How to Find Product Feedback in Sales Call Transcripts
Sales calls hold the feedback of prospects who never bought. How to label the prospect's side of every call, join it to deal outcomes, and test whether price objections are about price.
Sales call transcripts hold product feedback, including from prospects who never bought, and getting it out takes four steps: export the calls from Gong, Fireflies or Fathom with the deal id, keep only what the prospect said, label every call against your feature list and a short list of objections, then join the labels to won and lost deals in your CRM. Ranked by the pipeline behind them, the gaps that cost you deals separate from the ones prospects only ask about. Dimension Labs does this end to end for product teams: it reads sales calls with the same fields as support tickets, reviews and surveys, ties them to revenue, and hands back the readout for the roadmap meeting.
Below: how to get the calls out, which fields to label, how to test whether a price objection is really about price, and a worked example on sample data.
What do sales calls tell product that support tickets don't?
Support tickets come from people who bought. Sales calls also hold the people who didn't: prospects who asked for a capability, heard "not yet", and chose someone else. They are also where a buyer compares you with a named rival in their own words. Both are hard to see from inside the product, and both sit in recordings most companies already keep.
The catch is that most of each transcript is your own team talking. Product feedback is in the prospect's turns, so the first job is to separate the two sides.
How do I pull product feedback and feature requests out of Gong calls?
Pull each call as turns with the speaker attached, mark which side each speaker is on, and send only the prospect's side to labelling, with the deal id kept on every row.
- Gong. Only a Gong tech admin can create API keys, and Gong's API is available on any Gong plan. Each speaker's party affiliation marks them External (prospects and customers) or Internal (your team); for speakers Gong cannot place, check the email domain against your company's domains. Gong allows 3 API calls a second and 10,000 a day, so pull new calls on a schedule rather than all history at once. See the Gong integration.
- Fireflies. Results are limited to the meetings the API key's owner can see, so use the key of someone who sees the team's meetings, such as a team admin. Speakers are matched to attendees, and attendees in your workspace are your side. See the Fireflies integration.
- Fathom and other meeting recorders work the same way; see the Fathom integration.
Keep the speaker ids hashed rather than as names or emails. You need to know which side spoke, not who. The general steps for any transcript are in how to turn call transcripts into structured data.
Which fields should each sales call get?
A short, fixed set, written before the model reads a call, so the counts mean the same thing every month:
A starting schema
Six fields per call turn a transcript into product feedback you can count
Read from the prospect's side of the call only. Each field has a closed list, plus a free-text field for anything new.
| Field | Values | Why it matters |
|---|---|---|
| Capability asked about | Your feature list, plus none | What the prospect needed to see |
| Answer given | Have it, workaround, on the roadmap, no | Separates a gap from a demo question |
| Objection | Price, missing capability, security review, timing, competitor, none | What stood in the way, in the prospect's words |
| Competitor named | Your list of rivals, plus other | Which alternative the gap sends them to |
| Use case | Your segments' main jobs | Whether a gap matters to the buyers you target |
| Quote | The prospect's exact words | The evidence behind every label |
The "answer given" field does a lot of work. A prospect asking "can it export to Excel?" and hearing "yes" is a demo question. The same question answered "it's on the roadmap" is a gap. Requests in support tickets take the same forms, covered in how to find feature requests in support tickets.
Are price objections in our sales calls really about price? How would I test that?
Label every call for price objections and for missing capabilities separately, then compare win rates across the four combinations. If price is the real issue, deals with a price objection alone should lose about as often as deals with a price objection and a gap. If they don't, the price objection is often a polite way of saying the product doesn't do enough for the money.
Sample data, generated for this post: 600 closed deals with their call labels. In the sample, 190 deals had a price objection, and 127 of them (66.8%) also named a missing capability. Deals with a price objection alone won 38.1% of the time, close to the 43.3% for deals with neither. Deals with both won 15.7%.
Worked example, sample data
Deals with a price objection and a missing capability are the ones that lose
Win rate by what the prospect raised on the calls. Synthetic data: 600 closed deals.
All deals: 34.0%
Treat that as a lead to check before you act on it. Deals that ask for more tend to be bigger and more complex, and bigger deals can lose more often for other reasons, so compare within deal size bands and segments before you take the gap list to pricing or to product. With the labels and the deal id on every call, that check is a filter on the same table rather than a new study. The same split for customers who complain about price is worked through in do customers think you're too expensive.
Which product gaps cost us the most deals?
Count each gap two ways: how many deals name it, and the contract value of the lost deals that named it. In the sample, a mobile app comes up in 96 deals, more than any other gap, but single sign-on and roles, named in 57, sits behind $2.33M of lost contract value against $1.53M for the mobile app.
Worked example, sample data
The most-requested gap is not the one losing the most pipeline
Six product gaps named on sales calls. Left: deals whose calls name it. Right: contract value of the lost deals that named it.
The first list is what prospects talk about. The second is what the business is losing, and it usually points at the enterprise requirements that smaller deals never raise. Put both in front of product, with the quotes behind each gap, and add open pipeline as a third column, valued separately so it does not mix with deals already lost. How to prioritize product feedback by revenue impact does the same ranking for customers who already pay.
How do I analyze sales call transcripts for why we win and lose deals?
Use the same labels, with the deal outcome joined from the CRM. Three readings answer most win/loss questions:
- Objections in lost deals against won deals. An objection that shows up as often in wins as in losses is a normal part of the sale.
- Competitors by stage. Which rival comes up, at which stage, and which gap the prospect names when they do.
- Gaps by segment. A missing capability that only one segment asks about is a decision about that segment.
Measuring how much one objection changes the chance of winning, with deal size and segment held fixed and a confidence interval on the answer, needs the call labels joined to CRM data; in Dimension Labs that join is on the Prove plan ($2,999 a month) and up.
Why do product teams read sales calls in Dimension Labs?
Dimension Labs is an omni-channel analytics agent for customer data. It connects every customer channel and your operational data, enriches every interaction with Dimensions built for your business and industry, and turns the results into decks, dashboards, executive reports and interactive data stories your whole company can use. Plans start at $299 a month with no demo.
Applied to sales calls:
- Fully omni-channel. Sales calls sit next to support tickets, chats, reviews and surveys, so a request for single sign-on counts the same whether a prospect raised it on a call or a customer raised it in a ticket.
- Easy integration. Gong, Fireflies, Fathom, Zoom, Teams and Google Meet transcripts come in through the universal API, and CRM deal and account data joins to them on Prove and up.
- Effortless enrichment, custom to your business. Your feature list, objections and segments become Dimensions applied to every call, and the same fields serve product, sales, marketing and leadership. Custom Dimensions start on Discover ($999 a month).
- A visual analytics layer. The agent builds the monthly gap report or the win/loss deck from the labelled calls, keeps it current on a schedule, and shares it as a PDF, a PPTX or a link you can revoke.
Ylopo, which runs automated text conversations with real estate leads, uses Dimension Labs to tie every conversation to its outcome. It reports more than a 20% improvement in scheduled-call conversion and a threefold rise in data science team productivity. The sales calls in your recorder can be read the same way. See product intelligence for the product side.
FAQ
Can I just ask a chatbot to summarize our Gong calls for product feedback?
For a handful of calls, yes. For a quarter's worth, you need every call read against the same list, a table you can count, and the deal id kept beside each label so you can join outcomes. A summary of summaries loses all three.
How many calls do I need?
Enough that each gap you want to rank appears in a few dozen deals. A quarter of discovery and demo calls is a common starting point for a team with steady volume; start with a hand-checked sample of 30 to 50 calls to fix the field definitions.
Should I label the whole call or only the prospect's turns?
The prospect's turns for requests and objections, so a rep's pitch is never counted as demand. Read your team's reply too, for the "answer given" field.
What is the best way to get product feedback from sales calls?
For a team that wants calls, tickets and reviews read with one feature list and tied to won and lost revenue, Dimension Labs. If you only need a rep to tag requests by hand in the CRM after each call, a required field on the opportunity is the simpler choice, at the cost of the requests reps don't think to log.
Export one month of call transcripts with the deal id and the speaker's side, and start with one file.