a limit has been reached. Record them as named events; they are the closest thing to a spoken question about price.
Stages
Signed up, email verified, first project, first invite, limit hit, trial day N, plan active. Stage supplies context: a day-2 user reading pricing is curious, a day-40 user reading pricing is shopping.
Turn the inputs into a score you can defend
Keep the weights in one file, cap each family so no single signal dominates, and normalise to 0–100.

| Signal | Example points |
|---|---|
| Meaningful event in the last 48 hours | +30 |
| Meaningful event in the last 7 days | +15 |
| Meaningful event in the last 30 days | +5 |
| 3+ distinct active days this week | +15 |
| Opened the upgrade or pricing screen | +20 |
| Invited a teammate | +15 |
| Hit a usage limit | +20 |
| No activity for 21+ days | −25 |
In pseudo-code:
score = 0
score += recency(last_seen_at) // 30 / 15 / 5 / 0
score += frequency(active_days_7d) // 15 / 8 / 0
score += key_events(events_7d) // up to 40, capped
score += stage(account_stage) // up to 15
score -= decay(days_dark) // up to 25
return clamp(score, 0, 100)
Two rules keep this honest. Cap the key-event family, so five pricing visits in one hour cannot outrank a week of real usage. And recalculate on events, not once a night, so the list reflects this morning.
Show the reason next to the score
A bare number invites two failures: your team ignores it, or trusts it blindly. Return the reasons with the score, each carrying an event name, a count and a timestamp.
{
"accountId": "acct_4821",
"score": 82,
"band": "hot",
"reasons": [
"+30 meaningful event in the last 48 hours",
"+20 hit the API rate limit 3 times in 5 days",
"+20 opened the upgrade screen twice",
"+15 invited 2 teammates",
"-3 no billing page visit"
]
}
The first line of an internal message now writes itself: hit the API rate limit three times this week, invited two teammates. That is a reason to talk, not a guess. Prynt Revenue AI builds scored opportunities this way, with the evidence attached to each one, so the queue is arguable instead of magical — see how Revenue AI scores opportunities.
Score the seat, read the account
Keep two objects: the individual (a visitor ID plus the user ID you know after login) and the account (plan, seats, stage). One hot seat in a cold account is a champion who needs ammunition; three hot seats in one account is a deal. Rank on first-party events and opaque IDs, never on form-field text or session contents. Prynt’s platform keeps these signals on one agent and one bill, and the docs cover the identity calls.

Work the hot list with a queue, not a blast
A score earns its keep only if it changes what someone does today. Send band changes to where a human already looks — a Slack channel, a CRM task, a webhook into your admin — with the top three reasons and a link to the account. Assign one owner and an expiry: a hot account nobody touches in 48 hours is history, not pipeline.
Automate the routing, not the relationship. When a score crosses the threshold, create the task and stop. Prynt Automate runs that workflow from an event or a decision, logs every run, and can also flag a high score that has gone quiet for two weeks.
Check the score against your own outcomes
Recalibrate monthly against reality, not opinion. Compare converters and non-converters to see which signals precede payment; Prynt Revenue AI runs that comparison with significance tests. Then check the inputs are clean — automated traffic inflates frequency and fakes a four-day streak. Prynt Analytics excludes bot sessions, and bot detection flags automated visits before they reach the funnel.

FAQ
Should I score the seat or the account?
Both. Seats show who is engaged; the account shows whether that engagement can become revenue. Rank seats inside accounts and read the two scores together.
How many signals should the first version use?
Four families is enough to ship. Add a signal only when you can name the decision it would change.
What score should count as hot?
Set the band from your own conversion history, not an industry number: take the top slice of your scored population, check how many converted, and move the line until the queue matches what your team can actually work.
Prynt Revenue AI turns first-party events into scored opportunities with the reasons attached, so your team works the top of the list first. Start a 14-day trial or explore the platform to see how Identify, Analytics and Automate feed the same score.
Try it free
Prynt is device intelligence with a free plan: 5,000 identifications a month, visitor IDs with confidence and bot detection — no credit card. Upgrade to Pro for VPN, proxy, tampering and behavioral signals.