Blog
Voice, work, and what people actually think
Field notes on voice AI, workforce feedback, adoption measurement, and research at scale, from the team building Tadeus.
-
The Channel Audit: What Actually Reaches a Deskless Workforce
Your comms cascade loses the message at every step. A four point channel audit for reaching a deskless workforce.
-
The Response Rate Trap: Why Deskless Engagement Scores Measure the Wrong Thing
A high response rate proves a survey was finished, not that anyone was present. Behavioural signals separate real signal from completion mode.
-
The Missing Tool in Every Agent Stack: Asking a Human
Agents need ground truth from people. Forms return text with no reliability signal. Tadeus returns scored, structured voice answers an agent can trust.
-
How to Tell Whether an AI Interview Study Is Actually Reliable (A Checklist)
Most AI interview accuracy studies measure the wrong thing. Use these 5 tests, grounded in 2025-2026 research, to separate real evidence from marketing.
-
Tadeus is out of beta
Tadeus, the voice layer for workforce platforms, is live and API & MCP-native, so the agents your enterprise is building can already talk to your workforce.
-
The Action Gap: You Earned Their Honesty. Now What?
Listening programmes die at action, not collection. Honesty seen doing nothing trains a workforce to stop answering truthfully. Here's how to close the loop.
-
Your Next Buyer Might Be an Agent
Vendor evaluation increasingly runs through AI agents reading your docs. Machine-readable specs get evaluated; gated PDFs go invisible. Here's how to be legible.
-
How to Get a Works Council to Yes
Works councils approve workforce AI when they can see what it infers, who reads answers, how workers decline and what happens to the data. Bring the proof.
-
The Second Deadline: What December 2027 Actually Requires From Workforce AI
From 2 December 2027, high-risk employment AI must prove governance, logging, oversight and accuracy. Here is what a system must produce, and when.
-
A Word Count Is Not a Quality Signal
Word count, duration and sentiment measure output, not presence. The behavioural factors that make a response quality score reproducible and explainable.
-
AI Disclosure Becomes Law on 2 August
From 2 August 2026, any AI that talks to a person must disclose it. For workforce voice tools, the real test is proving disclosure happened.
-
The Compliance Conversation Record: A Proposed Specification
A v0.1 specification for audit-grade conversation evidence: fields, timestamps, abstention rules, quality scores, retention, and the provider-versus-deployer duty split.
-
The 80% Your Platform Has Never Really Heard From
Most listening tools assume a desk, email and screen time. That excludes the 80% deskless majority. Why reaching them is an instrumentation problem, and what voice fixes.
-
Employees Won't Talk to a Robot
Employees don't refuse AI, they refuse slow AI. Why response latency is both the adoption curve and a proxy for whether an answer meant anything.
-
From Zero to 10,000 Conversations in a Day: What Enterprise Rollout Actually Looks Like When Communication Is Two-Way
How a workforce-scale conversational campaign runs cohorts, invite links, live quotas and same-day structured data, versus weeks of manual scheduling.
-
How to Evaluate a Voice Layer Before You Embed One
Test a voice layer against the conditions that break it: latency under load, accents in noise, structured output, maintained languages, and a per-session quality score.
-
Build vs Buy: Why Every Major HCM Vendor Will Have a Voice Layer by 2028 (and Almost None Will Build It)
Every major HCM platform will offer real-time voice soon, but almost none will build it. Why embedding voice via API beats a multi-year in-house build.
-
Stop Running Surveys: The Case for Workforce Intelligence That Compounds
Quarterly surveys give a stale, gamed snapshot. Continuous conversation data benchmarks against its own history, gets sharper with use, and becomes hard to switch away from.
-
Checkbox Compliance Is a Liability: Why "They Clicked Accept" Won't Survive Your Next Audit
A ticked box proves a click, not comprehension. Auditors now test whether controls worked in practice, and a conversation record is the evidence that holds up.
-
The Silent Gap: Why Workforce Platforms Broadcast but Never Hear
Workforce systems broadcast but have no way to know if anyone understood. The cost of one-way communication, and why employee voice is an infrastructure gap.
-
It was never meant to be just a survey tool:
A completed onboarding survey and one filled in on autopilot look identical. Here's why conversational intelligence collects what surveys can't.
-
How do you know AI-collected data can be trusted?
Every AI vendor claims trustworthy data. Few can name their failure modes. How research-grade evaluation shapes the way Tadeus scores every conversation.
-
What happens when every employee and product has a voice
Conversational intelligence embedded in workforce platforms replaces completion rates with quality-checked signal. Here is why the infrastructure question matters.
-
Voice led Compliance
A click proves the screen advanced. A structured voice conversation proves comprehension. Here's why that difference is now a compliance liability.