Sales call recordings and transcripts linked to deal outcomes

A sales call transcript dataset is a set of recorded sales conversations, such as discovery calls, product demos and negotiations, with speaker-attributed transcripts, each linked to the CRM opportunity it belonged to and that deal's eventual outcome. SourceX sources these recordings from companies that capture calls in conversation intelligence tools such as Gong, Chorus or Zoom Revenue Accelerator, reviews recording disclosures and prospect confidentiality, and pseudonymizes buyers, sellers and companies before delivery.

Request this dataset Email the data team

Dataset manifest

Sourced to your spec

What it is: Recorded discovery calls, demos and negotiations, linked to the CRM deal and its outcome
Typical systems: Gong, Chorus by ZoomInfo, Zoom Revenue Accelerator, Salesloft, Outreach, Clari Copilot
Typical history: Varies by partner; begins when the team started recording calls
Modality: Meeting audio, transcripts and deal metadata; video and screen share where agreed
Delivery formats: Agreed per order; audio, JSON transcripts and a call-to-deal linkage table
Preparation: People and company names pseudonymized in speaker labels, transcripts and follow-ups
Licensing: Permitted use agreed per license; deals under restrictive prospect NDAs excluded
Availability: Depends on recorded calls with clear disclosures and deal links; not guaranteed

Request this dataset

What a delivery contains

Fields vary by source system and are fixed per order. A typical delivery includes:

Field	Type	What it holds
call_id	string	Pseudonymous identifier for the recorded meeting or phone call.
opportunity_id	string	The CRM deal the call was linked to, matching any pipeline history in the same delivery.
started_at	timestamp	Meeting start in UTC.
sequence	object	The call's position among the deal's recorded calls, with the IDs of earlier calls.
call_type	enum	Prospecting, discovery, demo, technical deep dive, pricing, negotiation or executive meeting.
deal_at_call	object	Stage, deal age, amount band and forecast category as they stood when the call took place.
participants	array	Pseudonymous people with their side (buyer, seller or partner), role on the deal and title band.
media	object	Medium (video meeting or phone), audio reference and, where in scope, video and screen-share files.
segments	array	Transcript turns attributed to participants, with start and end times in seconds.
trackers	array	Topics the recording tool flagged, such as pricing, competitors or security, with timestamps and detection method.
interaction_stats	object	Talk-to-listen ratio, longest monologue and question counts as computed by the recording tool.
follow_up	object	Next steps the rep logged and, where in scope, the follow-up email sent after the call.
coaching	array	Manager comments and scorecards pinned to moments in the call.
deal_outcome	object	The deal's next stage change after the call, and its final result, close date and loss reason.

Example record

{
  "call_id": "mtg_2d7a90",
  "opportunity_id": "opp_7e15c3",
  "started_at": "2024-05-14T16:00:12Z",
  "call_type": "demo",
  "sequence": { "call_number": 3, "prior_calls": ["mtg_1a04f2", "mtg_1c88d0"] },
  "deal_at_call": { "stage": "solution_validation", "deal_age_days": 23,
                    "amount_band_usd": "50k-100k", "forecast_category": "pipeline" },
  "participants": [
    { "id": "p_s01", "side": "seller", "role": "account_executive" },
    { "id": "p_s02", "side": "seller", "role": "solutions_engineer" },
    { "id": "p_b01", "side": "buyer", "role": "champion", "title_band": "director" },
    { "id": "p_b02", "side": "buyer", "role": "technical_evaluator", "title_band": "manager" }
  ],
  "media": { "medium": "video_meeting", "audio": "audio/mtg_2d7a90.m4a", "duration_s": 2710, "video": "not_in_scope" },
  "segments": [
    { "start": 612.4, "end": 631.0, "speaker": "p_b02",
      "text": "Before we go further, does this work with our own identity provider? Security will ask on day one." },
    { "start": 631.8, "end": 655.2, "speaker": "p_s02",
      "text": "It does, SAML and SCIM. Let me show you the admin side now." },
    { "start": 1904.6, "end": 1921.3, "speaker": "p_b01",
      "text": "Honestly, [COMPETITOR_A] came in cheaper. I need something to take to [PERSON_NAME]." },
    { "start": 1922.0, "end": 1950.7, "speaker": "p_s01",
      "text": "Fair. Let's build the cost case together. What does the manual process cost your team today?" }
  ],
  "trackers": [
    { "t": 612.4, "topic": "security_sso", "method": "keyword" },
    { "t": 1904.6, "topic": "competitor_pricing", "method": "keyword" }
  ],
  "interaction_stats": { "seller_talk_ratio": 0.58, "longest_seller_monologue_s": 142, "buyer_questions": 9 },
  "follow_up": { "next_steps": "Business case session next week; [CONTACT_NAME] to send security questionnaire." },
  "coaching": [ { "t": 1922.0, "author_role": "sales_manager",
                  "text": "Good pivot from price to the cost of doing nothing." } ],
  "deal_outcome": { "next_stage": "proposal", "days_to_next_stage": 6,
                    "result": "closed_won", "days_from_call_to_close": 71 }
}

Synthetic record for illustration. Field names, structure and format are agreed per order.

What AI teams use it for

Train AI sellers on discovery and objection handling

Real buyers raise budget, timing, security and competitor objections in their own words. Knowing which deals went on to close separates responses that worked from ones that only sounded good.

Score deal risk from what was said

Signals such as a missing economic buyer, no agreed next step or a late competitor mention can be tested against how each deal actually ended.

Draft follow-ups and CRM updates

Calls paired with the next steps and follow-up email the rep wrote afterwards teach a model to produce both, with the human version as the reference.

Evaluate sales assistants on held-out calls

Stop a recorded call where an objection lands, ask an agent for the next reply, and compare it with what the seller said and how the deal went.

Model coaching feedback

Manager comments and call scorecards are expert judgments of selling skill, attached to the exact moment they refer to.

Use-case guides: Sales agents, Voice agents and speech models

What makes this data valuable

Verified deal links

Each call is tied to the right opportunity, with the deal's stage at the time of the call.

Whole deal sequences

Every recorded call on a deal, from first meeting to signature, rather than isolated calls.

Buyer-side roles

Participants labeled as champion, economic buyer, technical evaluator or procurement, not just "external".

Late-stage coverage

Pricing, negotiation and executive calls, which some teams choose not to record.

Written follow-through

Follow-up emails, mutual action plans and logged next steps show what each call committed both sides to.

Coaching annotations

Timestamped manager comments and scorecards add expert judgment to individual moments.

The deal, not the call, is the unit

A single sales call rarely decides a deal. Between first discovery and signature there are usually more meetings, a stream of emails, a security review and a procurement process, and most of that never reaches a recording. Treating every call as an independent example labeled with the deal's outcome credits or blames one conversation for everything else that happened. The more useful unit is the deal: its recorded calls in order, the stage changes between them and the final result, so a model can see how buyer language shifts as a deal advances or stalls.

Coverage decides what an outcome label can mean. Some teams record discovery and demos consistently but switch recording off for pricing and negotiation, some prospects decline to be recorded, and retention settings may have deleted older calls while their deals remain in the CRM. Map which calls exist for each deal before deciding what to predict from them.

Leakage and confounds in conversation data

Late calls give the answer away. A call where the buyer mentions the signed order form, or explains why they chose another vendor, is good material for summarization but leaks the label in any prediction task. Cut each deal at a fixed point, such as a stage or a number of days before close, and use only the calls before it.

Sellers are a second confound. Strong reps run better calls and may also hold the better territories, so a model can learn who is speaking rather than what works. Seller pseudonyms that stay consistent across calls let you control for this and hold out by seller as well as by time.

Treat the recording tool's own outputs, such as topic trackers and talk ratios, as machine annotations rather than ground truth. Their definitions are set by the vendor and by the partner's admins, and they shift when a tracker's keywords are edited, so ask for the tracker definitions and their change dates along with the data.

What to check before licensing

Check how participants were told about recording, such as a visible recording bot, a spoken notice or the calendar invite. Prospects never accepted the seller's customer terms, so this may be the only notice they saw, and calls with participants in all-party-consent jurisdictions need particular care.
Ask whether mutual NDAs were signed with prospects during evaluations. Budgets, architecture and internal problems a buyer described on a call can fall under one, and those deals may need to be excluded.
Review video and screen shares separately from audio. Demo environments can show other customers' data, and a prospect's shared screen can expose their internal systems.
Spot-check call-to-deal links on the sample. Recording tools usually match calls to opportunities using calendar data and participants' email domains, which goes wrong when an account has several open deals or a reseller joins the call.
Map which call types were recorded across the deal cycle. Reps may stop recording for negotiation and some prospects decline, so coverage is rarely even.
Confirm that pseudonymization covers speaker labels taken from calendar invites, names spoken aloud, company and product names, and signatures in follow-up emails.
Agree how the partner's own pricing, discount approvals and roadmap commitments mentioned on calls are handled, since it may treat them as confidential.

How licensing works through SourceX

1
Define
Send the domain, modality, volume, format, timeline and permitted use you need.
2
Source
SourceX identifies businesses that hold matching data and are open to licensing it.
3
Qualify
Fit, rights and quality are checked, and you review samples before committing.
4
License
Scope, permitted use, exclusivity, price and obligations are agreed in writing.
5
Deliver
Approved data is prepared, de-identified where required and transferred securely.

Questions buyers ask

Are calls linked to whether the deal was won?

Yes, where the partner's recording tool or CRM linked them. Calls are usually matched to opportunities through calendar data and participants' email domains, so a delivery can carry the opportunity ID, the deal's stage at the time of the call and its final result. Calls with no matching deal, such as early prospecting, have no outcome, and links should be spot-checked because automatic matching makes mistakes.

Did prospects agree to be recorded?

Usually they were told the call was being recorded, which is not the same as agreeing to AI training, so the basis for licensing is checked before scoping. Recording tools usually announce themselves through a visible bot, a spoken notice or the calendar invite, and on calls that cross state or national borders the stricter jurisdiction's rules may apply. Prospects may never have seen the seller's privacy notice. Calls from periods or regions where disclosure was weak are left out.

How is confidential information from prospects handled?

With pseudonymization, masking and exclusions. Buyers describe budgets, systems, vendors and internal problems on sales calls. Company names are pseudonymized and identifying details in transcripts are masked, but context can still point to a company, so deals covered by restrictive NDAs or involving sensitive buyers are excluded during scoping. If full transcripts are not essential, summaries and metadata for some calls can widen what is licensable.

Is video included, or only audio and transcripts?

Audio and transcripts are the default; video is scoped case by case. Many recording tools capture video and screen shares, which add faces, slides and whatever was on screen, from demo environments to a prospect's internal tools. That content needs its own review before it can be licensed, so state whether you need video, and for what, when you submit the request.

How do sales calls differ from contact center recordings?

In structure, length and outcome. A sales call is usually a scheduled meeting with several people on each side, one of a series of conversations over weeks or months, and its result shows up later in the CRM. A contact center call is a shorter service conversation, usually one-to-one and resolved on the spot or escalated. Sales calls suit sales agents, deal-risk and coaching models; contact center calls suit ASR, support voice agents and QA automation.

Can I get the CRM history and emails for the same deals?

Yes, when the partner holds them and rights allow. Calls are most useful alongside the deal's stage history, logged emails and loss reasons, and pseudonymous IDs stay consistent across tables so you can join them. Scoping the combined set in one request avoids mismatched coverage, where a deal has calls but no outcome, or an outcome but no calls.

Sales CRM pipeline histories
Opportunity histories with stage changes, logged activities and won or lost outcomes
Contact center call recordings and transcripts
Recorded service and support calls with diarized transcripts, dispositions and QA scores
Approved workplace email and chat exports
Approved, de-identified exports of team email threads and chat channels
Human feedback and QA-scored work
Work items with scores, verdicts and corrections from the people who reviewed them

Evaluating this data for procurement?

Diligence packets are prepared per dataset. Rights, privacy processing and quality differ between datasets.

Request dataset diligence

Tell us what your models need

Send your spec — domain, volume, format, timeline and permitted use — and SourceX will match it against partner data and come back with what can be licensed.