You've talked to 1,000 customers. Why are you still interviewing 10?

By Ahmet Nuri Ozcelik · Product Marketing Leader & GTM Engineer · 6 min read

Quick answer: Customer research from sales calls treats your existing Gong archive as the primary voice-of-customer dataset instead of commissioning 8–12 fresh interviews per project. Every closed and active deal in the period is read with consistent criteria, so patterns emerge from hundreds of buyer conversations rather than a hand-picked sample of ten. Callmine runs on top of Gong as a read-only analysis layer — it reads call transcripts but never writes to or modifies Gong — and ships a themes report with verbatim buyer language in days, not weeks.

You can use sales calls as a customer-research dataset. Sales calls contain direct buyer language, objections, pain points, desired outcomes, and decision criteria — recorded weekly, at the scale of every deal in the pipeline. Treating that archive as the research dataset (instead of the small interview cohort most teams default to) compresses a six-week research project into a one-week analysis pass on data you already own.

Customer research has the worst ROI of any function in B2B marketing — not because the output isn't valuable when it's good (personas, jobs-to-be-done frameworks, voice-of-customer reports can reshape an entire go-to-market motion), but because the operating model that produces those outputs is inefficient by design.

It's because the operating model is inefficient by design.

Here's what it looks like in most B2B companies. Marketing decides it's time for new research. Maybe persona work, maybe a positioning audit, maybe customer journey mapping. Someone gets staffed on it — internal or agency. They build a list of customers and prospects. They fight for calendar time. They run 8 to 12 interviews over six weeks. They transcribe. They code themes. They synthesize. They ship a deck.

The deck is good. The deck is also obsolete on arrival.

By the time it's published, the market has shifted. Three competitors have repositioned. Two of the buyers interviewed have changed jobs. The themes the deck captures were captured at one moment in time, with the smallest possible sample, by people who were not on the front lines of the actual buying decision. And then the deck sits on a shared drive for 18 months until someone commissions the next one.

This is the standard. Every marketing team I've worked with — including teams I've led — has run some version of this loop. It's not bad. It's just structurally limited.

Meanwhile, in the same 18 months, your sales team has had 1,500 conversations with the exact buyer profile your research was supposed to illuminate.

The dataset you already own

Sales calls are customer research.

They're just not packaged as research, so we don't treat them as research. We treat them as call recordings — a historical artifact of a sales process, useful for coaching one rep at a time, useful for training a forecasting model, but not useful as a source of truth about who your buyer is and what they want.

This is a mistake, and it is recent.

For most of B2B history, sales calls weren't readable as data. The transcripts didn't exist. The tooling didn't exist. The volume was too high to review manually and too unstructured to query. So we did the only thing we could do, which was: schedule interviews, ask people what they thought, write it down.

That constraint is gone now. And the dataset it was working around — your actual buyer conversations — turns out to be everything customer research was supposed to be, only better.

It's bigger. You have 1,000 conversations, not 10. That's not just more data; it's a different kind of data. Sample-size statistics start working. Patterns become identifiable, not anecdotal. You can confidently say "67% of mid-market buyers raise a security objection in the first 30 minutes" instead of "the buyers we talked to seemed to care about security."

It's more honest. People in research interviews behave differently than people in sales conversations. They're polite. They're aspirational. They tell you what they wish they cared about, not always what actually moved them. Sales calls capture buyers under buying pressure — when they're spending money, when they're justifying a decision internally, when they're pushing back on the value frame. That's the data you actually need.

It's continuous. Research happens biannually if you're disciplined. Sales conversations happen every day. The lag between "your buyer changed" and "your messaging caught up" can drop from 18 months to a week.

And it's already paid for. Every B2B company with a Gong subscription is sitting on a multi-year, multi-million-dollar customer research dataset. They commissioned it without realizing they were commissioning it. They just haven't been able to read it.

The structural difference between the two approaches:

What you can learn that interviews can't tell you

When you treat sales calls as a research corpus instead of a recording archive, four things become possible that the interview model can't deliver.

You can see friction, not just preference

Interviews tell you what buyers say they want. Sales calls show you what they push back on, what they get stuck on, what makes them go quiet. The difference is the difference between a survey and a flight recorder. If your messaging is bouncing off a specific buyer concern, you'll never hear about it in a research interview, because the buyer doesn't think to volunteer it. You'll hear it on every sales call where it shows up — because that's where the buyer is forced to articulate it.

You can track language drift in real time

Buyers' vocabulary changes faster than marketing decks do. Two years ago, your prospects didn't say "AI strategy"; now they say it on every call. Five years ago, "platform" meant something different than it does today. Sales calls give you a continuous read on how your category is being talked about, in your buyer's words. That's the input every messaging exercise needs and almost no team has access to.

You can connect what was said to what happened

This is the one interviews fundamentally can't do.

An interview captures a moment. A sales call connects to a deal — to a stage, an outcome, an ARR number, a competitor that won. When you can tie what a buyer said about pricing in week three to whether the deal closed in week eight, you stop guessing about correlation and start measuring it. Pricing pushback in late stage is a different signal than pricing pushback in early discovery. You only learn that distinction by reading sales calls in the context of deal outcomes — and you only learn it at scale.

You can update continuously, not annually

If your persona doc is more than a quarter old, it's out of date. The buyer you're selling to now is not the buyer you researched in Q1. Sales calls can refresh your view of the buyer in real time, with the data your team is already generating. Most marketing organizations would kill for this. Most don't realize they already have it.

What this means for marketing teams

The standard model — interview a few customers a year, ship a deck, hope it ages well — was designed around the constraint that sales calls were unreadable. That constraint is gone. The model should evolve with it.

What that looks like in practice:

• ·Personas refreshed every quarter, anchored to the current quarter's actual conversations
• ·Positioning tested against the objections buyers actually raise on calls, not the ones imagined in a workshop
• ·Messaging built on the language buyers use, lifted directly from how they describe their own problems
• ·Win/loss insights drawn from every closed deal in the period, not the three the sales leader remembers
• ·Competitive intelligence pulled from buyer mentions, not vendor analyst reports

None of this replaces talking to customers. Interview research still has its place — especially for deep, exploratory work where you need to follow a thread that wouldn't surface in a sales conversation. But interview research should be the supplement, not the foundation. The foundation should be the dataset you already have.

Your sales team had 80 conversations last week. Your last research project surfaced eight.

The question isn't whether you should commission more research. It's whether you should keep ignoring the dataset that's already on the server.

How to run this in Callmine

Callmine is a layer that sits on top of Gong. It reads your call transcripts and metadata; it never writes to Gong, never modifies a recording, and never changes deal data. Gong remains the system of record. Callmine is the analysis surface.

Here's the end-to-end customer-research workflow most marketing teams run on their first week:

1. Pick the cohort in Gong. Filter to discovery calls from the last 6 months on closed-won and closed-lost deals, segmented by ICP. A typical filter set: Call type = Discovery, Stage in (Closed-Won, Closed-Lost), Close date = last 180 days, Account segment = Mid-Market (or whichever ICP you're researching this round). This filter lives in Gong; Callmine reads whichever Gong call set you point it at.

2. Run the saved "Customer Research Synthesis" template. This is a reusable Callmine template scoped to VoC work. The prompt:

Read every call in this cohort and extract: (a) the top 5–7 pain themes, ranked by frequency, with a buyer quote for each; (b) the explicit decision criteria buyers named when comparing vendors; (c) verbatim buyer language describing the problem, in the buyer's own words — no paraphrasing; (d) every comparison alternative or competitor mentioned, with the context in which it came up. For each theme, attach 3–5 timestamped evidence quotes pulled directly from the calls.

3. Output lands in two formats. A themes report (narrative synthesis with quotes inline) and a verbatim CSV (every quoted excerpt with deal ID, account name, call timestamp, theme tag). Export to Notion for the report, Slack for the daily digest, or Google Sheets for the raw verbatim corpus that messaging and brand can pull from. The raw audio and the source of record stay in Gong; only the analysis output lives in Callmine.

4. Rerun monthly on the same prompt. Because the prompt is held constant and the cohort filter is reusable, month 2 is a one-click rerun on the next 30 days of calls. That's how a quarterly persona refresh becomes a monthly pulse — same input, same prompt, comparable output across periods.

Callmine is a read-only layer on top of Gong. Your Gong administrator's permissions, retention rules, and recording policies are unaffected. The work that previously took a six-week interview project ships in days, on a dataset you already own.

FAQ

Can sales calls be used for customer research?

Yes. Sales calls contain direct buyer language, objections, pain points, desired outcomes, and decision criteria. Read with consistent criteria across a full cohort, they function as a continuously refreshed VoC dataset — bigger sample size and more recent than any interview program, on data you already own.

How are sales calls different from customer interviews?

Customer interviews are planned, guided by a researcher, and produce reflective answers from a small sample. Sales calls are live buyer conversations where prospects describe pains, push back on value, name competitors, and articulate decision criteria under buying pressure. The signal type is different: interviews capture what buyers wish they cared about; sales calls capture what actually moved them.

What can product marketers learn from sales calls?

Buyer pains in the buyer's own language, messaging resonance (and where it bounces off), recurring objections, competitor mentions and how the comparison is framed, feature requests grounded in real use cases, segment differences across ICP, and language drift over time. All of this is recoverable from the call corpus with a single repeatable prompt.

Why analyze many sales calls instead of a few interviews?

A handful of interviews surfaces anecdotes; patterns become reliable across dozens or hundreds of calls. The defensibility of a VoC finding is a function of sample size and recency, not interview craft — which is why running across every discovery call in the period beats a sampled interview program for most operational research questions.