An AI copilot for lead discovery helps a team turn a market question into a shortlist of accounts with signals, leads checks, and a reason to review each one. The useful output is a decision-ready research artifact, not an unexamined contact export. A person approves the shortlist before outreach or CRM changes.
In Dureach, this is a defined research job for a named specialist within an operator-reviewed AI workforce. The deliverable and approval point are explicit.
What should the lead discovery brief contain?
Name the business job first: find likely buyers, assess suppliers, or identify accounts for expansion. Give the researcher a product or service category, market, buyer role, exclusions, signal sources, and the decision the shortlist must support. Do not put passwords, private customer exports, or unnecessary personal data in the brief.
For example: “Find importers of this product category in this destination market. Show the shipment or customs signals, check company information, flag uncertain matches, and return a shortlist for review.” An HS code can help define scope, but product descriptions and company context still need checking.
| Brief field | What to specify |
|---|---|
| Scope | Product, market, account type, and exclusions |
| Signals | Shipment or customs records, company records, or another permitted source |
| Decision | Why an account qualifies and what would disqualify it |
| Output | Account, source, date, identity match, uncertainty, and next action |
How does Copilot turn a brief into a reviewable plan?
In Dureach, open Copilot at /app/copilot and enter the question and desired output. Use Deep Research from the Copilot tools when the job needs a saved plan and research artifact; the project lives under /app/ai-projects. Clarification is optional and should cover only gaps that change the scope or decision. Review and approve the plan before execution starts.
The research sequence is Prompt -> Clarify -> Plan -> Approve -> Research -> Validate -> Match -> Deduplicate -> Artifact -> Review. A run belongs to the server and survives browser closure. Progress should show observable work and signals summaries, never hidden model reasoning. The operator reviews the saved artifact and source trail before using it.
How should the team check signals and leads?
Work from Signals -> Leads -> Enrichment -> Decision -> Outreach -> Pipeline. For a trade-led search, inspect shipment or customs descriptions, dates, lanes, and repeated activity before inferring an account’s role. A shipment record can support a commercial hypothesis; it does not prove current purchase intent.
Check names, domains, locations, subsidiaries, and distributors against the intended company. Treat AI output as candidate data, keep authoritative records canonical, and use deterministic matching and deduplication before adding contact context. Keep provenance and replayable events so a reviewer can trace each decision. Logs must not contain secrets.
If signals is thin, records conflict, or a contact cannot be matched, mark the account uncertain and request a narrower pass. Do not fill missing facts with plausible text.
What should the reviewer receive and approve?
Give the reviewer a compact shortlist with the account, source and date, identity rationale, relevant product or market signal, confidence or open question, and proposed next action. The reviewer can approve, reject, or return an account for more research. Only approved findings should inform an outreach draft or CRM handoff.
Before anyone contacts a person, check the permitted use of each data source, contact provenance, applicable consent rules, opt-out and suppression status, restricted jurisdictions or use cases, retention, and sending limits. Review any claim based on trade data so the message does not imply private intent or surveillance. Record the owner and reason for each decision.
How does a shortlist reach the next workflow?
Pass the approved account summary, signals link, matched contact role, unresolved questions, owner, and next action to the outreach or CRM operator. Drafts and AI-suggested field changes remain candidates until reviewed. A reply or rejection should feed back into the account hypothesis and future research.
Measure approved accounts, rejected or duplicate matches, signals coverage, and handoff completeness. These measures show whether the discovery process produces usable decisions; list size alone does not.
For the app steps and visible states, see Use Copilot and Deep Research. To put a brief into a governed workflow, open /app/copilot, name the research job and required artifact, and review the plan before starting the run.
FAQ
What is the output of lead discovery?
A reviewed account shortlist with source signals, leads reasoning, uncertainty, and a proposed next action.
Does a trade record prove buying intent?
No. It is signals of recorded activity that needs context, date checks, and human interpretation.
When should I use Deep Research?
Use it when the question needs a saved plan, multiple sources, validation, and a reviewable artifact. It is a mode in /app/copilot; projects and artifacts live in /app/ai-projects.
Can AI write directly to CRM?
Treat AI-suggested changes as candidates. Validate identity, deduplicate, and approve the handoff before changing an authoritative record.
What if the signals conflicts?
Mark the match uncertain, keep both sources, and ask for a narrower search or human review. Do not silently choose one record.
What must be checked before outreach?
Review account fit, signals-based claims, contact provenance, consent where required, opt-outs, suppression, restricted uses, and sending limits.