Run a 50-lead outbound pilot

Needs and produces

You need a non-empty company description, offering, and ideal customer profile, plus an available sender for eventual delivery. You also need exclusions such as competitors, customers, and open deals. Do not use this play for cold email from a primary domain without warmup, an undefined customer profile, or an unreviewed scale beyond 50. A missing sender does not block building and validating the sequence, but it does block actual delivery.

The result is 50 qualified rows at most, a previewed three to five step sequence, a clear record of what is drafted, validated, published, enrolled, or started, and the held-back count. It is not an uncapped prospecting campaign. A workbook remains the audience boundary, so enrollment reads qualified worksheet rows rather than creating contacts one at a time.

Preflight the system and source the first batch

Architect first reads Context to confirm the company, offering, and ideal customer profile. It then finds a shipped outbound system, preflights it for a 25-lead pilot, and installs the system with automation disabled. Preflight and installation are free in the catalog posture. The tables work begins with a row source that can carry the specific target criteria, not a broad people search that forces expensive cleanup later.

The list-building skill attaches one source to the worksheet, puts every stated title, geography, and other supported criterion into that source, and runs a capped sample. Read the actual landed rows. An apparently successful filter can still return the wrong people, and a count alone does not prove relevance. Add the inexpensive and selective qualification evidence before contact data, then read the score or judgment for every criterion the operator named.

Validate the audience and message before delivery

Use the outreach skill to write one complete preview for a real qualified prospect. Every sending message should use the worksheet field that explains why this person is relevant: a hiring role, post text, company evidence, or another visible signal. The signal is about the recipient, not the method used to find them. First-name greetings use a first-name field, and every row-field value needs fallback copy that remains real when the evidence is empty.

Validate after every change and preview several rows, including a row with thin evidence. A LinkedIn cold sequence has a connection-request step and a separate DM step, whose body sends only after acceptance. Use the channels the qualified rows can actually receive. A row with a profile and no email supports LinkedIn steps, while email requires a populated email field. Review the final copy, sender, audience count, and pace in the sequence view before moving state forward.

1. 1

   Confirm the pilot boundary

   Load list-building and outreach, inspect Context and exclusions, then preflight the 25-row system. Check the preflight scope and estimate.
2. 2

   Source and qualify 25 rows

   Attach one source in a worksheet, run a capped sample, and read the rows. Add evidence and qualification before any contact lookup, then inspect qualified and excluded counts.
3. 3

   Draft real-row messages

   Write from source evidence and worksheet fields. Preview several rendered messages, including fallbacks, and validate every requested Sequence step.
4. 4

   Publish and enroll separately

   After review, publish the sequence and enroll a reviewed selection. The app result should show its saved state and the exact enrolled and held-back counts.
5. 5

   Start only when asked

   A start is a separate action. If explicitly requested, start under the chosen cap and read back the result. Otherwise report the Sequence as published and enrolled, not sent.

Credits and leading metric

Preflight and install do not spend credits according to the play catalog. The source and verification work can spend under their cap. Record the estimate before each live paid run and the charged credits from its completed result. The catalog calls a trial safe only a live run that stays below an explicit cap. Enrollment does not itself create provider charges, but delivery is still an outward action with a capped sender pace.

Track reviewed qualification rate for the first 25, then the quality of the 50-row set and the reply rate only after a start. Report the source filters, worksheet link, sample size, qualification rate, credits, sequence steps and channels, and the exact state proved by the final read. Do not claim delivery from a draft, publish, or enroll response.

When it goes wrong

If fewer than half of sampled rows match the intended audience, tighten or correct the source input, then rerun a small sample. Do not add layers of enrichment to a weak source. If a column is empty, failed, or waiting on an upstream field, read the column and its run, correct the binding or gate, and continue only the affected rows. Use Find out why something failed for a state that remains unclear.

If a sequence is silent, distinguish draft, validated, published, enrolled, and started. Also inspect sender connection, sending hours, and caps. Never retry an unknown external outcome until the sequence and conversation state have been read. If the sender is missing, finish the draft, preview, and validation, then record the connection needed rather than duplicating the Sequence.

For a published correction, read the saved steps before editing. Change only the requested copy or field binding. Validate again, then read several rendered rows. Do not create a near-duplicate sequence for the same audience. A completed step with a receipt is not run again. A pending state is not a failure and is not zero. A paced or capped state identifies a delivery limit. It does not justify widening the recipient scope. Report the observed state, the bounded repair, and the readback.