Salary and compensation details in email and chat archives

Why compensation needs its own exclusion rule#

Compensation needs its own exclusion rule because pay details identify people even after names are removed and cause real harm if exposed. A message about the raise for the only controller at the company's smaller branch names no one, yet everyone who works there knows who it means.

The rule is simple to state. Any message, file or field that reveals the pay, bonus, commission, equity, benefits elections, deductions or garnishments of an identifiable person, or of a group small enough to identify, is excluded or redacted. Company-level figures, such as a project budget or a department's total spend, are business records and follow normal review.

The value trade-off favors exclusion. AI developers license email and chat archives to learn how work gets done: how a quote is negotiated, how an exception is escalated, how a manager decides. Individual pay rarely helps with that, so removing it costs little.

Mailboxes and channels to exclude wholesale#

Some mailboxes and channels exist mainly to discuss pay and people, and excluding them wholesale is safer and cheaper than filtering them message by message. Build the exclusion list from your directory of shared mailboxes, distribution lists and channel names before any export is scoped.

Export settings can narrow or hide the problem. Slack offers exports covering private channels and direct messages only on Business+ and Enterprise plans, and Workspace or Org Owners must apply to use them; exports by conversation type or by member are listed only for Enterprise. On other plans the export may hold public channels only, which reduces exposure but does not remove pay talk from public channels.

Detection patterns for pay talk everywhere else#

Detection patterns for pay talk pair a currency amount or a raise figure with compensation vocabulary in the same sentence or a short window of text. Amounts alone flood review with quotes, invoices and purchase orders, and vocabulary alone catches harmless phrases such as pay attention.

Employees often discuss pay with coworkers, and in many workplaces they are legally protected in doing so, so expect those conversations in ordinary channels and direct messages, not only in HR spaces. That is not a reason to license them; it is a reason to scan everywhere.

Drop the message or redact the sentence?#

Whether to drop a message or redact a sentence depends on the message's purpose. If compensation is the point of the message, drop it along with the rest of its thread; if pay comes up in passing inside an operational discussion, redact the sentence and keep the rest.

Apply the decision consistently across email and chat. A thread dropped from Outlook should not survive as a forwarded copy in a Teams chat or as an excerpt pasted into a Jira comment.

• 1Drop: offer negotiations, raise requests, bonus decisions, commission disputes, payroll corrections.
• 2Redact: a project update that mentions one employee's overtime pay, or a quote thread that notes a named technician's rate.
• 3Generalize: staffing plans that list roles and budget totals without individual pay.
• 4Escalate: anything mixing pay with health, leave, discipline or legal claims, which usually means excluding the whole thread.

Illustrative: an industrial distributor's Microsoft 365 archive#

Illustrative: a fictional industrial distributor with branch sales teams wants to license Outlook email and Teams conversations about order exceptions, substitutions and customer escalations. Its sales reps earn commission, and branch managers often handle raise requests through chat.

Microsoft Purview treats Teams channel messages and Teams chats as separate locations, so the team could scope them separately for a clean first cut: channel messages from operations teams went into scope, while one-to-one and group chats stayed out. HR, payroll and the sales commission mailbox were excluded wholesale.

Pattern detection then ran on what remained. It caught commission figures posted in a branch channel during a month-end push and a few raise discussions in operations channels. Those messages were dropped, and incidental rate mentions in quote threads were redacted. The licensed scope kept the exception-handling workflow and none of the pay data.

Payroll records you still must keep#

Payroll records you must keep stay in your systems on their normal schedule, whatever the license scope. The Department of Labor's FLSA recordkeeping fact sheet says employers must preserve payroll records for at least three years, and records on which wage computations are based, such as time cards and wage rate tables, for two years.

Licensing preparation works on a copy. Excluding a payroll mailbox from the export does not delete or alter the original, and retention schedules, legal holds and audit needs continue to apply to the source systems.

How SourceX handles compensation data#

SourceX treats compensation as an excluded category in the Preparation step of the SourceX five-step transaction: Supply, Rights, Preparation, Approval and Delivery. Wholesale exclusions are agreed with the supplier during Supply, before any export is scoped.

The exclusion list, the detection patterns and the sample review are recorded in the privacy record of the SourceX Evidence Packet, and the supplier approves the result before release.