Observable ICP criteria
Define industry, geography, size proxies, event windows, and exclusions before reviewing results.
CRM enrichment and buying-signal research
Turn account signals into evidence your team can review
RB2B helps an AI agent research private companies, reconcile identity clues, and preserve the source and date behind material claims. Signals remain hypotheses until a person confirms relevance, permissions, and outreach readiness.
Open the installation runbookSix operating controls
Research that stays inspectable from brief to approval
Entity resolution
Compare operating name, domain, headquarters, subsidiaries, and dated events before merging records.
Signal chronology
Keep funding, leadership, acquisition, and hiring observations attached to their reported dates.
Role rationale
Explain why a professional function may matter without pretending that a title proves responsibility.
Provenance states
Separate verified evidence, configuration-dependent output, unresolved values, and analyst inference.
Human review gate
Check relevance, suppression, regional requirements, and contact verification before any sending step.
4 run statesCopy · Run · Configure · Review
3 evidence checksIdentity · Recency · Provenance
1 approval gateHuman decision before outreach
0 guaranteesCoverage, delivery, and replies vary
Data and security ledger
Trust is documented as an operating state
Connected sources
Configuration-dependent. Record each provider, permitted purpose, available provenance, and freshness behavior.
API credentials
Required practice. Keep secrets outside prompts, repositories, exports, screenshots, and shared diagnostic logs.
Runtime behavior
Inspect locally. Review package prompts, network destinations, retries, errors, and output locations before production use.
Retention and deletion
Policy-controlled. Decide access, correction, suppression, retention, and deletion for company and contact records.
Four-state runbook
Install, configure, and validate a known cohort
- 01 Copy
npx -y @okki-global/okki-go-taroballClipboard confirmation does not mean installation succeeded.
- 02 Run
Execute in a supported terminal; inspect the package prompt, permissions, and completion output.
- 03 Configure
Store provider credentials in runtime secrets with the minimum scope required by the task.
- 04 Review
Test known companies and check identity, source dates, unresolved fields, exclusions, and destinations.
Practical boundaries
Questions to settle before production use
Does a funding event prove purchase intent?
No. It is a dated organizational signal that may justify research. Budget, urgency, fit, and authority still require separate evidence and review.
Can every CRM field be filled?
No. Coverage varies by source and market. Preserve unknown values instead of replacing them with unsupported inference.
Where should credentials live?
Use an environment secret store with minimum scope. Never paste keys into prompts, source control, screenshots, or exported research.
Can the workflow send outreach automatically?
Research output should remain disconnected from sending until a person checks professional relevance, suppression, opt-out language, and applicable regional rules.
How should accuracy be tested?
Use a cohort containing known positives, exclusions, stale events, renamed companies, and ambiguous domains. Score fields and evidence separately.
What does installation confirm?
A successful run confirms package execution only. It does not establish data quality, lawful use, inbox placement, response rates, or commercial fit.