Prospecting with Claude, without the research black hole

Here is how we describe our best customers today:
{paste your rough notes, or three customer names}

Here is what those accounts had in common at the time
they bought: {paste}

Rewrite this as a scoring rubric with 6 to 8 criteria.
Each criterion must be checkable from public
information or from a CRM field. No criteria like
"innovative" or "growth-minded".

Then list our three strongest disqualifiers, and for
each one tell me the evidence that would prove it.

The line that does the work is “checkable from public information or from a CRM field”. Without it you get an ICP full of attitudes, and every account scores well because nobody can prove otherwise.

Account: {name}. I sell {product} to {buyer title}.
Here is what I have: their careers page, their last
two press mentions, and the CRM record.
{paste}

Give me, in under 250 words:
1. What changed at this company in the last 90 days
2. Who is likely to own the problem I solve, by title
3. One thing that is true here and not true at a
   typical company this size
4. The single fact that justifies contacting them
   this week rather than next quarter

Cite which source each point came from. If a point
is inference, write [inference] in front of it.

Point 4 and the [inference] tag are the two lines that matter. Point 4 kills accounts that are not ready, and the tag stops a plausible guess turning into a first line you cannot defend on a call.

Here is a job description {company} posted this week:
{paste the full posting}

Ignore the skills list. Tell me:
- What broke or got too big that made this role
  necessary
- Which team the role reports into, and what that
  says about who owns budget
- What tools the posting implies they already run
- Whether this is a replacement hire or a new
  function, and how you can tell

Then write one sentence I could open with that a
competitor reading the same posting would not write.

“Ignore the skills list” is the whole prompt. The skills list is the part every tool summarises. The reporting line and the tools named in passing are where the actual buying context sits.

Here are 12 events my alerts flagged this week
across my accounts:
{paste: funding, launches, exec hires, layoffs,
office moves, award wins}

For each one, answer two questions:
a) Does this change what {buyer title} is measured on
   in the next two quarters? Yes or no.
b) If yes, what would they now be trying to fix?

Sort into: act this week, watch, ignore. Be willing
to put most of them in ignore.

The permission in the last sentence is deliberate. Ask a model to rank twelve events and it will find merit in all twelve. Telling it that a mostly-ignore answer is acceptable is what gets you an honest three.

Scoring rubric: {paste the rubric from earlier}
Disqualifiers: {paste}

Here are 40 accounts with firmographics and any
history we have:
{paste}

Score each out of 10 against the rubric, with one
line of reasoning quoting the field that drove the
score. Show the distribution before the list.

Rules:
- Anything hitting a disqualifier scores zero and is
  listed separately
- If a criterion cannot be checked from what I gave
  you, mark it unknown rather than assuming
- Do not smooth the scores to spread them evenly

“Show the distribution before the list” is the check. If 34 of 40 accounts land between 6 and 8, your rubric is not separating anything and the ranking underneath it is decoration.

Here are my 15 highest scoring accounts, with the
last touch date, who we spoke to, and any trigger
from this week:
{paste}

Split them into three groups:
- Work today: something changed, and I have a reason
  to contact a named person
- Work this month: fit is there, no reason yet, tell
  me what I should watch for
- Park: say plainly why, and what would bring it back

For the work today group only, give me the person to
contact and the one fact the message should hang on.
Keep that group to five or fewer.

Capping the group at five is what makes the output usable. An unlimited priority list is just the same list in a different order, and you will still start at the top and run out of morning.