ICP Refinement Agent
Score each new account against your ICP model as it arrives, then use closed-won, renewal, and expansion patterns to suggest where the model should change.
- Trigger: an account is added
- AI steps: score the account, then suggest ICP refinements
- Output: ICP fit and confidence on the account, with refinement suggestions for RevOps
Agent breakdown
A blueprint to adapt, not a fixed recipe. Set the trigger, thresholds, and methodology that fit your team, and each step builds on the last, using the conversation and CRM context you already capture.
Score each account as it arrives
As soon as a new account is added, the agent starts scoring it against the ICP model. Running the score on arrival means the first person to work the account can see its fit without waiting for a monthly batch.
- Trigger: A new account is added
- Timing: Score the account as it arrives rather than in a monthly batch
Pull the account attributes and buying signals
Next, the agent pulls the account's size, sector, and stack, along with recent hiring and funding signals. The firmographics show whether the company looks like your ICP, while the signals help distinguish a good-looking account from one that may be ready now.
- Signals: Hiring and funding
- Window: The last two quarters
- Skip: Accounts already scored
Score the account's ICP fit
Using the account attributes and signals, the agent scores the account against the ICP model and gives its confidence out of five. It names the two attributes that had the greatest effect on the result and calls out missing information instead of filling the gaps with assumptions.
Score this account against the ICP model.
Input: the account's size, sector, stack, hiring signals, and funding signals from the last two quarters.
- Give the account an ICP fit score.
- Give your confidence in the score out of five.
- Name the two attributes that had the greatest effect on the score.
- List any attributes that are missing.
Rules: assess only the account information provided. Where an attribute is missing, say so rather than assuming the average.
Return the ICP fit score, confidence out of five, the two deciding attributes, and any missing attributes.
Save the fit score and confidence
After scoring, the agent writes the ICP fit and confidence to the account. When confidence is low, it records that uncertainty instead of rounding the result to a yes or a no, so anyone viewing the account can see where the model was unsure.
- Leave blank: Anything that remains unclear
- Re-score: When new signals arrive
- Keep: Any manual override
Suggest updates to the ICP model
Across scored accounts, the agent compares ICP scores with closed-won, renewal, and expansion outcomes. It brings the patterns together and suggests updates that make the model reflect the customers that prove to be the best fit.
Analyze how well the ICP scores for each account predict closed-won, renewal, and expansion outcomes.
Input: each account's ICP score and its closed-won, renewal, and expansion outcomes.
- Compare the ICP scores with the outcomes for each account.
- Identify the account attributes and signals that appear most often among closed-won, renewed, and expanded accounts.
- Identify attributes that receive strong ICP scores but do not predict those outcomes well.
- Aggregate the findings across the accounts.
- Suggest updates to the ICP model based on the patterns among the best-fit customers.
Rules: base every suggestion on the account scores and outcomes provided. Do not assume a pattern where the account data does not support one.
Return the aggregated findings and a list of suggested ICP updates, with the pattern behind each suggestion.
ICP fit scores and confidence on accounts, with refinement suggestions for RevOps
See how each new account fits your current ICP and how confident the model is in that score. You also get suggested ICP updates based on the attributes and signals shared by accounts that closed won, renewed, or expanded.
AI Agents that pair well
Same data, different jobs. Run them together for a fuller picture.
Learn more about GTM & Revenue Operations
RevOps Lab Podcast
50+ Free GTM & RevOps Cheat Sheets
Free RevOps Community (1k+ members)
.webp)