Persona-level intelligence contact analysis is the work of checking what the persona fields on a set of contact records actually say, how many records carry them, where each value came from, how often two readings of the same title disagree and how many routing decisions would change if a disputed value flipped. The result is a decision for each field: use it to route, use it only to order work, or hold it back.
Persona-Level Intelligence Contact Analysis: The Short Answer
- Most published "analysis" has a different unit. Vendors analyse one executive (Draup, LINX), one archetype per person (Conversion Realtor) or one persona across the customer's history (6sense). The unit here is the contact list and its persona fields.
- The audit has five measures. Coverage, the Re-Read Test, the Disagreement Table, routing effect and action yield. Each one needs only a sample and a current company page per person.
- The output is a per-field verdict, not a score. Function may pass for routing while level passes only for ordering. A single list-wide "persona accuracy" number would hide that.
- No vendor figure is needed. The cited vendor pages publish no error rate for function, level or persona assignment, so the audit measures your own list against your own reader.
What "analysis" analyses: five units
The same two words are used for four different jobs. Sort a source by its unit before using its method.
| Unit of analysis | What gets analysed | How a source describes it | What the description says about inputs |
|---|---|---|---|
| One executive | Career history, remit, priorities, signals | Draup's Persona Analysis Agent "creates a deep executive profile using leadership history, responsibilities, strategic priorities, hiring patterns, tech influence, and organizational signals". LINX says persona analysis is aimed at "the few who truly matter, i.e. executives", and lists personal insights such as family life among its contents. | Draup's page lists output sections, including account signals for the last 90 days; it does not state how each input was sourced. |
| One person as an archetype | Motivations, objections, timeline, communication style | Conversion Realtor profiles "each buyer and seller as a distinct decision-making archetype". | Not stated on the cited page |
| One persona across the customer's history | Which function-and-level combinations were engaged before won deals | 6sense's Persona Importance model "provides a persona's level and position within the list of contacts, comparing it with" the company's past outreach on efforts "that ended in a sale". | Three listed inputs: average profile fit, normalised CRM count and pre-win engagement |
| The role mix at an account | Which functions and levels are likely to be present | Bombora says persona-level insight gives "the likely mix of players across functions and levels". | Not stated on the cited page |
| The contact list's persona fields (this guide) | Coverage, agreement and routing effect of stored function, level and persona values | The audit below | Your own sample, read against current company pages |
Inference: the first four units describe a person or a role. None of them says whether the stored fields that route a hundred contacts into sequences are right. That is the gap the audit fills, and it sits upstream of the single-contact read in the persona-level intelligence contact profile analysis guide.
The Persona Layer Audit: five measures
| Measure | Question | How to compute it | What it supports | What it cannot show |
|---|---|---|---|---|
| 1. Coverage | How many sampled records carry a value in each persona field? | Count non-empty values for function, level and persona per field | Whether a rule that depends on the field can reach the whole list | Whether a value is right |
| 2. Re-Read Test | Does a person reading the verified title reach the stored value? | A reader sees only the current verified title and assigns function and level, then compares with the stored values | An estimate of mapping disagreement against one reader's judgement | Truth: the reader can also be wrong, so use two readers for border cases |
| 3. Disagreement Table | Do two sources agree on the same person? | For contacts present in two sources, cross-tabulate the level or function each gives | Whether the field is stable across suppliers | Which source is right |
| 4. Routing effect | How many routes change if the disputed value flips? | For each disagreement, check whether the contact would cross a rule boundary, such as a Director-and-above line | The cost of the disagreement | The size of the error in the rest of the list |
| 5. Action yield | How many audited-good contacts end in a named action? | Count contacts that survive checks 1 to 4 and have a dated, public event to cite | Whether the layer produces actions or only labels | Whether the actions convert |
The tiers used below are those of the Four-Tier Persona Ledger in the persona-level intelligence contact summary guide: Recorded fields are copied from a source that states them, Mapped fields classify a Recorded value, Modelled fields are computed scores and Guessed fields are attributed traits. The audit tests Mapped fields against Recorded ones. It does not test Guessed fields, because there is nothing to check them against.
How to run it
- Draw the sample. Take a random sample from the list, balanced across the sources that fed it. Fact (arithmetic): a share measured on 40 records carries a 95 per cent margin of at most about 15 percentage points either way, from 1.96 × √(0.5 × 0.5 ÷ 40). Inference: a sample of 40 exposes whether a field has a problem and which kind. It does not give a precise rate, so report counts, not percentages.
- Verify the title first. Check each sampled person's employer and title against a current company page or profile. A record whose title cannot be confirmed is dropped from the persona checks, because a classification of a title that may be wrong tests nothing.
- Measure coverage. Count the non-empty values in each persona field.
- Run the Re-Read Test. A reader who does not see the stored values assigns function and level from the verified title. Record each match and each mismatch. Where a title is compound or uses words such as Head, Lead or Principal, have a second reader decide.
- Fill in the Disagreement Table. For people present in two sources, cross-tabulate the two levels. Agreement on the diagonal is stability. Off-diagonal cells are the contacts whose route depends on the source.
- Compute routing effect. For every mismatch or disagreement, test whether the route would change under the other value.
- Write the verdict per field. The three outcomes are in the table in the decision section below.
A worked audit (invented)
Northgate Cloud Backup, an invented vendor, audits the persona layer of a 600-record list built from two data sources, called A and B. It samples 40 records. The persona rule routes Operations contacts at Director level or above to a priority sequence. Every number is invented.
| Check | Result | Count |
|---|---|---|
| Title verified against a current company page | Verified 36, not verified 4 | 40 = 36 + 4 |
| Coverage, function | Value present in all sampled records | 40 of 40 |
| Coverage, level | Value present in 38; blank in 2 | 38 + 2 = 40 |
| Re-Read Test, function (36 verified) | Reader agrees 33; disagrees 3 | 33 + 3 = 36 |
| Re-Read Test, level (36 verified) | Reader agrees 29; disagrees 7 | 29 + 7 = 36 |
The Disagreement Table covers the 22 sampled people who appear in both sources.
| Source A level \ Source B level | Director and above | Manager |
|---|---|---|
| Director and above | 9 | 4 |
| Manager | 3 | 6 |
Check: 9 + 4 + 3 + 6 = 22. The sources agree on 9 + 6 = 15 and differ on 4 + 3 = 7.
Routing effect: of the seven Re-Read Test level disagreements, four cross the Director line (for example "Head of Network Operations", which one reader maps to Director and another to Manager). The three function disagreements would each send a contact to a different sequence. In this example those seven are different people, so the routes of 7 of the 36 verified contacts depend on a value the reader disputed, and 29 of 36 keep the same route under both readings.
Action yield: of the 29 stable contacts, 11 are Operations at Director level or above, and 4 of those 11 have a dated public event at their company that bears on the product. Four named actions result.
The verdict follows from the counts, not from a threshold set in advance. Northgate reads them this way.
| Field | Evidence from the audit | Verdict |
|---|---|---|
| Job function | Full coverage; reader differed on 3 of 36; the sources were not compared on function | Use for routing; send the 3 mismatches to a reader |
| Job level | 2 blanks; reader differed on 7 of 36; sources differ on 7 of 22; 4 border cases change a route | Use for ordering; route by level only after a person reads titles with Head, Lead or Principal |
| Persona name | Derived from function and level, so it inherits the level problem | Use only after the level verdict improves |
From audit to sales action
A verdict becomes an action through three rules.
- Route on what passed. Contacts whose function and level survive the Re-Read Test go into the persona sequence. Contacts that did not go to a reader, not to a sequence.
- Write to the buyer from Recorded fields only. The message cites the verified title and the dated event. It never cites a Mapped label such as "Director-level" or a Modelled importance score.
- Re-run the audit when the inputs change. A new source, a vendor taxonomy change or a new persona rule changes the results, so the audit repeats on a fresh sample.
For the account-level question that comes first, which accounts are in market, Percepture's guide to using intent data to identify sales qualified leads states that "a lead is not sales qualified because it showed activity". To choose among contact sources before auditing one, see the guide to the best B2B contact database. For the four meanings of the phrase and a field-by-field layer sheet, see contact persona-level intelligence. To keep a stored record's fields apart by tier, use the persona-level intelligence contact summary, and to cut a record to what one seller reads, the contact persona-level intelligence brief. Percepture is a company Bob Generale leads.
What to ask a data source so the audit is possible
- Does the export give each persona value with its source and the date it was assigned?
- Is the verbatim title included beside the mapped function and level, so a reader can re-read it?
- Can two sources be exported for the same people?
- Is the taxonomy documented? Demandbase, for example, documents that non-English titles are translated into English and mapped to standard roles, and that coverage "works best for commonly used titles".
- What does any confidence score cover? ZoomInfo's Contact Accuracy Score page (March 4, 2026) describes employment at the profiled company and the email address. It does not mention function, level or persona.
Lead Seeker is relevant to the first step of the audit, not the persona layer: its product pages describe a verified person, retrieved contact data, company context and the public signal that surfaced the person, re-verified when the search runs. It does not publish a persona taxonomy, a confidence score or a buying-group map. See how Lead Seeker works and Lead Compass for the account side. Lead Seeker is the product this site publishes.
Limitations
- The reader is not ground truth. The Re-Read Test measures disagreement with one person's judgement. Two readers on the border cases reduce that problem and do not remove it.
- A sample does not give a rate. At 40 records the margin can reach about 15 percentage points, so counts are reported and the list is not scored.
- The audit does not test Modelled or Guessed fields. A score such as 6sense's persona importance depends on the customer's own CRM counts and engagement history, so audit its inputs, not the score.
- Personal data rules still apply. Reading and classifying an identifiable person's title is processing of personal data. GDPR Article 14 requires information to be given to a person whose data was obtained from a source other than the person, within one month and at the first communication, and Article 21 gives a right to object, including to profiling for direct marketing.
- The example is invented. Northgate Cloud Backup, sources A and B and every count are illustrative.
Frequently Asked Questions
What is persona-level intelligence contact analysis?
It is a check of the persona fields on a set of contact records: how many carry a value, whether a reader and a second source agree with it, and how many routing decisions depend on a disputed value. The output is a verdict per field.
How big should the audit sample be?
Large enough to show which kind of problem a field has, and small enough to read by hand. At 40 records the margin is at most about 15 percentage points, so report counts and avoid rates.
Is a persona the same as a job title?
No. A title is a recorded string. A persona is a classification of that string: 6sense defines it as job function plus seniority level, and Demandbase builds one from title keywords, job functions and job levels.
Does a confidence score tell me whether the persona is right?
Not on the page examined. ZoomInfo describes its Contact Accuracy Score as the likelihood that a contact is employed at the profiled company with a correct, current email, and the page does not mention job function, level or persona.
How to Read the Tables
The Persona Layer Audit, the Re-Read Test, the Disagreement Table and the Northgate Cloud Backup example are this guide's own editorial methods, not vendor documentation or benchmarks. Labels Fact, Inference and Opinion mark what a source states, what follows from it and what the author recommends. "Not stated" is scoped to the document cited. This is an October 2026 snapshot.
Sources
- 6sense, Persona Importance Model, published August 14, 2025.
- Demandbase, Job Level and Job Function Categorization and Create and Manage Personas for Buying Groups (updated March 23, 2026).
- ZoomInfo, Overview of the Contact Accuracy Score, March 4, 2026.
- Bombora, Unlocking personas within buying groups is key to GTM success, October 20, 2025.
- Draup, Persona Analysis AI Agent.
- LINX, Persona Intelligence Analysis Is a Must-Have for Sales Teams, December 6, 2024.
- Conversion Realtor, Persona-Level Intelligence: Contact Analysis for Real Estate.
- GDPR, Article 14 and Article 21.
- Lead Seeker, Contact Persona-Level Intelligence.
- Percepture, How to Use Intent Data to Identify Sales Qualified Leads, July 6, 2026. Percepture is a company Bob Generale leads.
- Lead Seeker, How it works and Lead Compass. Lead Seeker publishes these pages.
About the Author
Bob Generale is President of Percepture. He works across SEO, AI search, digital PR, sales intelligence and AI-powered revenue systems, with a focus on connecting visibility, buyer intent and sales action.
Next Steps
Pull 40 records from your current list, verify each title against a current company page, and run the Re-Read Test on function and level before the next sequence launches. Write the verdict per field, and send only the records that passed into the persona route. To start the sample from verified people, try Lead Seeker free or see how it works.