Prompts to build customer intent scoring
This resource is based on The autonomous email engine behind 200% conversion growth, featuring Adam Cain of ElectricityRates.com, published on the AI Lab by ActiveCampaign.
Get the prompt template
How to use these prompts
These prompts are pulled from the scoring workflow Adam Cain built at ElectricityRates.com to power autonomous email campaigns. Copy, paste, and swap in your details where you see [BRACKETS]. The sequence mirrors how Cain evolved his prompts: start broad, then layer in constraints as you learn what the AI surfaces
Prompt 1: Open-ended pattern discovery
- Best for: Starting from scratch — finding patterns you haven’t explicitly programmed
- Use with: Google Gemini, ChatGPT, Claude, or any LLM that accepts file uploads
Open-ended pattern discovery
I’m sharing a dataset of [NUMBER] customers for [DESCRIBE YOUR BUSINESS].
Each record includes [LIST YOUR DATA FIELDS: e.g., contract end date, email engagement history,
website visits, past purchase behavior, potential savings amount].
Look at this data and tell me:
- Which customers do you think have the highest likelihood to [YOUR CONVERSION ACTION]?
- What patterns do you see in customers who take action?
- Are there any signals I’m not currently tracking that seem predictive?
Variables to fill in:
- [NUMBER]: your sample size (500–1,000 records is enough to start)
- [DESCRIBE YOUR BUSINESS]: e.g., “a SaaS company,” “a home services provider,” “an energy comparison platform”
- [LIST YOUR DATA FIELDS]: include everything you have, not just what you think is relevant
- [YOUR CONVERSION ACTION]: e.g., “renew their subscription,” “switch energy providers,” “make a second purchase”
What to expect: The AI will surface combinations you didn’t consider. Cain’s team discovered Gemini could identify customers who should receive outreach a specific week because a favorable rate had appeared in their area, a temporal signal no one had programmed.
Follow-up prompt
Based on the patterns you found, which 3–5 variables are most predictive?
What would a simple scoring formula look like?
Prompt 2: Assign a baseline readiness score (v1)
- Best for: Generating an initial 0–100 score for each customer record
- Use with: Google Gemini, ChatGPT, Claude, or any LLM
Assign a baseline readiness score
Based on your analysis, assign each customer a Switch Readiness Score from 0–10 that reflects their likelihood to [YOUR CONVERSION ACTION].
Use these data points to inform the score:
- [SIGNAL 1: e.g., contract end date]
- [SIGNAL 2: e.g., email engagement in last 30 days]
- [SIGNAL 3: e.g., potential savings percentage]
- [SIGNAL 4: e.g., number of website visits]
Explain your scoring logic and what the top-scoring customers have in common.
Variables to fill in:
- [YOUR CONVERSION ACTION]: same as Prompt 1
- [SIGNAL 1–4]: the top signals the AI identified in Prompt 1, plus your own known predictors
What to expect: A scored list with an explanation of the model’s logic. Review this carefully — the first version will surface the method, which you’ll refine in Prompts 3 and 4.
Prompt 3: Add a timing constraint (v2)
- Best for: Narrowing to customers who are actually actionable right now
- Use with: Google Gemini, ChatGPT, Claude, or any LLM
Add a time constraint
Revise the scoring model. A score above [THRESHOLD: e.g., 70] should only be assigned
IF [YOUR TIMING CONSTRAINT: e.g., the customer’s contract end date is less than 60 days away].
Customers who score above [THRESHOLD] but don’t meet the timing constraint should be
scored between [LOWER_RANGE: e.g., 40–69] based on their other signals.
Re-score the dataset and show me the updated distribution.
Variables to fill in:
- [THRESHOLD]: Cain used 70 as his mid-tier cutoff; adjust based on your conversion volume
- [YOUR TIMING CONSTRAINT]: The window that makes a customer genuinely actionable (contract end, trial expiry, renewal date, etc.)
- [LOWER_RANGE]: how to classify interested-but-not-yet-ready customers
What to expect: A tighter list of immediately actionable customers. The distribution should shift—fewer customers in the high-score range, but those who qualify are genuinely ready.
Prompt 4: Layer in a value threshold (v3)
- Best for: Ensuring outreach only fires when there’s enough upside to justify it
- Use with: Google Gemini, ChatGPT, Claude, or any LLM
Layer in value threshold
Add one more constraint. A score above [HIGH_THRESHOLD: e.g., 85] should only be assigned
IF [TIMING CONSTRAINT from Prompt 3]
AND [YOUR VALUE CONDITION: e.g., available savings are 7% or greater / discount is above $X /
inventory discount exceeds threshold].
Customers who meet the timing constraint but not the value condition should score
between [MID_RANGE: e.g., 70–84].
Re-score and show the final distribution with count at each tier.
Variables to fill in:
- [HIGH_THRESHOLD]: Cain used 85 for his top-tier segment; this is the group that gets your most aggressive outreach cadence
- [YOUR VALUE CONDITION]: the minimum upside that justifies sending: savings percentage, deal size, discount level, urgency signal
- [MID_RANGE]: customers who are close—they should be in a lighter nurture sequence
What to expect: Three distinct tiers: high-intent + high-value (immediate outreach), high-intent + lower value (nurture), and low-intent (drip or no outreach). Each tier maps directly to a campaign type in ActiveCampaign.
Prompt 5: Identify external trigger opportunities
- Best for: Mapping which real-world events should automatically activate a campaign for high-scoring customers
- Use with: Google Gemini, ChatGPT, Claude, or any LLM
Identify external trigger opportunities
I have a customer segmentation model with readiness scores.
My business is [DESCRIBE YOUR BUSINESS].
What external events or signals in my industry should automatically trigger
outreach to my high-intent customers? For each trigger, suggest:
- What the trigger event is
- Which customer segment it applies to
- What the email angle or message should be
- How quickly after the trigger the email should go out
Variables to fill in:
- [DESCRIBE YOUR BUSINESS]: your industry, product, and customer type
What to expect: A list of trigger-campaign pairs. For ElectricityRates.com, the primary trigger is a utility rate change—the system detects it, filters for customers in that service area with high scores, and fires a campaign automatically. Your equivalent might be a price drop, a competitor announcement, a policy change, or a new product launch.
Tips for better results
- Start open, then narrow. Don’t skip Prompt 1, the open-ended discovery step is where the most useful patterns surface. Cain’s team found Gemini surfaced signals they hadn’t programmed it to find.
- Let the model show its work. Always ask for an explanation of the scoring logic, not just the scores. You need to understand why a customer ranks high before you build an automated system around it.
- Test thresholds against real conversion data. If you have historical data on who actually converted, validate your scoring formula against it before connecting the score to live campaigns.
Ready for the full story?
Read The autonomous email engine behind 200% conversion growth, featuring Adam Cain of ElectricityRates.com, published on the AI Lab by ActiveCampaign.
Related
More data from the AI Lab.