The prospecting sheet is the least glamorous document in sales and the one that decides everything downstream. Traditionally it is built by a VA over a week, arrives as a spreadsheet of 2,000 half-checked rows, and quietly poisons the campaign built on it. Here is the pipeline we use to go from a raw Sales Navigator search to a campaign-ready list in an evening, as we did recently for lists of 2,125 construction prospects and 1,121 marketing managers.
Step one: over-collect on purpose
Start with a deliberately broad Sales Navigator search and export everything. Trying to make the search itself perfect is a trap, because Navigator's filters are approximations: titles are self-reported, company sizes lag reality, and industry tags are chaos. Collect wide, then let scoring do the discrimination. A 2,000-row raw export that gets cut to 1,300 good prospects beats a "precise" 800-row search that misses half your market because someone titled themselves Commercial Director instead of Sales Director.
Step two: AI cleans before AI scores
The pass most teams skip. Before any scoring, the AI normalises the sheet: deduplicates people who appear under two companies, flags profiles with no activity in twelve months, catches the job seekers whose headline says open to work, and standardises company names so the same firm does not appear four ways. On a typical raw export this pass removes or repairs 10 to 15% of rows. Every one of those rows would have been a wasted connection request or a duplicate embarrassment.
Step three: score everything, explain everything
Each prospect gets the 0-100 score across fit, timing and reachability, with a one-line reason attached. The reason is not decoration. When the founder skims the sheet and sees "skip: 2,300-employee company, in-house sales function" they either agree in one second or correct the model, and every correction sharpens the next thousand rows. A score without a reason cannot be argued with, and a model that cannot be argued with cannot be trusted.
Step four: enrich only the pursues
Emails, company websites and phone-adjacent context get looked up only for prospects that survived scoring. This ordering sounds obvious and is skipped constantly: teams pay to enrich entire raw lists, spending on skips and maybes. Enriching after scoring cuts data spend dramatically on a big list while covering everyone who could actually enter a sequence.
Step five: the sheet becomes a system, not a file
The finished list does not go back to a spreadsheet. It loads into the campaign with its scores and reasons attached, the CRM rows are created, and from that point every event, connection accepted, reply received, intent tagged, meeting booked, writes back automatically. The prospect sheet stops being a document anyone maintains and becomes a live view of a database that maintains itself. Six weeks later, nobody is asking which version of the spreadsheet is current, because there is no spreadsheet.
What this replaces, priced honestly
A VA-built list: a week of elapsed time and £150 to £300 per thousand rows, with quality you discover only after the campaign burns through it. A data vendor list: instant and broad, but built from the same tired database your competitors bought, with none of your timing signals. The AI pipeline runs in an evening, applies your judgement at row level, and the output improves every time you correct it. The list is the campaign. Build it like it matters.
This pipeline, list build through daily execution, is what AI Sales runs end to end. The scoring layer on its own is free: ICP Scorer, paste a list and see what gets cut.