Most website visitor intent scoring dies quietly. A number appears in the CRM, a rep opens the account, sees a red badge and no explanation, and two weeks later the field is ignored. The score was probably not wrong. It was unexplainable, so nobody could defend it in a pipeline call.
The fix is not a bigger model. It is a score built from three inputs your team can already argue about: which pages an account viewed, how recently, and how often across distinct sessions. Decompose the score into those parts and it becomes something a rep can check, a manager can tune and a forecast can lean on.
This guide shows how to build that score inside Prynt Reveal: how to weight pages, decay recency, count frequency with bots removed, and ship the evidence next to the number.
Why reps ignore intent scores
Most scores fail for one of three reasons. They arrive as a single opaque number with no explanation, so a rep cannot tell whether 80 means anything. Their inputs drift: page rules change, thresholds move, nobody wrote down what the score means. Or the data underneath is dirty, and refresh loops and crawlers make a status page look like a hot account.
Trust comes from the opposite: a score you can decompose, definitions that stay written down, and identification you can audit. Reveal names a company only when the visit comes from that company’s own network, with a confidence level and the evidence behind it; home broadband, mobile carriers, hosting and VPN traffic stay anonymous, as our accuracy page explains.
The three inputs that survive scrutiny
Pages: weight the page, not the pageview
Start with a short list of page rules, not a hundred. Pricing, product, integration docs, security and compliance pages carry buying intent. Blog posts and glossary entries carry research intent. Careers pages carry almost none.
| Page type | Weight |
|---|---|
| Pricing | high |
| Product and features | high |
| Docs and integrations | medium-high |
| Security and compliance | medium-high |
| Blog and glossary | low |
| Careers | near zero |
Cap the contribution per session: ten pricing pageviews in one afternoon is one signal, not ten.
Recency: decay, don’t cut off
A hard 30-day window creates a cliff. Use a half-life instead: pick a decay period (14 days is a reasonable start) and score each visit as weight × 0.5^(days_since_visit / 14). The score fades the way interest fades, and every contribution has a formula behind it.
Frequency: distinct sessions, bots excluded
Count distinct sessions per day with a daily cap, and distinct days visited in the window. One session with eight pageviews is curiosity; four sessions across four days is a buying process. Remove non-human traffic first: Prynt Identify combines fingerprinting, bot and AI-agent detection and VPN/proxy signals, so crawlers and polling scripts do not inflate the count. Our platform overview shows how the products share one browser agent, one account and one bill.
Turn the inputs into one explainable score
Pick weights you are willing to defend in a pipeline call, then write them down. A workable start is pages 45, frequency 30, recency 25, each normalised to 0–100 first.
Account A, worked:
- Pages: pricing and the integration doc — 40 page points, capped.
- Frequency: 3 sessions on 3 different days in the last 7 — 30 points.
- Recency: last visit 2 days ago at a 14-day half-life — 25 × 0.91 ≈ 23.
- Total: 93, flagged high, with the three lines above attached.
That is the whole trick. The number summarises; the three lines argue. A rep who disagrees can point at a line, not at a model.
Keep ICP fit as a separate field. A perfect-fit account reading a blog post is not in-market today, and blending the two hides both facts.
Ship the evidence with the score
A score that lives in a dashboard changes nothing. Send it where the rep works — Slack, HubSpot, Salesforce, Pipedrive or a signed webhook — with hot pages, last seen, session count and ICP fit in the same payload.
Then automate the boring part. A rule-based playbook can read: ICP fit high, pricing page viewed, three sessions in seven days → create a task with the evidence and draft an outreach note for a human to review. Nothing sends itself: Reveal drafts, a person reviews and sends. People are attached only from first-party data with a consent record, and opted-out accounts are suppressed, as described in our Reveal privacy notice.
Review the score, don’t worship it
Once a week, pull your ten highest-scoring accounts and ten mid-scoring ones. Ask whether the evidence matches the score and whether reps acted on it. If high scores never convert, the page weights are usually wrong — blog pages counting as buying intent is the classic case. If reps still ignore the list, the evidence is not visible where they work.
Re-tune monthly and keep a changelog of weight changes.
FAQ
What is website visitor intent scoring?
It is a per-account number that estimates how close a visiting company is to buying, built from observable behaviour on your site — which pages, how recently, how often — with the evidence shown next to the score.
Should intent scores include ICP fit?
Keep them separate. ICP fit says whether an account could buy; intent says whether it is active now. Two fields let a rep prioritise the intersection instead of guessing what a blended number meant.
Can you score anonymous visitors?
No, and you should be suspicious of anyone who says otherwise. Reveal identifies a company only when the visit comes from that company’s own network; people are linked only from first-party data such as forms, logins and CRM imports, with a consent record.
Ready to see it on your own traffic? Turn on Reveal on one site, let a week of identified visits accumulate, then set your three weights and backtest them against deals you already closed. The Free plan covers 100 companies a month on one site with 7-day history — enough to check whether your page weights match how your buyers actually behave. Create your account at pryntid.com.
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.