What to Tell Agents
Agents will find out what the system does. Telling them first is a legal requirement in several jurisdictions and an operational advantage everywhere.
Analytics deployed without explanation is experienced as surveillance. The explanation costs nothing and changes how the programme is received for years.
What to tell them, specifically
What is recorded. Calls, screens, chat, and whether breaks or personal calls are excluded.
What is analysed, and whether it is every interaction or a sample.
What is automated. Which scorecard items are machine-determined and which are human.
What is not done. This item matters as much as the others. If emotion inference is off, say so. If real-time supervisor alerting is not enabled, say so. If voiceprints are not created for agents, say so.
What triggers a supervisor alert, in real time, if anything does.
Who can access their data, and whether that access is logged.
How long recordings and transcripts are kept.
How to see their own evaluations and the underlying calls.
How to dispute, and what happens then.
What the data is used for — coaching, compliance, process improvement — and what it is not used for.
What it is not used for
Stating the limits explicitly is the part that builds trust, and the limits should be real.
Not for productivity surveillance beyond the defined measures.
Not for monitoring union or employee representative activity, which is protected in many jurisdictions.
Not for identifying who raised a concern.
Not for performance management without human review.
Not for anything outside the stated purposes.
A system used for one of these once will be assumed to be used for all of them permanently, and the assumption will be reasonable.
The legal position
In several jurisdictions this is not optional.
Transparency about employee monitoring is a requirement, not good practice.
Consultation with employee representatives may be mandatory before deployment, and in some countries a deployment without it is unlawful and its output unusable in a disciplinary process.
Data subject rights mean agents can request their data regardless of whether you offer it.
Establish which of your jurisdictions require consultation before you buy, not after.
How to communicate it
Before deployment, not when the first finding lands.
In writing, available permanently, not in a briefing people forget.
In plain language. A privacy notice written by lawyers satisfies the legal requirement and communicates nothing.
With a route to ask questions that produces answers.
Repeat it when anything changes. A new capability enabled quietly is the thing that destroys the trust the original disclosure built.
What agents actually ask
Anticipating these makes the disclosure useful rather than defensive.
"Is it listening all the time?" Answer precisely, including breaks and personal calls.
"Can my supervisor listen live?" Yes or no, and under what conditions.
"Does it judge my tone?" Answer honestly. If sentiment is computed, say whether it reaches their record.
"What happens if it gets it wrong?" The dispute route, and evidence that disputes are upheld.
"Will this be used to reduce headcount?" The most common unasked question. An honest answer, whatever it is, is better than an evasion that everyone reads correctly.
The operational benefit
Agents who understand the system report faults in it, which is how you find out the recogniser mishears a product name.
Agents who do not understand it work around it, which produces the scripted, defensive behaviour that degrades customer experience.
The disclosure is the cheapest quality intervention available and it is skipped because it feels like it invites objections. The objections exist either way; disclosure means they arrive as feedback rather than as attrition.
The one-page agent notice
The document itself, written for agents rather than for lawyers.
What is recorded: calls, screens, and what is excluded.
What is analysed: every interaction or a sample, and by what.
What is automated: which scorecard items, and what a person decides.
What is not done: emotion inference, live supervisor alerting, voiceprint enrolment — whichever of these are genuinely off.
Who can access it, and whether that access is logged.
How long it is kept.
How to see your own data.
How to dispute a finding, and what happens then.
What it is used for, and what it is not used for.
One page, plain language, permanently available, reissued on any change. The legal notice can exist separately; this is the one people read.
External reference: worker monitoring and data protection.