TalkScore Hub is a real-time analytics platform for AI-powered candidate assessments. It gives your team a unified view of everything that runs through TalkScore — pipeline health, scoring consistency, quality alerts, and individual candidate reports — across all campaigns and AI agents at once.
Who is TalkScore Hub for?
Role | What to focus on |
Recruiters | Reports tab — search and filter candidate reports, review individual scores and transcripts, export data for hiring decisions. |
People Analytics / QA | Metrics and Score Calibration tabs — analyze whether TalkScore predicts hiring outcomes, track scoring consistency, identify calibration issues and skill gaps. |
Operations / Hiring Managers | Snapshot and Insights tabs — monitor pipeline health in real time, track completion rates and dropout patterns, act on quality alerts and agent health statuses. |
How to Access TalkScore Hub
Go to talkscore-hub.talkpush.com and log in with your Talkpush email address.
On the Select a Company screen, choose the company you want to work with. Each company card shows its status, country, number of users, and link to the TalkScore instance.
Click Open to enter that company's dashboard.
If you manage multiple companies, you can return to the company selector at any time by clicking the company name in the top-left corner of the sidebar.
Navigating the Hub
The left sidebar is your main navigation. Each tab serves a distinct purpose:
Section | What it does |
Your at-a-glance dashboard. Key metrics for the selected period — calls, completions, scores, and per-assessment breakdowns. | |
Deeper analytics split into General (volume, completion timelines, call status distribution, duration analysis) and Outcome Analysis (hired vs. not-hired comparisons, score predictiveness, CEFR alignment). | |
Tracks whether your AI scoring is consistent over time. Shows standard deviation trends, healthy vs. poor scoring waves, score compression, and per-skill breakdowns. | |
Quality monitoring across four tabs: Health (agent-level quality flags), Feedback (human reviewer opinions), Quality (searchable log of every flag), and Alerts. | |
Candidate-level report list with scores, CEFR levels, and dates. Click any candidate to see their full report — transcript, recording, per-skill scores, sentiment analysis, and quality flags. | |
Your AI agents. Each assessment is a configured voice agent with its own questions, rubric, and system prompt. | |
Settings | User management and notification preferences. Add team members, assign roles, and configure email alerts. |
Key Concepts
Before diving into the tabs, a few terms come up throughout TalkScore Hub that are worth understanding upfront. They'll make the data much easier to read once you're inside.
Scores are on a 0–5 scale
Every candidate receives an overall score from 0 (no usable evidence) to 5 (excellent), calculated as the average of individual soft skill dimension scores. Each dimension (e.g., "Professionalism and Courtesy," "Active Listening") is scored independently based on evidence from the transcript.
Standard deviation measures scoring consistency
You'll see "Std. Dev" or "±" values throughout the Hub. A standard deviation of ±0.5 means most candidates score within half a point of the average — that's consistent. Above ±1.1, scores vary widely and the rubric may need recalibration.
Quality flags are AI-generated
After every completed call, the system reviews the full transcript and raises flags for issues like hallucination (the agent stated something false), repetition, or unprofessional tone. These are grouped into three categories: Agent Communication Quality (12 flags), Bias and Fairness (9 flags), and Candidate Behavior (2 flags). See the Insights and Reports article for the full taxonomy.
Scoring anomalies catch contradictions
Beyond conversation quality, the Hub runs logic-based checks to catch contradictions between scores, completion status, and CEFR levels — for example, a high score on an incomplete call, or a completed interview with no score at all.
CEFR measures language proficiency
If your assessments include language evaluation, candidates receive a CEFR level from A1 (beginner) to C2 (mastery). This is the international standard for rating English fluency.
Waves track scoring over time
The Hub groups scored candidates into "waves" — batches scored within a time window — so you can see whether scoring consistency is improving, stable, or degrading over time.
What happens when a call ends
Call ends — The candidate completes their voice assessment.
Hub processes — Within seconds, TalkScore Hub scores the data across all dimensions, generates the report, and runs quality and calibration checks.
Flags are raised — Quality issues are tagged with severity (Critical, Warning, or Info) and categorized.
Dashboards update — All KPIs, charts, and drill-down reports refresh in real time.
Report delivered — The full report appears in the Hub and is delivered back to the candidate's Talkpush record.
The Snapshot Dashboard
The Snapshot is the first screen you see when you open a company in TalkScore Hub. It gives you an at-a-glance health check of your AI interviews — what's happening right now, what changed compared to last period, and which assessments need attention.
AI-Generated Summary
When you load the Snapshot, the Hub generates a natural-language summary highlighting the most important patterns: peak interview hours, completion trends, notable anomalies. This updates based on the data in your selected time window.
The KPI Cards
Calls Started — The total number of interviews where a candidate actually connected. Excludes "Not Taken" calls. A sudden drop usually means a campaign was paused or a scheduling link broke. A sudden spike may mean a new campaign went live.
Completion Rate — The percentage of started calls where the candidate finished the full interview. If completion drops below 50%, check Peak Dropout to see where candidates are leaving, then go to Insights → Health to look for agent quality issues.
Avg Score — The mean overall score across all scored candidates in the selected period. Most useful when read alongside the standard deviation. An average above 4.0 combined with low deviation may indicate score compression.
Score Std. Deviation — How spread out scores are around the average:
≤ 0.7 — Healthy. The rubric is scoring consistently.
0.7 – 1.1 — Warning. Some inconsistency; worth monitoring.
> 1.1 — Poor. Scores vary widely; the rubric may need recalibration.
Peak Dropout — The interview stage where the most candidates abandon. High dropout at "Intro" often means the opening message is too long or confusing. High dropout at a mid-interview question may mean that question is unclear.
Hallucination Flags — Total AI hallucination detections across all agents in the period. Any count above zero is worth reviewing. High counts across multiple agents indicate a systemic issue — contact your Talkpush representative immediately.
Dropout by Stage
The Snapshot includes a question-by-question funnel showing what percentage of candidates remain at each stage of the assessment. High dropout at the Intro usually means a landing page or agent configuration issue, not candidate quality.
The Agent Health Table
Below the KPI cards, every active agent is listed with its real-time health status: Healthy (no action needed), Warning (monitor closely and check the Insights tab), or Critical (requires immediate investigation — check Insights and contact your Talkpush representative).
Each agent row shows: assessment name, status, total calls, completion %, average score, hallucination flags, standard deviation, and last activity.
Generating a Report
Click the Generate Report button in the top-right to export a downloadable summary of your Snapshot data. This is useful for sharing with stakeholders who don't have Hub access.
Q: Where should I start when using the platform?
A: Here are some common first tasks:
Check overall health: Go to Snapshot and look at the completion rate and average score for the past 7 days. If completion is below 60% or scores cluster at the extremes, investigate further.
Review quality flags: Go to Insights → Health to see what percentage of conversations have issues and which flag types are most common.
Look at a candidate report: Go to Reports, click any candidate, and explore the full transcript, scores, sentiment, and agent quality analysis.
Check scoring consistency: Go to Score Calibration and verify the average standard deviation is ≤ 0.7. If it's trending upward, review the "Poor Waves" and the per-skill breakdowns.
Set up notifications: Go to Settings → Notifications and enable Critical Flag Alerts so you're notified immediately when something goes wrong.
Q: How do time filters work?
A: Most screens include a time filter in the top-right area (e.g., "Last 7 days," "Last 30 days"). Changing this filter updates all metrics on the current page. You can also filter by specific assessment or campaign where available.
