Dual Personalization Confidence Gate
Score person and company evidence before automating personalized outreach.
- Difficulty
- Moderate
- Time to result
- ~weeks to results
- Steps
- 5
- Confidence
- 98%
The Dual Personalization Confidence Gate treats individual and company personalization as separate evidence problems. The system gathers data about the person, scores whether it has enough reliable material for a personal anecdote, and independently scores whether it understands the company well enough to make business-specific claims. If a score clears its threshold, that layer of personalization can be used. If it does not, the system omits the unsupported element or sends the case to a human. When both scores are strong, the message combines a human-level connection with company-level relevance. This prevents an AI agent from manufacturing intimacy merely because its template expects a personalized opening.
Origin
Kieran Flanagan and Kipp Bodnar developed the gate while discussing why AI performs better for people and companies with abundant online information.
Core principles
- 01Person-level and company-level personalization require separate evidence.
- 02The amount and reliability of available data should determine automation confidence.
- 03Low confidence should remove unsupported personalization or trigger human review.
- 04More online information generally improves AI's ability to personalize accurately.
How to run it
- 1
Separate the targets
Create distinct evidence collections for the individual and the company rather than treating all personalization data as interchangeable.
Pro tip Label every fact with its target, source, and retrieval date.
Watch out A strong company profile does not justify personal claims about an employee.
- 2
Score individual confidence
Assess the volume, relevance, recency, and source reliability of information about the person.
Pro tip Require multiple corroborating public signals for a personal anecdote.
Watch out Do not reward data volume when the material is sensitive or irrelevant.
- 3
Score company confidence
Assess whether available business information supports a credible problem and solution connection.
Pro tip Prioritize operational signals over generic company descriptions.
Watch out Company news may not imply a need for the product.
- 4
Apply thresholds
Use person-level and company-level content only when their respective scores exceed predefined minimums.
Pro tip Calibrate thresholds against factual-error and response-rate data.
Watch out A single global threshold can hide different risks across the two layers.
- 5
Degrade or hand off safely
When evidence is insufficient, omit the weak personalization, use a simpler message, or route the prospect to a human researcher.
Pro tip Record why each case failed the gate to improve future data collection.
Watch out Never fill a missing personalization slot with an invented detail.
In the wild
An agent finds numerous interviews and posts from one executive, producing a high individual-confidence score, while another buyer has only a job title and LinkedIn page. It uses a validated personal opening for the executive but omits that component and requests human review for the second buyer.
→ Automation adapts to evidence quality instead of fabricating equal personalization for every recipient.
Common mistakes
Using one combined score
Combining person and company data can hide the fact that one personalization layer has no supporting evidence.
Treating online volume as permission
Public availability does not automatically make every detail appropriate for prospecting.
Automating below the threshold
Generating a required anecdote despite low confidence defeats the purpose of the gate and encourages hallucination.
Is it for you?
Best for
AI prospecting systems that create individualized B2B messages from web, CRM, and company data.
Not ideal for
Fully manual, low-volume outreach where a person can directly validate every fact.
From the transcript
“You could break personalization into personalized for the individual, personalized for the company.”
“And then you probably need some sort of confidence score, right?”
“And if it doesn't have the confidence score, it knocks it over to a human.”
From the episode
How To Master Sales Prospecting With Ai In 2024