Skip to main content
NeuronLens QA scoring replaces manual spot-checking with automated, consistent evaluation of every call your agents and bots handle. You define the quality criteria that matter to your business — the specific things an agent should say, disclose, offer, or do on each call — and NeuronLens scores every conversation against them. The result is an objective, reproducible quality score for each call, aggregated into per-agent and per-team leaderboards you can act on immediately.

​
How QA Scoring Works

NeuronLens reads the full transcript of a call and evaluates it against your scorecard parameters. Each parameter is scored individually — binary (yes/no), on a numeric scale (0–10), or via keyword-presence detection — and the weighted scores combine into an overall call score between 0 and 100. You receive a score and an explanation for each parameter, not just a number.

Binary Parameters

Yes/no checks: did the agent do this thing or not? Best for mandatory steps like opening greeting, required disclosure, or call close.

Scale Parameters

A 0–10 score with model-generated reasoning. Best for subjective quality dimensions like empathy, product knowledge, or objection handling.

Keyword Presence

Checks whether specific required phrases or terms appeared in the transcript. Best for compliance-adjacent disclosures and regulatory language.

Weighted Aggregation

Assign a weight to each parameter so critical items (like regulatory disclosure) carry more impact than optional best-practice items.

​
Configuring a QA Scorecard

You set up scorecards in the NeuronLens dashboard or via API. Each scorecard is reusable across campaigns and agent teams.
1

Open the Scorecard Builder

Navigate to NeuronLens → QA → Scorecards → New Scorecard in your dashboard.
2

Name and Scope Your Scorecard

Give the scorecard a name (e.g., “BFSI Collections – Standard”) and optionally assign it to specific campaigns or agent groups so it applies automatically to matching calls.
3

Add Parameters

For each quality criterion, set a name, a description (used by the model to interpret the call), a scoring method (binary / scale / keyword), and a weight (1–10). Add as many parameters as you need.
4

Save and Activate

Save the scorecard and toggle it to Active. NeuronLens will apply it to all new calls that match the assigned scope, starting immediately.
The following parameters are representative of a scorecard for a collections call in the BFSI sector. Adjust weights and scoring methods to match your quality standards and regulatory requirements.
#ParameterScoring MethodWeightDescription
1Proper opening greetingBinary7Did the agent greet the customer by name and identify themselves and the organisation at the start of the call?
2Disclosed purpose of callBinary9Did the agent clearly state the reason for the call — specifically, the outstanding loan or EMI — within the first 90 seconds?
3Correctly stated outstanding amountKeyword + Scale10Did the agent mention the exact outstanding amount, due date, and account reference? Scored on accuracy and clarity.
4Offered payment optionsBinary8Did the agent proactively mention at least two payment methods (e.g., UPI, bank transfer, payment link via SMS)?
5Compliance with RBI guidelinesBinary10Did the agent avoid prohibited language — no threats, no contact with third parties, no calls outside permitted hours?
6Professional call closeScale (0–10)6Did the agent summarise the agreed next step, confirm the customer understood, and close the call courteously?

​
Fetching QA Scores via API

​
Single Call

curl "https://api.vinfer.ai/v1/analytics/qa?call_id=call_abc123" \
  -H "Authorization: Bearer YOUR_API_KEY"

​
Response

{
  "call_id": "call_abc123",
  "scorecard_id": "sc_bfsi_collections_v2",
  "overall_score": 84,
  "pass": true,
  "scored_at": "2024-01-15T10:47:00Z",
  "parameters": [
    {
      "parameter_id": "p_001",
      "name": "Proper opening greeting",
      "method": "binary",
      "weight": 7,
      "score": 1,
      "passed": true,
      "explanation": "Agent said 'Good morning, am I speaking with Ramesh Iyer? This is Priya calling from VInfer Financial.' at 00:00:04."
    },
    {
      "parameter_id": "p_003",
      "name": "Correctly stated outstanding amount",
      "method": "scale",
      "weight": 10,
      "score": 7,
      "passed": true,
      "explanation": "Agent stated the outstanding amount of ₹14,200 correctly but did not mention the due date clearly. Customer had to ask for clarification."
    },
    {
      "parameter_id": "p_005",
      "name": "Compliance with RBI guidelines",
      "method": "binary",
      "weight": 10,
      "score": 1,
      "passed": true,
      "explanation": "No prohibited language detected. Call was made within permitted hours. No third-party contact occurred."
    }
  ]
}
​
overall_score
integer
Weighted aggregate QA score for the call, on a 0–100 scale.
​
pass
boolean
Whether the call met your scorecard’s minimum passing threshold (configurable per scorecard).
​
parameters
array
Array of individual parameter results. Each entry includes the parameter name, scoring method, weight, raw score, pass/fail flag, and a natural-language explanation of how the score was determined.

​
Bulk QA Scoring

Submit a batch of calls for QA scoring and receive results via webhook or polling: 
curl -X POST https://api.vinfer.ai/v1/analytics/qa/batch \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "call_ids": ["call_abc123", "call_def456", "call_ghi789"],
    "scorecard_id": "sc_bfsi_collections_v2",
    "webhook_url": "https://your-server.com/webhooks/qa-results"
  }'
NeuronLens processes the batch asynchronously and posts each result to your webhook as it completes. You can also poll the batch status endpoint using the returned batch_id.

​
Agent Leaderboards

QA scores aggregate automatically into per-agent leaderboards in the NeuronLens dashboard under QA → Leaderboards. For each agent you see:
  • Average overall QA score for the selected period
  • Score trend (up or down vs. previous period)
  • Pass rate (percentage of calls meeting the minimum threshold)
  • Weakest parameter — the single criterion where the agent scores lowest on average
  • Number of calls scored
Use leaderboards to identify your top performers for recognition and your bottom performers for targeted coaching. Clicking an agent opens their individual score history and links to the specific calls that drove their averages.
QA scoring requires transcription to be enabled for the call. When you submit recordings via the API, include "transcription" in the features array. Without a transcript, NeuronLens cannot evaluate scorecard parameters and the call will be skipped during scoring.

​
QA Scoring API Reference

​
call_id
string
The unique call identifier returned when the transcription job completed.
​
scorecard_id
string
The scorecard to apply. If omitted, NeuronLens uses the default scorecard assigned to the campaign this call belongs to.
​
from
string
Filter bulk QA requests to calls processed after this date (YYYY-MM-DD).
​
to
string
Filter bulk QA requests to calls processed before this date (YYYY-MM-DD).
​
agent_id
string
Filter results to a single agent’s calls.