For a first pass, I would usually start with 10 to 25 redacted leads. That is enough to see whether the workflow has missing owners, unclear next actions, duplicate requests, stale status labels, or records that should not be contacted. It is small enough that a person can inspect the result instead of trusting a summary without checking it.
The number is less important than the mix. Do not choose only the cleanest rows. Include a few recent inquiries, older open records, possible duplicates, records with a completed task, records with a missed call, and at least one row where the next step is unclear. If every row comes from one channel or one owner, the sample may hide the handoff problem you are trying to find.
## What to keep in the sample
Useful columns usually include a lead or row identifier, received date, source, owner, status, last contacted date, next action, context, and contact-permission signal. Keep only the information needed to answer the review questions. Remove passwords, payment data, full unredacted conversations, health or legal records, recordings, and unrelated customer history.
For each row, ask three plain questions:
- What event created this record?
- What evidence supports the current next action?
- What would stop another customer-facing message?
If the answers are clear, the row may be Ready for a human to review. If an owner or decision is missing, use Hold. If two rows describe the same event, use Duplicate while preserving the relationship. If an opt-out or complaint is present, use Do Not Contact. If the source event cannot be found, use Missing Context.
## Why not start with the full export?
A full export can create more privacy exposure and more noise before the team understands the classification. It can also make a weak rule look authoritative because the output is large. A small sample lets the owner correct the categories, identify the fields that matter, and decide whether a broader review is appropriate.
The Missed Lead Recovery review accepts a small redacted sample in the browser. It does not require a CRM password, send messages, change records, or decide contact permission. The owner reviews the evidence and decides what may happen next.
The first review should answer whether the workflow is understandable and safe to inspect. It does not need to predict revenue or prove that every old lead can be recovered. Starting with 10 to 25 representative rows gives a team a concrete repair list without asking it to share more customer data than the decision requires.
If the first sample is mixed, keep the categories visible instead of forcing a single score. A row with a clear request but no owner is different from a duplicate with a known owner. A row with an opt-out is different from an old inquiry whose source cannot be found. These distinctions help the owner decide what to repair first and make it easier to explain why some records should not move forward.