A useful lead research pipeline produces a reviewable account brief, not a confident paragraph with no sources. Separate collection, normalization, scoring and human review so each claim can be traced and corrected. Research access does not itself authorize outreach; keep sending and publishing as distinct actions with their own approval.

Before you start

Define the target account and the business problem you can actually help with. List permitted sources and the fields you need. Start with public company information and keep the scope proportionate. Record source URLs and observation dates so a reviewer can distinguish a current fact from an older or uncertain signal.

The workflow

  1. Collect candidate accounts using explicit criteria. Preserve the reason each account entered the list. Deduplicate by a stable company identifier or reviewed domain match rather than assuming similar names identify the same business.
  2. Normalize the data into a small schema. Separate company facts, observed signals and interpretations. An unavailable field should remain unknown; do not ask a model to create a plausible replacement.
  3. Attach evidence to each material claim. A company website can support what the business offers. A visible operational issue may suggest a research question, but it does not prove a budget, buying intention or internal problem.
  4. Score against a documented rubric. Make the supporting evidence visible beside each score. Keep hard exclusions separate from positive points so a polished summary cannot compensate for an account outside the intended market.
  5. Send the brief to a person for review. Include uncertainties and a proposed next research step. A reviewer decides whether the account is relevant and whether any outreach is appropriate before another system acts.

Worked example

Illustrative example: a studio fits the target service category and publicly shows a multi-step inquiry process. The researcher records those observations with URLs and dates. It proposes checking whether handoff delays are a real problem, but does not claim the studio loses leads. The reviewer can accept the account for further research while rejecting the unsupported inference. No message is sent as part of this example.

Your worksheet

  • Account ID, name and verified domain: ___
  • Why it entered the list: ___
  • Fact → source URL → observed date: ___
  • Possible problem, labeled as hypothesis: ___
  • Score and evidence for each point: ___
  • Unknowns and next research action: ___
  • Reviewer decision and reason: ___

Test it

  1. Remove a source and check that its claim is flagged as unsupported.
  2. Include similarly named companies and verify identity matching.
  3. Ask the reviewer to explain a score using only the linked evidence.

What to measure

Track duplicate rate, missing evidence, reviewer rejection reasons and the time needed to verify a brief. Lead volume alone does not establish pipeline quality, and a research score is not proof of willingness to buy.