Your employer deployed ClarityLift. Here is what it sees about you.

Imagine it is a Tuesday at 2:14pm. You typed a message in a channel your employer turned on for ClarityLift analysis. Below is everything that happens after you hit send. Every step is enforced in code, not in a policy document.

1. 1

   Slack or Teams delivers the event to our webhook.

   The event is metadata plus message text. Your employer only granted channel-read permissions, so this only fires for messages posted in channels they have explicitly turned on. Direct messages, group DMs, and channels your admin has not enabled never reach us in the first place.

2. 2

   The handler checks if the channel is a DM. If yes, it drops the event.

   Even if Slack accidentally sent a DM to our webhook, the first line of the handler rejects it before any other code runs. We do not store it, classify it, or log it. The server returns 200 and moves on.

3. 3

   A fast pattern-matcher decides if the message is worth classifying.

   Most messages are not. "ok", "lol", "thanks", and the majority of the day's chatter fail this filter and are discarded on the spot. Only a small fraction of messages proceed to the next step.

4. 4

   If it passes, the message text is sent to a language model once, in memory.

   The model sees the text long enough to decide what category of signal the message fits (team friction, communication health, culture drift, retention, strategic alignment, or none). The model returns a category and a confidence score. That is all it returns. It does not return your name. It does not return a quote. It does not return a summary.

5. 5

   The model's provider does not retain the text either.

   We use providers under zero-retention agreements. On Azure OpenAI, the request stays inside Microsoft's Azure boundary and is not used to train anything. On Anthropic, a contractual zero-data-retention clause prevents storage. The message text is discarded after the call.

6. 6

   The row that gets written to our database has seven fields.

   Organization id, team id, channel id, signal category, severity, confidence, detection timestamp. That is it. No user id. No message id. No text. No quote. No author. If an admin wanted to look up "what did employee X say," the data to answer that question does not exist.

7. 7

   Nothing shows up on any dashboard unless the team has 10 or more members.

   Before any output is visible to your manager, a database check confirms the team has at least 10 people. Teams below that threshold produce no output at all. Not a "hidden" output. No row is emitted.

These are not settings. A workspace admin cannot toggle them. A paying customer cannot negotiate them. They are built into the schema and the ingest pipeline.

Your manager, or whoever holds the ClarityLift admin seat at your company, sees a dashboard built from the signal rows described above. Here is what is on it.

• Team-level scores, per team, per week.

  Five pillar scores: team friction, disengagement, communication health, culture drift, retention signals. Alignment to stated strategic goals is on the roadmap. Each score is a number with a trend line. Each is computed from the aggregated signal rows for that team over the window.

• Which signal categories fired for the team.

  A count of signals by category over the period. Twelve friction signals, three communication-health signals, zero retention signals. Not which messages caused them. Not who said them.

• Anomalies against the team’s own baseline.

  After four to six weeks of baseline, the scoring engine flags unusual weeks for the team relative to its own history. Not relative to other teams. Not relative to an industry benchmark.

• Written insights, generated weekly.

  A short paragraph summarizing what moved on the dashboard. The paragraph is generated from aggregate scores only. It contains no quotes. It cannot contain a quote because no quote was ever stored.

The difference between a permissions-based limit and a structural limit matters. A permission can be granted. A structural limit means the data does not exist to reveal, even to the CEO, even under subpoena, even if every engineer at ClarityLift cooperated. These fall into the second category.

• Your messages. Full stop. The text is not stored anywhere.
• Your identity on any signal. Rows carry a team id, never a user id.
• Which message triggered which signal. There is no link from a dashboard number back to a source message.
• Your DMs. The integration never had access to them.
• Your private-channel messages, unless the admin explicitly turned that channel on.
• A per-employee score. Every output is scoped to a team of 10+.
• A list of who is a flight risk. The product has no such view, because it has no per-person data.
• A transcript of any conversation. Not for any reason, including legal discovery, because no transcript exists on our side.

You have rights under your local privacy law, and you have a shorter list of things ClarityLift commits to regardless of jurisdiction. Both are below.

Everything on this page is backed by code or contract you can read, or by a claim your own admin can verify inside their own Slack or Teams workspace.

• We never analyze DMs.

  Your admin can open the ClarityLift OAuth consent screen and see the scopes we requested. There is no DM scope. If we had wanted access, we would have had to ask.

• We never store message text.

  The signal-row schema is public in our privacy architecture document. Seven fields, none of them text. A DSAR export contains signal counts and scores. There is no row type that could contain a message body.

• No output is produced for a team below 10 members.

  Your admin can create a test team of 3 people and turn on ClarityLift for a week. The dashboard for that team will be empty. This is not because the data was hidden. It was never written.

• The product cannot be used for HR decisions.

  The acceptable-use policy is in every signed order form. Outputs do not name individuals, so an admin could not use them this way even if they wanted to. If it happened anyway, every email address on the AUP reporting list is a notification endpoint.