People ops teams sit on three data streams that rarely meet: engagement pulses, experience surveys, and manager 1:1 notes. The synthesis work is slow, the framing is sensitive, and the cost of a sloppy quote is real. A people ops agent can read all three streams, cluster the signal, and draft an insights memo. The HR lead reviews each claim against the source row before the memo ships. The platforms keep the raw responses. Dock keeps the interpretation, the reviewer, and the timestamp.
Culture Amp, Qualtrics, and Lattice stay the system of record for the raw responses, scores, and 1:1 notes. Dock is the system of record for what the agent interprets from that data. Each Dock row carries a pointer back to the source record (cultureamp_response_id, qualtrics_response_id, lattice_note_id), the agent identity that drafted the synthesis, the HR reviewer who signed off, and the decision timestamp. When the agent needs current state, it re-fetches via fresh API reads against each platform.
The synthesis table
| theme | source pointers | agent draft (1 line) | confidence | reviewer | status |
|---|---|---|---|---|---|
| Manager 1:1 cadence slipping in EMEA eng | cultureamp_response_id: 88412, qualtrics_response_id: 7741, lattice_note_id: L-3320 | EMEA eng reports 1:1 frequency dropped after Q1 reorg; 14 verbatims cite "skipped weeks" | 0.81 | priya@ | approved |
| Onboarding clarity dipped for Aug cohort | cultureamp_response_id: 88501, lattice_note_id: L-3401 | Aug new-hire pulse shows -12 pts on "I know what is expected"; concentrated in support org | 0.74 | priya@ | needs-revision |
| Recognition signal strong in sales | qualtrics_response_id: 7802 | Sales EX score +9 pts QoQ; verbatims credit weekly shout-out ritual | 0.68 | mei@ | approved |
One worked workflow
The agent pulls the latest Culture Amp pulse, the open-end fields from the Qualtrics EX survey, and the past quarter of Lattice 1:1 notes that managers have shared with people ops. It clusters comments by theme, scores each cluster for confidence, and writes one row per theme into the synthesis table above. Each row links back to the exact source records.
The HR lead opens the row, clicks through to the Culture Amp response and the Qualtrics verbatims, and either approves the draft, requests a revision, or rejects it. Approval flips status, stamps the reviewer, and unlocks the memo draft in a Dock doc. Publishing the memo to the leadership channel is gated. The agent cannot post without a second approval. That is the consent gate.
Why this matters
Employee research is high-stakes interpretation. Qualtrics found that only 44% of new employees intend to stay longer than three years, which means the synthesis you ship lands on leaders who are already anxious about retention. A misquoted verbatim or an over-confident cluster does real damage.
SHRM reports that 81% of HR professionals list maintaining employee morale and engagement as a top priority. The bottleneck is not data collection. It is synthesis the HR lead trusts enough to forward. An agent that drafts, cites, and waits for review collapses the cycle from weeks to days without removing the human judgment.
The architecture also gives you an audit trail. Every claim in the memo traces to a source record, a model version, and a named reviewer. When a VP asks where a number came from, you click the row.
Read the full Dock for people ops pillar for the broader pattern, or jump to the Dock for HR overview if you handle the full employee lifecycle.
FAQ
Q: Does the agent see raw 1:1 notes by default? A: Only notes a manager has explicitly shared with people ops. The agent inherits the same permissions as the HR lead and cannot widen scope. See agent identity for the inheritance model.
Q: What stops the agent from publishing a bad memo? A: A consent gate. Drafting a row is one permission. Publishing the memo to a leadership channel requires a second, named approval. See agent audit and compliance.
Q: How does this differ from research synthesis for product or UX? A: The pattern is the same. Sources differ, reviewers differ, sensitivity differs. The Dock for research writeup covers the product-research variant.
Q: Can the same agent identity work across Culture Amp, Qualtrics, and Lattice? A: Yes. One agent identity holds scoped read tokens for each platform. Tokens rotate independently. The identity is the unit of trust, not the platform login. See agent identity.
