Personas

Personas

What this skill does

Traditional personas degrade into assumptions. They start with strong intent — a deck neatly organized by role and pain — and within 6 months they're stale slides nobody opens. When personas drift, everything downstream (messaging, GTM, product focus) starts to misfire.

This skill builds personas as living profiles — synthesized from multi-source data (CRM, calls, reviews, community signals, behavioral telemetry, qualitative interviews) rather than assumption. The output is structured so personas can be refreshed continuously, not rebuilt from scratch every year.

When to invoke

• The user has no personas, or has personas that are >6 months old.
• The user is starting a positioning, messaging, or launch project and references stale personas.
• The user mentions a new buyer type emerging in deals (e.g., procurement showing up in cycles that used to be CMO-led).
• The user has a pile of research (interview transcripts, sales call notes, reviews, survey data) and needs synthesis.

Prerequisites

1. Read pmm-context.md first when available. If it is missing, offer to run pmm-context. If unavailable or declined, collect a minimum brief covering product, ICP, suspected persona segments, available evidence, buying context, and known constraints. Mark assumptions and continue.
2. Ask the user what inputs they can provide. Personas grounded in evidence beat personas built from assumption. Useful inputs:

• Sales call recordings or transcripts (Gong, Chorus exports)
• Customer interview transcripts
• CRM data — closed-won and closed-lost reasons, deal stages, time-to-close
• Product usage data
• Review platform data (G2, TrustRadius, Capterra)
• Community discussion (Reddit, Discord, LinkedIn)
• Support ticket themes
• NPS / survey data
• Win/loss interviews

1. If user has limited inputs, ask which 1-2 personas to focus on first. Better to do 2 personas well than 5 personas thinly.

The four-question structure

Every persona must answer four questions, each with AI-augmented evidence:

| Question | Traditional approach | AI-enhanced approach | |---|---|---| | Who they are | Static demographics/firmographics from research | Real-time CRM enrichment + clustering on closed-won deals | | What they feel | Interview quotes (small N) | Sentiment analysis across reviews, support tickets, community | | Why they buy | Hypothesized motivations | Pattern detection on call transcripts surfacing real triggers | | How they buy | Generic journey map | AI journey mapping from CRM stages + touchpoint data |

Workflow

Step 1 — Scope

Pick 1-3 personas to build or refresh. Resist the urge to do all 7. Ask the user: "If you only had time for one, which one would change the most decisions downstream?"

Step 2 — Gather core data

For each persona, collect inputs across two dimensions:

Quantitative (firmographic + behavioral):

• Closed-won deal characteristics (company size, industry, role of primary contact, deal cycle length, ACV)
• Closed-lost reasons by segment
• Product usage patterns (which features they use, frequency, drop-off points)
• Acquisition channels (where they came from)

Qualitative (motivations + language):

• Interview transcripts (3-10 conversations minimum; quality beats quantity)
• Sales call themes (top objections, top questions, top "why now" triggers)
• Review/community quotes in their own words
• Support themes ("they keep asking us X")

Step 3 — Synthesize

For each persona, generate the four-question profile. Use the template in TEMPLATE.md.

Synthesis rules:

• Quote, don't paraphrase. Use the persona's actual words wherever possible. "Saves time" is paraphrase. "I used to spend my Sunday nights pulling this together for Monday standups" is voice.
• Cite sources. Every claim → a source (call ID, review ID, survey question). Without sources, the persona is fiction within 6 months.
• Cluster, don't average. If you see two distinct sub-personas, split. Personas built on averages describe nobody.
• Tag confidence. Mark each claim High / Medium / Low confidence so users know what to test.

Step 4 — Pressure-test

Before publishing, run these 5 tests on each persona:

1. The Sales Test. Show it to 2 sales reps who close this segment. Do they recognize it? Does it predict objections they hear?
2. The Customer Test. Show it to 2 customers in this segment (anonymized). Do they see themselves?
3. The Disqualification Test. Can a rep use this to disqualify a bad-fit prospect? If not, the persona is too vague.
4. The Trigger Test. Does the persona name a specific trigger event (not "they want to grow") that makes them start looking?
5. The Drift Test. Is this materially different from the persona we had 12 months ago? If not, did anything actually change in the market — or is it just stale?

Step 5 — Set up monitoring

Personas should refresh, not rebuild. Define:

• Which signals would tell us this persona is shifting? (e.g., role title changing in CRM data, sentiment trending negative on a specific theme)
• Who owns the monitoring? (PMM, RevOps, customer marketing)
• Cadence for review (quarterly is typical; monthly for fast-moving segments)

Outputs

A. Persona profile (one per persona)

See TEMPLATE.md. Structured around the four questions, with sources cited and confidence tags.

B. Day-in-the-life narrative

A 200-300 word narrative of a typical day for this persona. Names a problem they encounter, the tools they use to solve it, the people they talk to about it, the moment they'd consider buying. Used in sales training and content.

C. Persona-to-channel map

A table showing where each persona shows up (watering holes), what content resonates, what objections come up most. Used in demand gen and content strategy.

D. Monitoring plan

Which signals to watch per persona, who owns review, refresh cadence.

Application guide

Once personas exist, downstream applications:

• Messaging: adaptive-messaging skill uses persona language for talk tracks
• Positioning: positioning skill uses persona pain to identify best-fit segments
• Launch: launch skill maps tier and channel selection to persona behavior
• Sales enablement: persona-specific battlecards, email templates, demo flows
• Personalization: dynamic content for landing pages, ads, lifecycle email

Quality bar

• No fictional names with stock photos. "Marketing Mary, 34, drinks lattes" is decoration, not insight. Skip the stock-photo persona.
• Evidence per claim. Every "they care about X" line is linked to a source.
• Disqualifiers are mandatory. A persona must include who is NOT a fit, not just who is.
• Language is the persona's, not the model's. Match their vocabulary. If they say "stack," don't write "technology portfolio."

Anti-patterns to refuse

• Building personas with no evidence beyond user assumption (push back, ask for inputs)
• Creating 5+ personas at once (force prioritization)
• Personas that are demographic profiles instead of motivational profiles
• Personas without a named trigger event
• Personas without disqualifiers

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: dirknicol
• Source: dirknicol/pmm-skills
• License: MIT

Install and usage instructions live in the source repository linked above.