Skip to main content
Use this workflow when you have a list of accounts and need to decide which ones deserve sales attention first. It replaces the old lead scoring, signal classification, account scoring, win/loss, and closed-lost recovery pages. The pattern is simple: enrich the account, detect the signals that matter for your market, score the account against your ICP, and write the reason back to a CSV or CRM.

When to use it

  • Rank a target-account list before outbound
  • Turn closed-won and closed-lost history into a better ICP score
  • Find niche signals that separate good-fit accounts from noisy lookalikes
  • Revisit closed-lost accounts when something material changed
  • Give agents a short, inspectable reason for each score

Run it from an agent

With Codex:

Workflow

1

Start with the account list

Use domains when you have them. Company names work, but domains reduce match ambiguity.
2

Enrich firmographics

Pull company size, industry, location, funding, and other structured context with company enrichment.
3

Detect current signals

Check for signals that change over time: hiring, fundraising, leadership changes, expansion, competitive mentions, product launches, and relevant news.
4

Compare against your own history

If you have CRM exports, compare enriched closed-won and closed-lost accounts. Look for signal differences that explain why deals won or lost.
5

Score and explain

Score each account against the ICP and detected signals. Write the tier, score, reason, evidence URL, and recommended next action.

Output schema

Useful variants

Niche signal discovery

Use this when your ICP is too generic.
Good outputs include signal definitions, examples, and caveats. Avoid treating one noisy correlation as a rule. Ask the agent to show the evidence it used.

Closed-lost recovery

Use this when a lost account may have a new reason to buy.
Route only accounts with a clear change. A lost deal with no new signal should stay out of the campaign.

New signal detection

Use this when you know a trigger matters but do not yet have a saved play.
Be specific. “Growth signals” is too broad. “Hiring two or more RevOps roles in the last 90 days” is usable.

Quality rules

  • Pilot on 5-10 rows before running the full file.
  • Keep the scoring model small enough to inspect.
  • Require evidence for every non-firmographic signal.
  • Separate fit from timing. A perfect ICP account with no current trigger may still be a lower-priority account this week.
  • Do not use AI to invent missing firmographics. If a field is unknown, leave it unknown.