Post: Predictive Retention: 9 Cross-Departmental Data Signals to Track
Predictive retention is not an HR-only discipline. The leading indicators of a regrettable departure show up first in calendar density, project routing, and comp benchmarking — three datasets HR rarely owns. This is a practical list of nine signals worth pulling from across the org, with the Make.com orchestration pattern for each.
The framework for connecting those data sources lives in AI-Powered Workflow Automation for Strategic Talent Acquisition — Complete 2026 Guide — the OpsMesh™ pattern for HR teams that need data from systems they do not own.
How the nine signals compare
| Signal | Source system | Lead time | Predictive strength |
|---|---|---|---|
| Calendar density drop | Google Calendar / Outlook | 30-60 days | High |
| 1:1 frequency drop | Calendar / HRIS | 30-90 days | High |
| Project routing change | Jira / Asana / Teamwork | 14-45 days | Medium |
| Code-review participation drop | GitHub / GitLab | 30-60 days | High (engineering only) |
| Internal Slack message drop | Slack | 14-30 days | Medium |
| External LinkedIn activity rise | LinkedIn signals | 60-120 days | Medium |
| Comp band drift | Comp benchmarking tool | 90-180 days | High |
| Comp ratio decline | HRIS | 180+ days | Medium |
| Engagement survey response drop | Engagement platform | 30-90 days | Medium |
1. Calendar density drop
An employee’s calendar usually fills up as they get more embedded in the work. A sustained drop in meeting count over 30 to 60 days is the earliest leading indicator of disengagement we see. The Make.com pattern pulls weekly calendar metadata, computes a rolling average, and flags employees whose density drops more than one standard deviation below their baseline.
- Source: Google Calendar API or Microsoft Graph
- Lead time: 30-60 days before voluntary departure
- Privacy posture: aggregate metadata only, never meeting content
- Make.com pattern: weekly poll, rolling-average comparison, alert to retention dashboard
2. 1:1 frequency drop with manager
Regular 1:1 cadence is one of the strongest engagement signals an org can track passively. When the cadence slips from weekly to biweekly to skipped, the employee is checking out — or the manager is. Either failure mode is worth surfacing. Make.com pulls the calendar pattern, joins it to the HRIS manager-of-record, and flags pairings whose 1:1 cadence has dropped below 75 percent of the org baseline.
- Source: calendar + HRIS
- Lead time: 30-90 days
- Make.com pattern: scheduled scenario, pairwise calendar analysis, HRIS lookup, dashboard alert
3. Project routing change
When an employee stops getting the high-leverage projects their peers get, they read it correctly as a signal about their trajectory. Project management systems hold this data in the task-assignment history. A drop in assigned tasks relative to peers at the same level, over 14 to 45 days, is a medium-strength leading indicator worth correlating with calendar density.
- Source: Jira, Asana, Teamwork, or equivalent
- Lead time: 14-45 days
- Make.com pattern: weekly poll of task-assignment data, peer-cohort comparison, retention dashboard alert
4. Code-review participation drop (engineering only)
For engineering teams, code-review participation is one of the highest-fidelity engagement signals. An engineer who reviews three PRs a week and then drops to one is signaling something — burnout, disengagement, or a side project taking attention. The Make.com pattern pulls GitHub or GitLab review counts weekly and flags drops greater than 50 percent of the engineer’s rolling average.
- Source: GitHub / GitLab API
- Lead time: 30-60 days
- Make.com pattern: weekly poll, rolling average, threshold alert
5. Internal Slack message volume drop
Slack participation tracks engagement — an employee who has stopped posting in team channels has usually mentally exited before formally exiting. The metric is volume in team channels, not DMs (privacy line). A 50 percent drop over 14 to 30 days is the threshold worth flagging.
- Source: Slack analytics
- Lead time: 14-30 days
- Make.com pattern: weekly Slack analytics pull, rolling average, threshold alert
6. External LinkedIn activity rise
An employee updating their LinkedIn profile, increasing connection requests, or following more competitors is a 60 to 120-day leading indicator. The signal is medium strength because plenty of employees update their profile for legitimate reasons unrelated to job search. Use it to correlate with other signals, not as a standalone trigger.
- Source: LinkedIn public profile changes
- Lead time: 60-120 days
- Make.com pattern: monthly profile-change polling for tenure-flagged employees
7. Comp band drift relative to market
The strongest comp-related retention signal is not the employee’s salary in isolation — it is their salary relative to market for the same role. When market moves up faster than the employee’s comp, the gap becomes the resignation incentive. Make.com pulls a comp-benchmarking source quarterly, joins to HRIS, and flags employees whose comp ratio has drifted below market by more than 10 percent.
- Source: comp-benchmarking source + HRIS
- Lead time: 90-180 days
- Make.com pattern: quarterly comp-benchmark pull, HRIS join, gap analysis, retention dashboard
8. Comp ratio decline within the band
An employee whose comp ratio drifts toward the bottom of their band has stalled. Causes include slow merit cycles, missed promotions, or comp bands that moved while the employee’s comp did not. The signal is medium strength on its own but high strength when combined with calendar and project signals.
- Source: HRIS
- Lead time: 180-plus days
- Make.com pattern: quarterly comp-ratio calculation, peer-cohort comparison
9. Engagement survey response drop
An employee who has historically responded to engagement surveys and stops responding is signaling disengagement with the org itself. Most engagement platforms surface response rate as a metric — pull it via API, join to HRIS, and flag employees whose participation has dropped below their personal baseline by more than two consecutive cycles.
- Source: engagement platform
- Lead time: 30-90 days
- Make.com pattern: quarterly engagement-data pull, response-rate trend analysis
Building the cross-departmental data mesh
The reason this is hard is not the analytics — it is the data plumbing. Each of these nine signals lives in a different system owned by a different team. The OpsMesh pattern is to pull a small extract from each system into a single retention analytics store, normalize the schemas, and run the correlation analysis at that layer. Make.com is the orchestration tool that pulls and normalizes. The analytics tool on top is your choice — Looker, BigQuery, or a lightweight equivalent works for most HR teams.
Expert Take
Most HR retention models fail because they only see HR data. Add calendar and project signals and the predictive power doubles. Add code review or sales activity data, where it applies, and the model goes from interesting to actionable. The blocker is rarely the model — it is the data integration. Solve that and the analysts on the team can build the model in a week.
Privacy and consent posture
Every signal on this list is built from metadata, not content. Calendar density uses meeting count and density patterns, never the meeting subject or invitees outside the org chart. Slack participation uses message count by channel category, never message content. Code review uses participation counts, never PR content. Document the data scope, get HR ops and legal sign-off before turning on any signal, and publish the policy to employees. The work-as-easier promise only holds if employees trust the data is being used responsibly.
How we evaluated
Signal strength is based on retention model performance in client engagements where the signal was deployed. Lead time is the median time between signal threshold breach and voluntary departure, across teams that have measured it. Source system is the most common implementation — substitutes work where available.
Related reading
RELATED POST
