Lusha company intelligence evidence workflow

Sales intelligence inside your AI agent

Turn company research into evidence your team can review

Use Lusha-oriented prospecting workflows to discover accounts, resolve company identity, evaluate decision-maker relevance, and preserve source context before a person approves any outreach.

Open the installation runbook
Firmographic contextDecision-maker researchHuman approval

Six controls for dependable prospecting

Research depth without a black-box handoff

01

ICP constraints

State firmographic criteria, technographic clues, geography, and exclusions before discovery begins.

02

Entity resolution

Compare operating name, legal identity, domain, location, and dated events before records merge.

03

Role relevance

Connect a decision-maker’s function and seniority to the buying problem instead of relying on title alone.

04

Source provenance

Keep observation dates and evidence context visible so reviewers can challenge stale or conflicting fields.

05

Verification states

Separate verified, config-dependent, and unresolved information rather than filling gaps with inference.

06

Outreach gate

Require human review of relevance, suppression, opt-out handling, and regional requirements before sending.

Operational facts, not performance promises

4Copy, run, configure, and review states
3Identity, recency, and provenance checks
1Named human approval gate
0Guaranteed matches, delivery, or replies
Transparent unknownsMissing evidence stays visible
Least privilegeCredentials remain in runtime secrets
Auditable changesCorrections retain their reason
Human-in-the-loopResearch never authorizes sending

Installation runbook

Four states from command to first reviewed result

  1. 01 Copynpx -y @okki-global/okki-go-taroball

    Success means the exact command is on your clipboard; no package has run yet.

  2. 02 Run

    Execute it in a supported terminal, inspect the package prompt, and record the installed version.

  3. 03 Configure

    Connect only approved sources and store API keys outside prompts, repositories, exports, and shared logs.

  4. 04 First result

    Use known companies, then check domain identity, dates, provenance, exclusions, and role relevance.

Review questions

What teams should decide before production use

Does copying the command install anything?

No. Copying changes only the clipboard. Installation begins only when an operator runs the command and accepts the package prompt.

Where should API keys live?

Use a runtime secret store with minimum scope. Never place credentials in an AI prompt, source file, exported record, or support screenshot.

Is a funding event the same as intent?

No. Funding, hiring, and leadership changes are observations that may justify research; they do not prove budget, timing, or buying interest.

Can the workflow send outreach automatically?

Research and drafting should remain separate from sending. A person must review relevance, verification, suppression, opt-out language, and applicable regional rules.

How should coverage be measured?

Test a documented cohort containing known positives, exclusions, ambiguous identities, and stale events. Report unresolved fields separately from rejected matches.

What should a first-result audit retain?

Keep the source, observation date, accepted field, rejected alternative, reviewer, and downstream decision so corrections remain explainable.