SDR Call Intelligence

Configure how calls are evaluated, distinguish live conversations from automated calls, review evidence-based scores, and coach SDR performance.

SDR Call Intelligence turns available phone-call and meeting recordings into a searchable review, coaching, and performance workspace. Marcenta identifies what kind of interaction occurred, transcribes eligible audio, evaluates live conversations against your organization's calling criteria, extracts outcomes and signals, and preserves evidence for manager review.

Automated analysis is decision support. Recording quality, transcript accuracy, missing context, and source data can affect the result, so important conclusions should be checked against the recording and transcript.

How SDR Call Intelligence works

For each eligible imported recording, Marcenta:

imports the recording and available participant metadata from the source;
processes the audio or supplied transcript;
separates and identifies speakers where the evidence allows;
classifies the interaction as a live human conversation or a non-human call type;
performs deeper evaluation only when the interaction contains a real human conversation;
evaluates the call against the organization's saved script, discovery, next-step, persona, and scoring settings;
extracts the outcome, objections, buying signals, conversation dynamics, coaching opportunities, and manager-review recommendation; and
makes the result available in the recording detail, analytics, and reporting data.

Import, recording retrieval, transcription, classification, deeper analysis, and result storage are separate processing stages. A call can appear in the recording list before every stage is complete.

Calls classified by Marcenta

Marcenta assigns one canonical call class:

Call class	Meaning
Human conversation	A real exchange occurred between the representative and another person
Screening	A device, carrier, or call-screening system intercepted the call
IVR	An automated voice menu or interactive voice response system answered
Voicemail	A voicemail or answering-machine greeting was reached
No connect	The call rang without an answer, contained only hold or system audio, or ended without a conversation
Other non-human	The recording is non-human but does not fit one of the more specific classes
Uncertain	The available evidence is insufficient to classify the interaction confidently

Screening can be further identified as:

iPhone or device gatekeeper;
call-screening gatekeeper; or
carrier or system gatekeeper.

Classification prevents IVR, screening, voicemail, ringing, and other non-human audio from being treated as SDR performance. These calls remain counted and classified, but they do not receive a live-conversation quality score.

A short or unsuccessful conversation with a real person is still a human conversation. Outcome and quality are separate questions: a human conversation can end with no interest, a callback request, a referral to the correct person, or another result without being reclassified as non-human.

Data sources

Apollo is a supported source for outbound call records and recordings. Connect it before expecting Apollo calls to appear. See Apollo for connection, API-key, import-window, and recording-access requirements.

Qualifying meetings booked through Marcenta can also enter a meeting-focused recording and analysis flow when Record and analyse meeting is enabled in the matched routing outcome and the booking contains a supported conferencing link.

Phone calls and discovery meetings use different analysis profiles:

Outbound phone calls focus on call classification, SDR execution, outcome, objections, signals, scoring, and coaching.
Discovery meetings focus on pains, business impact, urgency, use cases, desired outcomes, current processes and tools, stakeholders, budget, timelines, decision criteria, competition, risks, commitments, and verified next steps.

Configure what calls are evaluated against

Workspace owners and admins can open Settings → Call Analysis to define the shared evaluation criteria. The saved settings apply across the organization; individual representatives do not configure a separate rubric for their own calls.

Provide the following inputs:

Input	What to provide	How it is used
Pitch script	The complete approved call flow or master playbook	Evaluates whether the representative followed the expected structure and milestones
Required pitch points	One required idea, phrase, proof point, or talk-track milestone per line	Identifies which expected points were covered or missed
Required discovery questions	One required question or discovery topic per line	Evaluates whether the expected discovery was performed when the conversation allowed it
Allowed next-step types	One acceptable next step per line, such as demo, assessment, technical review, or follow-up call	Evaluates whether the representative attempted and secured a concrete approved next step
Ideal titles	One target role or job title per line	Helps evaluate whether the contacted person matches the intended persona
Ideal seniorities	One target seniority indicator per line	Adds seniority context to persona evaluation
Personas to avoid	One out-of-scope persona per line	Identifies contacts who should generally not be treated as the intended buyer
Scoring weights	The relative contribution of each scored category	Produces the overall quality score

Use specific, observable criteria. “Mention customer proof after establishing relevance” is more useful than “deliver a great pitch.” For discovery questions, describe the information that must be uncovered rather than requiring an exact sentence unless exact wording is mandatory.

Avoid duplicating or contradicting entries. If “Technical review” is an allowed next step but another entry implies that only demos are acceptable, managers may find the resulting assessment difficult to interpret.

Defaults and blank settings

Blank settings do not stop analysis. Marcenta provides default guidance for:

the outbound calling flow;
required pitch points;
discovery questions;
allowed next steps;
ideal titles; and
ideal seniorities.

Organization inputs replace the corresponding defaults where supplied. If you provide a custom pitch script but leave required pitch points empty, Marcenta evaluates the custom script without separately requiring the default pitch-point list. Personas to avoid remains empty unless your organization supplies entries.

Do not leave a field blank merely to disable its scoring category. A default may still apply. If an area should have no effect on the overall score, set its scoring weight to zero and redistribute the remaining weights so the total remains 100%.

Scoring categories

Every eligible live human conversation can be evaluated across seven weighted categories:

Category	What Marcenta evaluates	Default weight
Script adherence	Whether the representative followed the supplied playbook flow and covered expected milestones	20
Discovery	Whether required discovery questions or topics were addressed	20
Objection handling	How effectively the representative recognized and responded to explicit resistance	15
Next-step control	Whether an allowed, specific, and verifiable next step was attempted or secured	20
Prospect persona accuracy	Whether the contacted person fits the supplied title and seniority criteria and avoids excluded personas	15
Brevity and relevance	Whether the representative stayed concise, relevant, and avoided unnecessary monologues	5
Fluency	Clarity, filler words, verbal clutter, awkward phrasing, and conversational smoothness	5

Owners and admins can change these weights, but they must add up to exactly 100 before the settings can be saved.

How the overall score is calculated

Each evaluated scoring category receives a score from 0 to 100 with an assessment and supporting evidence. The overall quality score is the weighted average of the categories that could be evaluated for that call.

A category can be marked Not evaluated when the required context or conversation event does not exist. For example:

objection handling is not evaluated when no genuine objection occurred;
a non-human call does not receive the SDR rubric;
a category that lacks usable evidence can remain unevaluated; and
a setting-dependent category is not guessed when the required reference context is unavailable.

Unevaluated categories are excluded from the calculation, and the remaining active weights are normalized. They are not silently treated as zero. If no weighted category can be evaluated, the overall score is unavailable.

The overall score is therefore not a direct measure of revenue impact, effort, or employee value. Compare scores only after considering the rubric version, source quality, call type, sample size, and whether the same categories were evaluable.

When settings changes take effect

The active organization settings are attached when a recording is dispatched for analysis. Saving a new rubric affects calls analyzed after that point.

Completed historical calls are not automatically rescored. Use Reprocess on a representative sample when you need to understand how a revised rubric changes results. Avoid reprocessing the entire history until managers have validated the new criteria.

Frequent rubric changes make scores across time less comparable. Record major changes internally and tell managers when the scoring basis changes.

Outcomes detected

For live conversations, Marcenta can identify outcomes such as:

meeting booked;
follow-up requested;
information requested;
callback requested;
not interested;
wrong person;
already a customer; and
using a competitor.

Outcome classification also preserves a concise summary, the underlying reason, and transcript turns supporting the conclusion when available.

An outcome is separate from the source platform's disposition. Marcenta derives it from the analyzed conversation, while the source may contain a manually selected or provider-generated call outcome. Review both when they disagree.

Objections and buying signals

Marcenta looks for explicit prospect objections and classifies them into practical categories:

need or interest;
price or budget;
timing or urgency;
trust or credibility;
authority;
existing vendor; and
other.

For an objection, the analysis can include the prospect's words, the turn where it appeared, the representative's response, a response-quality rating, and qualitative feedback. Use Play Objection, Play SDR Response, or the transcript-turn link to verify the exchange.

Marcenta can also identify buying signals, positive momentum, risk or stall moments, prospect concerns, disengaged portions of the conversation, and competitor references when supported by evidence. Absence of a detected signal means the analysis did not find sufficient evidence; it does not prove the signal never occurred.

Opener analysis

The analysis identifies the representative's actual opener after any IVR, screening, receptionist, or routing exchange. Supported opener patterns can include:

permission-based;
direct reason for calling;
mutual connection or referral;
pattern interrupt;
problem-led; and
unknown when the real opener cannot be determined.

Marcenta records the opener style, the earliest supporting turn, opener feedback, and time to hook where timing evidence is available. Opener grading is shown on a 1–5 scale and is separate from the 0–100 weighted quality score.

Conversation dynamics

For supported live conversations, the Performance view can show:

representative and prospect talk share;
the representative's longest monologue and word count;
average and median response latency;
interruption count;
prospect turn count;
opener performance;
conversation-health observations; and
evidence turns behind scored dimensions.

These measurements require usable speaker separation and timing data. Missing values can indicate unavailable evidence rather than neutral performance.

Talk share has no universal ideal. A discovery call, cold outbound call, qualification call, and technical meeting naturally have different conversational patterns. Use the metric with the call purpose and transcript.

Manager review

Marcenta determines whether a manager should review a call using the analyzed evidence. When review is recommended, the result can include:

low, medium, or high priority;
a review theme;
a specific reason;
supporting transcript turns;
an evidence summary;
discussion anchors;
a top coaching priority;
a suggested manager coaching script; and
focus areas for the next call.

High priority is intended for serious missed opportunities, risky claims, compliance concerns, or repeated severe behavior supported by evidence. A low score alone should not be treated as the complete review reason.

The manager-review recommendation does not itself complete a human review or take action against a representative. Managers should listen to the cited moments, account for context, and document their conclusion.

Review the recordings table

Open SDR Call Intelligence → Recordings to see imported calls. Available columns include:

contact, caller, and company;
call date and duration;
source;
whether a human conversation occurred;
call outcome and outcome type;
analysis status;
quality score; and
manager-review requirement and priority.

Use search and filters to narrow recordings by caller, manager-review state, outcome, source, and date. View settings controls visible columns and their order. Columns can also be sorted and resized.

The recording list and its filters are scoped to the current workspace. Recent calls can show incomplete analysis while background processing continues.

Review a phone call

Open a recording to access the audio player, call details, transcript, and analysis.

The detail view provides:

Overview — call summary, verified next step, decision signals, key call moments, objections, and available prior-call activity;
Performance — overall and category scores, opener feedback, conversation health, coaching evidence, and conversation dynamics;
Review — manager-review reason, priority, evidence, discussion anchors, and action plan; and
Collaboration — the workspace comment thread for the recording.

Use a cited turn to jump to the corresponding transcript and, where timestamps are available, play the recording from that point. Transcript-linked evidence is the best way to determine whether an automated conclusion fairly represents the conversation.

You can move to the previous or next recording without closing the detail view and refresh it while processing is still underway.

Review a discovery meeting

Discovery meetings use a meeting-focused view instead of the standard outbound-call scorecard. Depending on available evidence, it can organize:

executive summary and disposition;
customer pains and business impact;
urgency, use cases, and desired outcomes;
current process, tools, and incumbent vendors;
stakeholders, authority, and missing participants;
budget and commercial constraints;
timeline, compelling events, and target decision dates;
decision criteria and decision process;
competitors and alternatives;
objections, risks, blockers, and unanswered questions;
qualification signals;
commitments and verified next steps; and
coaching observations and discovery coverage.

Findings can link back to transcript turns. Discovery results should still be verified before updating opportunity stages, forecasts, or customer commitments.

Collaborate on a recording

Use the Collaboration area to add context for managers and representatives. Workspace users can add comments, and a comment author can edit or delete their own comment.

Useful comments capture:

the manager's conclusion after reviewing cited evidence;
business context unavailable in the recording;
the coaching action agreed with the representative;
a corrected interpretation; or
the follow-up that should occur.

Do not paste credentials, access tokens, unrelated customer data, or other secrets into comments.

Reprocess a recording

Use Reprocess when:

valid audio or transcript exists but analysis failed;
an analysis is incomplete and should be retried;
the evaluation rubric changed and you intentionally want a new result; or
improved source content has become available.

Reprocessing runs asynchronously and can take time. It does not repair a recording that the source never supplied, create missing audio, or guarantee a usable transcript from poor-quality content.

Reprocessing replaces the analytical interpretation used for the recording. If the earlier result influenced a management or customer decision, preserve that decision context outside the generated analysis before rerunning it.

SDR Call Intelligence analytics

Open SDR Call Intelligence → Analytics to review aggregate performance for the displayed period. The default analytics period covers the current day and the preceding 29 days.

The page includes:

total calls and human-conversation rate;
meetings booked and meeting rate;
manager-review count and review rate;
buying signals and objection counts;
outcome mix;
manager-review volume by priority;
an SDR leaderboard with calls, meetings, quality score, and meeting rate; and
opener effectiveness with calls, meetings, meeting rate, and average time to hook.

Aggregate quality scores use only calls with available analyzed scores. Compare representatives cautiously when their call volumes, territories, call types, personas, or evaluable rubric categories differ.

SDR Call Intelligence data is also available to Reporting, where supported dimensions and metrics can be combined into custom reports and funnels.

Troubleshooting

Symptom	What to check
No Apollo calls appear	Apollo connection, Master API key access, configured import window, call dates, recording availability, and latest sync
A call appears without analysis	Analysis status and whether recording retrieval, transcription, or background processing is still underway
A call has no transcript	Whether the source supplied usable audio or a supported transcript and whether processing completed
An IVR or screening call has no quality score	This is expected; non-human calls are classified but excluded from live-conversation scoring
A real conversation is marked non-human	Listen to the recording, check transcript and speaker quality, then reprocess if usable evidence exists
A category says Not evaluated	Required rubric context, whether the relevant event occurred, and whether evidence was sufficient
The overall score differs from a simple average	Category weights and exclusion of categories that could not be evaluated
Scores do not reflect new settings	Confirm the settings were saved and that the call was analyzed or reprocessed afterward
Scoring settings cannot be saved	All seven weights must total exactly 100, and only an owner or admin can change them
Evidence opens the wrong moment	Transcript timing and speaker alignment quality; use surrounding turns to verify context
Manager-review totals seem low	Analysis completion, selected period, source coverage, and whether calls actually required review
Analytics differs from the recording list	Date coverage, filters, processing completion, and availability of projected analytics data
Reprocessing fails	Usable stored audio or transcript, source availability, current processing state, and retry status

See SDR Call Intelligence Settings for the administration workflow.

Call recordings and transcripts can contain personal, confidential, and employment-related information. Use SDR Call Intelligence only with appropriate recording notice or consent, restrict access to authorized workspace users, and follow applicable retention, privacy, and employment policies. Do not use an automated score as the sole basis for an employment decision.

On this page