The problem this solves
Most ICP scores were set once, during a workshop, on whatever data the CRM held that week. Since then the database has doubled, the product has moved, and the market has shifted, but the scores still reflect the old snapshot. Sales works the list in score order anyway, which means the ranking quietly decides where human attention goes long after the inputs behind it stopped being true. A stale score is worse than no score, because it carries the authority of a number.
Re-scoring by hand is a project nobody volunteers for. It means re-pulling firmographics for thousands of accounts, reconciling them with whatever intent signal you have, running the rubric, and writing everything back without breaking the fields other automations read. So it happens annually at best, usually attached to a planning cycle, and the other eleven months run on decay. The accounts that changed the most since the last pass, the ones growing or adopting adjacent tools, are precisely the ones the old score misranks.
The fix is to make re-scoring cheap enough to be routine: live inputs pulled at run time, the rubric applied consistently across every account, scores written back with their reasoning attached, and the ranked book delivered somewhere the revenue team actually works. When the whole loop costs one instruction, quarterly re-scoring stops being a project and becomes how the database simply behaves, and the ranking finally deserves the authority the score field has always carried.
How the mission runs
- Assemble the scoring inputs. The agent pulls every account from HubSpot, then gathers current inputs per company: size and industry from Firmographics, stack and buying signals from Intent Data. Inputs are collected fresh at run time, so the score reflects what the account looks like today rather than at the last cleanup.
- Apply your ICP rubric consistently. Your fit criteria - size bands, industries, stack markers, intent thresholds - are applied identically to every account. Consistency is the point: a rubric a human applies drifts across a thousand accounts, while the agent scores account one and account one thousand with exactly the same weights.
- Handle the unknowns explicitly. Accounts with missing inputs get scored on what exists, with the gaps flagged in a confidence column rather than hidden inside the number. You can see which scores rest on complete data and which are provisional, which is exactly the distinction a rep needs before trusting the ranking.
- Write scores back to HubSpot. Each account's score is written back idempotently, with a note on the record explaining the score's components and the evidence date. Re-running the mission updates scores in place without duplication, and the note means any rep can open a record and see why it ranks where it does.
- Deliver the ranked book in Sheets. The full ranked book lands in Google Sheets: one row per account with score, component breakdown, confidence, and key firmographic facts. Sorted and filterable, it is the working document for territory conversations, outbound prioritization, and the always-awkward question of which accounts deserve attention this quarter.
The prompt
This is the exact objective the agent receives. Swap the obvious placeholders for your own domain, segment or channel and run it as-is from the console, Slack, or the API.
What comes back
Every account in HubSpot carrying a current ICP score with its reasoning noted on the record, and a Google Sheet holding the ranked book of business: score, component breakdown, confidence level, and supporting firmographics per account. The sheet is the artifact the revenue team argues over; the CRM scores are what routing, alerts, and reporting quietly consume from that day forward.
Make it yours
- Run it on new accounts weekly and the full database quarterly, so fresh signups get ranked within days while the whole book stays broadly current.
- Ask for a delta view: alongside the new ranking, a tab showing the accounts whose scores moved most since the last run, which is where sales attention should shift first.
- Split the output by segment or region into separate tabs so each team lead gets a ranked book scoped to the accounts they actually own.
Frequently asked questions
Where does the score's reasoning live?
In two places. Each HubSpot record carries a note listing the score components and the evidence behind them, and the Sheets book breaks the same components into columns. Every figure traces back to the tool call that produced it, so a disputed score is a checkable claim rather than an argument.
Will re-scoring disrupt automations that read the score field?
Writes are idempotent and update the same field in place, so downstream automations see a value change exactly as if a human had edited it. If your workflows trigger on score thresholds, run the delta variation first to preview how many accounts will cross a threshold before you commit the write.
How is my ICP defined for this?
You state it in the prompt: the size bands, industries, stack markers, and intent signals that make an account yours, with rough weights if you have them. The agent applies that rubric verbatim. Teams that want help deriving the rubric from their own won-deal history usually run an ICP definition exercise first.