Intelligence

Intelligence groups processed calls by agent and turns call analytics into agent-level results: call counts, average scores, sentiment, risk and customer-satisfaction measures, trends, and profile-specific insights.

Intelligence is enabled per account by Speechlytics.

Agent attribution

Speechlytics associates a call with an agent using identifying call metadata such as username or local, or the agent details supplied by an integration. When a call names an agent that does not exist yet, the agent is created automatically.

For direct API submissions, send identity information in the serialized metadata field:

{
  "username": "agent@example.com",
  "local": "1001",
  "remote": "+15551234567",
  "externalId": "provider-call-42"
}

Use consistent identifiers: formatting the same person differently (for example 1001 and +1 1001) creates separate agent records.

Listing agents

GET /api/v1/intelligence?Page=1&Rows=10&DateFrom=2026-09-01&DateTo=2026-09-30

Parameter	Meaning
Page, Rows	Paging.
DateFrom, DateTo	Period of calls to aggregate. Defaults to today.
Name, Username, Phone	Filter by agent identity.

The response contains the agents, each with these measures over the period, plus the total number of agents:

Field	Meaning
totalCalls	Analyzed calls attributed to the agent.
averageScore	Average overall call score, 0–100.
customerSatisfaction	Average customer-satisfaction rating, 0–100.
positiveSentimentPercentage	Share of calls whose overall sentiment is positive.
riskPercentage	Share of calls rated high risk.
isImproving	true when the agent's newer half of calls in the period averages a higher overall score than the older half (needs at least two calls).

Agents without attributed, analyzed calls in the period do not contribute.

One agent

GET /api/v1/intelligence/id?id={agentId}&from={date}&to={date}

id, from and to are query parameters. An unknown agent returns 404. The response follows the analysis profile used by most of the agent's calls in the period, reported as dashboardType (0 Sales, 1 CustomerSupport).

Common to both profiles

Field	Meaning
averageScore, totalCalls	Headline score (0–100) and number of calls.
positiveSentimentPercentage, neutralSentimentPercentage, negativeSentimentPercentage, riskPercentage	Sentiment split and high-risk share.
radarScores	Six average ratings, 0–100, in a fixed order (below).
topStrengths, topWeaknesses	Recurring strengths and improvement areas across the calls.
topTopics	Most frequent topics with counts.
customerEmotion, agentEmotion	Emotion distributions.
calls	The agent's calls in the period, with date, type, score, sentiment and risk, and the call id to drill into.

radarScores order

#	Sales	Customer Support
1	Customer Satisfaction	Customer Satisfaction
2	Agent Clarity	Agent Clarity
3	Professionalism & Tone	Professionalism & Tone
4	Resolution Effectiveness	Resolution Effectiveness
5	Sales Opportunity Management	Issue Resolution Quality
6	Sentiment & Relationship Building	Empathy & Active Listening

Sales profile: salesOpportunityRate, upsellRate, churnSignalRate, resolutionEffectivenessRate and a businessImpact summary.

Customer Support profile: firstContactResolutionRate, escalationRate, retentionConcernRate, issueResolutionQualityAverage, empathyActiveListeningAverage, and distributions of issue category (Technical, Billing, General Inquiries, Complaints, Other), urgency (Critical, High, Medium, Low), follow-up action (Callback Needed, Email Follow-up, Escalation Required, No Follow-up Needed), risk (High, Medium, Low) and resolution status (Resolved, Partially Resolved, Escalated, Pending), plus recurring improvementThemes.

Interpreting agent results

• Rating averages (customer satisfaction, radarScores) use only calls that contain the rating. In averageScore, a call without an overall score counts as 0, so analyze calls consistently before comparing agents.
• Percentages can swing widely when a period contains few calls; consider the call count alongside them.
• Compare similar roles and call types, over representative periods.
• Review source calls before using a score for coaching or evaluation.
• Transcript and speaker-separation quality affect every derived result.

Troubleshooting

Symptom	Likely explanation
Empty results	Intelligence is not enabled for the account, or no calls in the period identify an agent.
An agent has no calls	Submitted or imported calls did not contain a matching identity.
The same person appears twice	Their calls used inconsistent username, phone or local identifiers.
Sparse results	Some analytics were not available on calls in the period.
Sales and support fields differ	The calls used a different analysis profile.

See the API reference for the exact schema.