LinkedIn activity signals are observable changes or interactions—not proof that a buyer is in market. Use them to update research, priority, or message context only after identity, source meaning, freshness, account fit, and permitted action pass a written acceptance contract.
The safest default is a research task. A public post, profile view, or role change may justify learning more; it does not automatically justify a pitch.
Know what LinkedIn actually surfaces
LinkedIn's Sales Navigator help documents alerts for saved leads and accounts including company updates, hiring and growth changes, role and job changes, lead shares, some profile and content engagement, funding, and senior hires. Availability varies by plan and setting, and many alerts depend on saving the lead or account.
Record the exact product label and source page rather than translating every event into “intent.” A saved account raised money, a lead shared a post, and a lead viewed a profile are different observables with different identity and privacy implications.
Capture the subscription and configuration context too. LinkedIn's documentation distinguishes alert availability across Sales Navigator editions and embedded experiences. A workflow tested with one edition cannot assume the same event coverage in another. Treat a missing alert as unknown coverage, not evidence that nothing happened.
Keep platform-derived events separate from facts a rep observed on a public page. Store whether the signal came from the alert feed, a saved-lead notification, an embedded CRM experience, or a rep's current profile view. That provenance determines how the event can be reproduced, explained, and corrected.
Classify activity without inventing intent
| Class | What it establishes | What it does not establish |
|---|---|---|
| Person change | Reported job or role update | New budget, buying authority, vendor interest |
| Company change | Reported funding, hiring, growth, news, or update | A project connected to your category |
| Public activity | A post, share, comment, or visible engagement occurred | Consent to private outreach or purchase intent |
| Relationship event | Connection accepted or known relationship changed | Permission for repeated unsolicited messages |
| Product-labeled buyer interest | The platform's documented event occurred | Individual identity or a purchasing decision unless explicitly provided |
Preserve LinkedIn's own label, timestamp, subscription context, and caveat. The B2B buying signals guide separates broader event types.
Use evidence tiers rather than one warmth score
Tier one is a directly observable event with a stable person or account match. Tier two adds independent account evidence relevant to a real workflow problem. Tier three adds a current relationship or explicit buyer action that permits a specific next step. Each tier changes what the rep may do; it does not claim a probability of purchase.
For example, a lead sharing an operations article is tier-one context. The same event plus a company announcement about rebuilding operations may justify research. A direct request or active conversation belongs to a different relationship state. Do not let multiple weak activities masquerade as one strong event merely because a score adds them together.
Build the signal acceptance contract
Before an event enters a rep queue, require: source URL or event ID; observed time; person and company IDs; match confidence; exact observable; freshness rule; fit rule; current relationship; owner; allowed action; suppression conditions; and correction path.
Reject or review namesakes, subsidiaries mapped to the wrong parent, former employees, stale profile data, topic ambiguity, disputed identities, customers, active opportunities owned by someone else, opt-outs, and repeated alerts for the same event. Suppression must outrank scoring.
Define freshness by event type. A role change remains an account-structure fact after it stops being timely outreach context; a public engagement can become irrelevant quickly; a company update may be superseded by a correction. Store both event time and observed time, then expire the action while retaining permitted audit evidence.
Define merge rules before volume arrives. The same company post may generate an account update, a lead-share alert, and a rep-observed event. Merge related evidence under one account review rather than issuing three rep tasks. Preserve each source record so the reviewer can see why the group formed.
Choose a permitted action
- Ignore: no fit, stale, duplicate, or irrelevant observable.
- Monitor: relevant change but no useful action yet.
- Research: verify context and determine whether the event changes a real account hypothesis.
- Update account plan: change stakeholder map, open questions, or next review date.
- Draft for review: only when relevance, channel policy, recipient identity, and supported claims pass.
Use the signal-based outreach contract before any buyer-facing action. Never reveal sensitive or surprising provenance merely because the system made it visible.
Work through edge cases explicitly
| Edge case | Safe handling |
|---|---|
| Lead changed role and company | Resolve effective employer and archive the former-account relationship separately |
| Post mentions a topic to criticize it | Classify the observable only; require human interpretation before relevance |
| Several employees engage with one post | Retain individual events but avoid claiming a buying committee |
| Profile view identity is not provided | Do not guess the person from company-level data |
| Alert conflicts with CRM ownership | Route to the current account owner; do not create parallel outreach |
| Buyer asks how the rep knew | Use truthful, non-invasive provenance or withdraw the claim |
Measure signal usefulness
Track received events, accepted signals, identity corrections, stale events, duplicates, suppressions, research tasks, reviewed drafts, accepted actions, no-action decisions, and critical errors by source and event type. Measure against eligible events, not only the signals that produced a response.
Audit a labeled sample regularly. A high-volume source can be less useful than a small source whose events are correctly identified, fresh, explainable, and connected to a permitted action. Do not claim outcome lift without a suitable design.
Start in read-only mode. Sample across alert types, account tiers, geographies, relationship states, and no-action cases. Have two reviewers independently label event identity, relevance, freshness, suppression, and permitted action. Adjudicate disagreements before evaluating the system.
Then run shadow routing and compare proposed owners and actions with the approved labels. Move to a bounded rep queue only after wrong-person and suppression errors meet the team's hard gates. Revalidate after an edition change, alert-definition change, identity-rule change, or connector update. Stop the workflow on cross-account disclosure, repeated duplicate tasks, or buyer-facing claims that exceed the observable event.
Where Gangly fits
Gangly repository facts describe selected LinkedIn profile, job-change, and company-update context through an extension, with outreach drafted for rep review rather than auto-sent. Evaluate Signal Detection on source coverage, identity, freshness, explanation, suppression, and accepted decisions; verify account requirements and current implementation.