Design a governed lead-intake workflow without automated extraction

An engineering guide for turning authorized professional observations into reviewable lead candidates while protecting identity, purpose, provenance, and provider boundaries.

Educational research only. LinkedIn account access remains decision-gated. No DewEngine LinkedIn connector is approved or implemented.

Talk to us ↗Read technical guides →

Product framing

Define a lead candidate before discussing automation

A lead is a business judgment, not a profile copied into a database. Specify the target customer problem, required attributes, disqualifiers, owner, and next human decision. DewEngine's catalog names profiles, companies, relations, and search only as targets within a decision-gated LinkedIn product. It does not authorize automated extraction, and the workflow must remain useful with manual or customer-provided inputs if the gate stays closed.

Separate observed identity from commercial qualification
Document the decision that each requested field supports

Input boundary

Accept only authorized and attributable observations

The intake layer should record how each candidate entered the system: a user-selected record, a customer import, an authorized search result, or another approved source. Free text and URLs are hints, not proof of identity. Reject flows that ask a worker to crawl pages, defeat access controls, or impersonate normal browsing. A source label and observation timestamp must survive every later transformation.

Require an access decision for each source class
Preserve the original customer reference without copying unnecessary page content

Entity resolution

Build a candidate queue instead of automatic merges

People change employers, share names, use multiple languages, and appear in more than one customer dataset. Match suggestions can combine provider-scoped identifiers, customer-known email, company context, and prior manual links, but uncertain candidates belong in a review queue. A wrong merge can expose one person's messages or notes under another person's record, so confidence never substitutes for tenant-scoped access checks.

Show the evidence behind every proposed person match
Allow split and correction without losing the audit trail

Qualification

Keep scoring inputs inspectable and limited

If a product prioritizes candidates, use explicit customer criteria such as supported geography, company segment, or an existing relationship. Avoid hidden inferences about protected or sensitive traits, and do not treat profile completeness as intent. For example, a role change might trigger a research reminder, but it should not automatically create a sales stage or message. Users need to see which observations influenced the suggestion.

Prefer rules a customer can explain and override
Do not infer willingness to be contacted from professional visibility

Downstream control

End extraction before any provider-visible action

Lead intake, CRM creation, campaign eligibility, and message approval are distinct transitions. Each needs its own actor, reason, and outcome. A candidate may be valid research but suppressed for outreach, owned by another representative, or already in an active conversation. Idempotent imports and deterministic deduplication prevent repeated jobs from creating duplicate records or enrolling the same person twice.

Require a separate approved command for invitations or messages
Check suppression and existing conversation state at action time

Evidence plan

Measure quality and reversibility, not harvested volume

Qualification should prove approved access, tenant isolation, duplicate resistance, correction, retention, deletion, and graceful stop on account restriction or revoked authorization. Sample accepted and rejected matches for human review. Record false merges, stale observations, and unsupported fields. No production release can proceed without the LinkedIn legal and provider gate, consented conformance evidence, action budgets, audit, and a tested kill switch.

Track reviewed precision rather than profiles collected
Test deletion across normalized records, caches, exports, and logs

Questions

Before you build.

Is automated lead extraction an available DewEngine feature?+

No. LinkedIn account access is decision-gated, and DewEngine has no implementation or conformance evidence for this workflow. The guide describes a product model that could accept authorized inputs after the access decision is resolved.

Why not create a CRM contact as soon as a likely profile is found?+

A likely match can still be the wrong person, an outdated role, or a duplicate owned elsewhere. A reviewable candidate record preserves evidence and gives the customer a safe point to confirm identity, purpose, ownership, and contact eligibility.

Build with us

Does this match the workflow your users need?

DewEngine is in development. Real use cases decide what ships first.

Share your use case ↗