Outbound Account Research Sprint

Goal

Run a daily outbound desk that researches 15 target accounts, writes a first line only someone who looked would write, and stages a four-touch sequence a rep approves before anything sends.

Tools required

Direct is the vendor's public site. Affiliate appears only when ShipPlays is enrolled in that program.

• Apollo.io

  Direct

  Build the account list. Pull a work email only after the row is approved.

• Clay

  Direct

  Put hiring, stack, and a page from their site on one row per account.

• Claude

  Direct

  Research synthesis and drafting. Any long-context model you are allowed to use will do.

• Instantly

  Direct

  Send only the rows a named human marked approved. Do not connect this on day one.

Step-by-step setup

1. Write the lane in one sentence

   A stranger should be able to apply it. Include who they sell to, the trigger that makes this week relevant, and who you will not contact. Example: B2B software, 20–150 people, hired a first head of sales in the last 90 days, US or Nordics, not agencies. If two reps would pick different accounts from the sentence, it is too soft. Do not open a model until this sentence exists.

2. Create the 15-row sheet

   Columns: company, domain, trigger evidence URL, contact name, role, email, first line, touch 2, touch 3, breakup, status, source links. Status starts at draft. Leave email blank until a human approves the row. The sheet is the system of record. The chat is not.

3. Pull candidates, then enrich only the maybes

   Export companies that match the firmographic half of the lane. Drop them into Clay and ask for three evidence fields only: a hiring signal, a stack signal, and one sentence from their site about who they serve, each with a URL. A row with no evidence URL does not get a first line.

4. Run the research pass on rows that have evidence

   Paste up to 15 rows into the research prompt. Require a paraphrase plus the URL. Reject any first line that would still be true if you swapped the company name. That test catches almost every generic opener.

5. Run the critic before anyone approves

   The critic kills invented numbers, fake familiarity, and lines that compliment a post nobody read. It returns keep, rewrite, or disqualify. The rep spends the daily 45 minutes on rewrites and approvals, not on blank-page writing.

6. Approve by hand, then send

   Paste approved rows into the sequencer yourself for the first two weeks. Add a sync only after those weeks are clean. Log replies for 10 business days with tags: not now, wrong person, interested, unsubscribe, angry. On Friday, paste the angry and wrong-person notes into the critic as negative examples.

Exact prompts

You are the research desk for an outbound sprint. You do not send email. You do not invent facts, metrics, mutual connections, or posts.

Lane (apply strictly; disqualify anything outside it):
{ICP_SENTENCE}

For each account, use ONLY the evidence pasted with it. If a row has no URL, status is disqualify.

Accounts:
{ROWS_WITH_EVIDENCE_URLS}

Return one block per account, in the same order, with exactly these lines:
- Company:
- Status: draft or disqualify
- Evidence: one sentence paraphrasing the source, then the URL
- Why this week: the trigger, or "none"
- First line: one or two sentences a rep could send. It must become false if the company name is swapped. No "loved your post." No invented numbers. No "quick question."
- Do not claim: anything you almost wrote but could not source

A short disqualify list is a successful run. Never fill a weak row with a plausible line.

You are the critic for outbound first lines. You do not rewrite the whole batch. You judge.

Lane:
{ICP_SENTENCE}

Drafts:
{RESEARCH_PASS_OUTPUT}

For each company return:
- Verdict: keep, rewrite, or disqualify
- Reason: one sentence
- Failure mode, if any: generic (survives a company-name swap), invented fact, fake familiarity, wrong lane, no URL, or none
- Replacement first line: only if verdict is rewrite. Same evidence URL. If you cannot write a specific line, verdict becomes disqualify.

Also list up to three patterns the rep should ban in the next batch.

Write touches 2, 3, and a breakup for this already-approved first line. Do not change the first line. Do not introduce a new fact that is not in the evidence.

Approved first line:
{APPROVED_FIRST_LINE}

Evidence URL and paraphrase:
{EVIDENCE}

Constraints:
- Touch 2 adds one useful observation from the same evidence, then a single ask.
- Touch 3 is shorter than touch 2 and refers to the same trigger.
- Breakup is two sentences, no guilt, no fake deadline.
- No calendar links unless I pasted one.
- No percentages, customer counts, or "teams like yours" unless they appear in the evidence.
- Reading level: plain spoken English. No "circling back," "synergy," or "leverage."

Return the three touches labeled, plus a one-line note on what a rep might still want to cut.

Expected results

• 15 researched rows on a running day, with a visible disqualify pile.
• First lines that cite a linked fact and fail the company-name swap test.
• A four-touch sequence the rep can edit in under ten minutes.
• A reply log that tells you within two weeks whether the lane is real.
• Zero sends that did not pass a named human.

Guardrails

• Do not auto-send. The sequencer receives approved rows only.
• Do not invent metrics, mutual connections, or events.
• Do not buy or guess personal inboxes. Use a work email from a source you can name.
• Honor opt-outs the same day. Suppress the domain, not only the thread.
• Stay inside the sequencer's limits and your domain warm-up. A better prompt does not fix a cold domain.
• If you cannot link the evidence, you do not send the line.

Common mistakes

Related

• Two-Minute Inbound Lead Router
• Proposal Assembly Line