Buyer Persona Research: Why Your Sales Calls Are the Dataset

By Ahmet Ozcelik, Product Marketing Leader & GTM Engineer — Published 2026-07-10

Quick answer: Buyer persona research is the process of gathering real, evidence-backed data on your ideal customer's goals, language, and buying questions — traditionally done through 5-10 scheduled interviews per persona. The same insight already exists at much larger scale inside your Gong-recorded discovery and demo calls, where buyers answer these exact questions unprompted and in their own words. Clustering that call language by job title turns hundreds of existing sales conversations into a persona dataset the standard interview-only approach can't match for sample size or authenticity.

Most buyer persona research projects start the same way: book eight interviews, offer a gift card, hope enough people show up. Here's the part nobody says out loud — you probably already have 400 of those interviews recorded, unread, sitting in Gong.

What Buyer Persona Research Actually Requires

Buyer persona research is the process of building a research-based profile of a target buyer's goals, challenges, and decision criteria — not a demographic slide with a stock photo and a made-up name. Most "personas" built from a marketing workshop are guesses dressed up as research. A real persona traces back to evidence: something a buyer actually said, a pattern repeated across accounts, a decision criterion you can point to and defend.

The standard methodology combines two data types. Quantitative inputs — firmographic and CRM data, deal size, win rates by segment — tell you who your buyers are structurally. Qualitative inputs — interviews, surveys, call notes — tell you how they think and talk. Both matter, but the qualitative layer is where most persona docs fall apart, because it's expensive to gather well.

What teams actually need out of this process isn't a framework slide. It's real buyer language — the exact phrase a VP of RevOps uses to describe a broken handoff, not a paraphrased summary a product marketer wrote after the fact. 71% of companies that exceed revenue goals have documented buyer personas, which tells you the artifact matters. It says nothing about whether the underlying research was any good.

The Interview Bottleneck: Why Most Persona Docs Rest on 5-8 Conversations

The accepted wisdom, repeated across nearly every persona guide, is that you need somewhere between five and ten interviews per persona. Adele Revella, founder of the Buyer Persona Institute, has said the same thing from decades of doing this work — in one interview she put it plainly: we learn everything we're going to learn after about ten interviews around a particular buying decision with a well-focused segment.

That's sound advice for the interview method itself, and it assumes a single buyer per persona. Most B2B deals aren't decided by one person — B2B buying groups typically run five to sixteen people across multiple functions, which means the ten interviews you managed to schedule may still miss entire roles that influence the purchase.

The bigger problem is what happens between "we should do ten interviews" and "we actually have ten scheduled." Recruiting real buyers — not friendly customers who'll say yes to anything — means chasing CS for warm intros, offering incentives, and hoping enough people show up with something useful to say. Most persona refreshes stall at three or four completed conversations and ship anyway with a thin evidence base.

Small samples carry a second risk: one articulate buyer can define the entire persona. If your five interviews happen to include one unusually vocal Director of Sales Ops who hates a specific competitor, that opinion ends up baked into messaging for an entire segment — not because it's representative, but because it's the loudest data point you have.

None of this is a knock on the interview method. It's a knock on treating it as the only method, when a much larger, already-recorded dataset is sitting one login away.

The Overlooked Dataset: Your Discovery and Demo Calls Are Already Persona Interviews

Every discovery call already asks the questions a persona interview would ask. What are you trying to accomplish this quarter? What's not working today? What would need to be true for this to be worth changing? Reps ask versions of these questions constantly, and buyers answer without the artificiality of "this is a research interview, please be candid" framing — they're mid-evaluation, motivated, and talking in their own words because they want the deal to move forward.

That's a structurally different, arguably better, research condition than a scheduled interview. A buyer being pitched to explains their problem the way they actually think about it, with real stakes attached. A buyer interviewed for research six months later reconstructs a memory, softened and generalized in hindsight.

The dataset exists at a scale interview-only research can't touch — and it's almost entirely unmined. Discera's internal usage data shows 97% of Gong calls go unread after the deal they were recorded for closes or dies. That's true of any team relying on Gong's native call view, which is built for reviewing one call at a time, not clustering language across hundreds. Worth reading more broadly on mining sales calls for customer research if this is the first time you're thinking about calls as a research corpus rather than a coaching tool.

This isn't limited to sales calls, either. Customer success and renewal calls carry the same signal for expansion and retention personas — what a buyer says six months post-close about whether the product delivered is exactly the evidence a "renewal decision-maker" persona needs, sitting in the same Gong workspace as your discovery calls.

What Does Job-Title-Level Call Data Reveal That Interviews Miss?

Job title clustering — grouping call language by the buyer's actual title rather than by deal or account — surfaces distinctions a small interview sample usually can't. A VP of RevOps and a Director of Sales Ops might buy the same category of tool, but describe the same underlying problem with different words and different success metrics, because they're measured on different things. Ten interviews rarely give you enough of each title to notice the pattern. Hundreds of calls, clustered by title, do.

Scale also changes what "recurring" means. In a five-interview sample, a question that comes up twice looks important. Across 300 discovery calls filtered to a single job title, you can rank questions by actual frequency — the third-most-common question a Director of Marketing Ops asks isn't a guess, it's a count.

Objections behave differently in live calls than in interviews, too. A buyer on a real sales call raises a concern because they need it resolved before they'll move forward — pricing, integration risk, internal buy-in. The same buyer, weeks later in a research interview, will often soften or omit that objection, because the stakes are gone. Live calls capture friction; interviews capture retrospection.

How to Extract Buyer Persona Research From Gong Calls at Scale

If your team is on Gong, this stops being a methodology question and becomes a workflow question. Here's how I'd set it up in Discera.

Start with filters, not a blank prompt. Pull Gong calls by call type — Discovery and Demo — over the trailing two quarters, enriched with the HubSpot contact job-title field so every call is tagged to a real title rather than a generic "buyer" bucket. If you haven't set this up before, segmenting Gong calls by deal stage covers the same filter logic — call type, date range, CRM enrichment.

Then write the prompt to cluster, not just summarize. Something close to what I use:

"For each call, identify the buyer's job title from the HubSpot contact record and extract: (1) the specific questions they asked in their own words, (2) the outcome or metric they said they're measured on, (3) the exact phrase they used to describe their current problem. Group all findings by job title and return the five most frequent needs and questions per persona, with verbatim supporting quotes."

Specificity matters more than people expect here — a vague prompt gets you a vague summary. Writing effective Gong call analysis prompts covers the gap between prompts that produce generic themes and prompts that produce usable, quotable output.

Discera runs that prompt across the filtered set in parallel — up to 30 concurrent jobs — so a batch of a thousand calls typically comes back in around five minutes rather than the days a manual read-through takes. The output ships as a DOCX export for the persona draft, plus a scheduled monthly summary posted to the #product-marketing Slack channel, so the dataset stays current instead of going stale the moment the doc is finished. Saved templates for voice-of-customer research and messaging validation cover adjacent work with this same call corpus.

Worth being direct: Discera reads and analyzes Gong data, it doesn't write back to or modify anything in your workspace. It's an analysis layer on top of calls you've already recorded, not a replacement for Gong, and Gong is required at every tier.

Where Interviews Still Matter: Combining Call-Mined Data With Direct Conversations

Call-mined persona data has a real limitation, worth stating plainly rather than overselling: it reflects what reps happened to ask and buyers happened to volunteer, not a structured interview guide designed to cover every angle. If reps never ask about budget approval process, that topic won't show up no matter how many calls you analyze. It's the same structural gap that shows up in win/loss analysis from Gong calls — the closed-lost corpus is thin because reps talk less on calls they're losing, so analysis can't fully substitute for outreach there either.

Interviews also do something calls structurally can't: post-decision reflection. A buyer three months into using a product, or one who chose a competitor, can answer "why did you really choose us" in a way no in-flight sales call captures, because the decision is already made.

The practical answer isn't choosing one method over the other. Use call-mined data to build the first draft — goals, recurring questions, objections, exact phrasing — then use that draft to decide which two or three interviews are actually worth scheduling. You'll ask sharper questions because you're validating specific hypotheses instead of starting from a blank page.

Turning Verbatim Call Language Into a Persona Document

Once you have clustered call data, translating it into a persona document is mostly mapping, not writing. Take the top questions and needs per job title and slot them into the standard fields — goals, day-in-the-life context, objections, decision criteria — the structure any persona template already uses.

The one habit worth protecting: keep quotes verbatim, not paraphrased. If a buyer says their team is "drowning in manual QBR prep," that phrase belongs in the persona document exactly as spoken, not smoothed into "faces operational inefficiencies." The paraphrase is marketing-speak nobody outside your building uses. The verbatim quote is language your next piece of messaging can borrow directly, because it's already proven to resonate with a real buyer under real conditions.

Re-run the same clustering prompt quarterly rather than treating the persona as a one-time project. Buyer language shifts as your product, competitors, and market change, and a persona document built once and left alone for two years is functionally a demographic slide again — accurate the day it was written, stale by the time anyone uses it.

FAQ

What is buyer persona research?

Buyer persona research is the process of gathering evidence-backed data on a target buyer's goals, challenges, and buying language, usually through interviews, surveys, and CRM analysis, so a persona document reflects how real buyers talk rather than internal assumptions.

How many interviews do you need to build a buyer persona?

Standard methodology, including guidance from the Buyer Persona Institute, points to roughly eight to ten interviews per persona once you're working with a reasonably focused segment. Some frameworks suggest starting with a larger pool and narrowing down as patterns repeat.

Can you build buyer personas without customer interviews?

You can build a strong first draft without scheduling new interviews by mining language already in recorded discovery, demo, and customer success calls. Still validate with a small number of direct conversations, particularly for closed-lost buyers and post-decision reflection that calls in progress won't capture.

How often should buyer personas be updated?

Most teams should revisit personas at least every two quarters, and sooner after a pricing change, a new competitor, or a shift in ideal customer profile. Buyer language typically drifts faster than static persona documents get refreshed.

What's the difference between a buyer persona and a negative persona?

A buyer persona describes who you want to sell to and why they buy, while a negative persona describes who looks like a fit on paper but consistently churns, stalls, or never closes. Documenting both keeps teams from chasing accounts that resemble your ICP without actually converting.

If your team already runs on Gong, the fastest way to test this is to run the clustering prompt above against your own discovery and demo calls from the last two quarters — you'll likely find your persona doc is thinner than your call library. Start a free trial at discera.ai.