A prospect list is a sheet you can sort, not a chat

I keep seeing "AI built me a prospect list" screenshots. They are chat windows. A blob of companies, half with the wrong URL, sitting in a paragraph like a grocery story.

That is not a list. A list is a sheet you can sort, filter, and hand to someone who has to live with it on Wednesday.

The artifact is the sheet. If you cannot open it in the morning and delete the rows that are junk, you bought a vibe.

Criteria, then research, then enrich, then you

You write the criteria like a hire. B2B landscapers in three states, 10 to 80 people, already running ads, must have a real booking page, skip anyone who is hiring a CEO this month. Ugly and specific. "Interesting companies" is not criteria.

Scout takes that file and a computer. It opens directories you named, homepages, jobs pages, the occasional news post. It fills the columns you defined: name, site, city, employee band if a page says it, booking URL, a one-line why-they-fit, source link.

Empty cells are allowed. Invented employee counts are not. If it cannot fetch it, it leaves it blank and you decide whether to pay a data platform for that column.

Then enrich, still in the sheet. Not a new essay. Same rows, more facts, still sourced.

Then you. You cut the garbage. You mark the ten you actually want. Outreach drafts wait. Send is a tap.

That sequence is the job. Skip criteria and you get a tourist list. Skip the sheet and you get a chat. Skip your approval and you get banned.

Data platform vs manual vs a teammate

Pay ZoomInfo, Apollo, the rest, when the value is the database. Historical firmographics, direct dials, the graph you cannot browse into existence by Tuesday. They win there. Do not pretend a browser is a data warehouse.

Do it manually when the list is twenty accounts and you already know them. A teammate would be theater.

Hire Scout when the sources are public, the list is tens to a few hundred, and the columns are custom enough that a data platform will not have "has a booking page" as a filter. LangChain put research and data around 24.4% of agent deployments. This is that job with a spreadsheet at the end.

The computer is for fetching pages, not for claiming a private graph.

Would this force a context switch. If Scout also writes the sequence and also decides ICP, you mixed strategy, research, and Mail. Split. You pick ICP. Scout fills. Mail or Maker drafts. You send.

Can they fetch without inventing. Homepages, yes. Intent scores, no. LinkedIn behind a wall they should not fake, no.

Do research rules poison outreach rules. Yes, if the researcher is allowed to get clever in the first line. Keep the sheet boring. Clever happens later, in a different head, after you have picked the rows.

Honesty about what they actually do

They browse. They compile. They fill columns you defined.

They do not magically know a cell number. They do not "find anyone who needs this." They do not replace your taste for who is worth a Tuesday. First-time invention of the ICP stays yours. The tenth refresh of the same sheet is a job.

They miss. JS-heavy directories. Pagination they stop early. A company that lists the team on a PDF. You will still delete rows. Budget for that, the same way you budget for a human researcher who is not a wizard.

Watch is a cousin, not the same hire. Watch writes a weekly brief on pages you already care about. Scout builds the sheet of people you do not know yet. Different trigger. Different artifact.

What a good Tuesday looks like

The sheet has a frozen header. Scout ran at 7. Forty new rows, each with a source URL. Twelve blanks in employee count. You sort, kill fifteen, star eight.

Mail has drafts for the eight, in your voice file, parked. You send three. Cap for the day. Clerk is not in this meeting.

Chief only exists if this lane is one of several. If prospecting is the only job, you are the Chief. Fine.

Octopus is $99 for the computer, the board, OSS models. Bring ChatGPT, Claude, or Grok. No markup. Own email for Scout so replies have a door. Vault for any logins you actually want to give it. Dedicated workspace. We never train on the sheet. Seven-day money-back.

If your vendor shows you a chat full of names and calls it pipeline, ask for the CSV. If they cannot produce a working sheet, they did not build a list. They told you a story.