This week's priorities
1. Idris Mwangi is leaving in about 30 days, and his exit takes $2.3M of vendor renewals with it. Idris matches all four historical pre-departure indicators (88% confidence), the strongest signal we have. His comp sits 11% below market, he has been stuck at L7 for 20 months when peers move up in 14, his manager Carmen Espinoza cancelled 3 of his last 4 one-on-ones, and his meeting load is 71% over baseline. The part leadership has not connected: he is the Director of Vendor Operations sitting on four renewals worth $2.3M due in 60 days, and knowledge transfer has not started. If he walks, those negotiations land on someone who has never run them. This is not a backfill problem, it is a financial-exposure problem with a 30-day clock. Recommendation: counter-offer conversation this week (comp and a real promotion path), and start documenting the renewal pipeline in parallel regardless of whether he stays. Treat the $2.3M as the thing at risk, not the headcount.
2. Marcus Hale is probably leaving, and he is the only person who understands payments. 78% probability of resignation inside 60 days. The full pre-departure quartet is lit: engagement down 27 points (worst on manager relationship), Jordan Chen cancelled 4 of his last 5 one-on-ones, his LinkedIn says "open," and he has run 32 hours of meetings a week for six straight weeks. Here is what makes this tier-1, not routine: Marcus is bus-factor 1 on the payments integration, sole owner of tier-1 critical infrastructure. His departure is an operational outage waiting to happen, not a vacancy. Note the pattern with Idris: both have managers (Jordan, Carmen) who stopped showing up to one-on-ones right before the engagement crash. That is a manager-behavior signal, not a coincidence. Recommendation: get a second engineer shadowing payments now, this week, before any retention conversation, because the knowledge risk outlives Marcus either way. We do not see his comp history; if a recent adjustment was made, lower the resignation weighting.
3. The Marketing agent is writing to HubSpot ungoverned, and the EU AI Act window closes August 1. The Asset Repurposing Pipeline (Sigil AI pilot) runs with no supervisor and no risk classification while writing to a system of record. 95% confidence this is a compliance flag once enforcement opens, with 88 days to fix it. The consequence leadership has not priced: this is one ungoverned agent we can see. It writes to HubSpot, which means it can corrupt customer-facing records with no human accountable. The real question is whether Demand Gen stood up others we have not catalogued. Recommendation: assign Pavel Novak (Director of Demand Gen) as supervisor and classify the risk tier this week. Cheap to fix now, expensive after August 1.
Watching
Zoë Martin, 21 days to act. 60–65% resignation risk inside 90 days. Engagement cratered 34 points on compensation fairness; she is 14% below market with no adjustment in 18 months, and her 3-year anniversary lands in 21 days. Fix the comp before the anniversary. After it, the conversation anchors differently and money stops solving it. This sits underneath the same comp-fairness problem dragging on Idris.
Sienna Quinn's Mid-Market Sales team. Lost 3 of 14 in 90 days, 4.8x baseline. Exit interviews blame "manager fit," not rep quality. Sienna is a first-time manager 290 days in. Pair her with a senior Enterprise coach for 8 weeks before promoting the two new AEs into her team, or we will burn those two as well.
Engineering at 91% utilization, rising. Six points past the quality-degradation line. This is the soil Marcus's burnout grew in. Pushing harder here makes every other retention risk worse.
Customer Success at 81% change-absorption. Four simultaneous changes in 90 days (re-org, new comp plan, tooling migration, ICP change). They are near the ceiling. Add nothing else this quarter.
Wins
Growth Infrastructure cross-team collaboration is up 18% month-over-month after the platform/product reorg. That reorg is working. Worth studying what went right here, because it is the opposite of what is happening on Sienna's team.
System evolution
Two calibration items for your approval. First, the cancelled-one-on-one signal showed up before both Marcus's and Idris's engagement crashes, with the manager initiating every cancellation. We propose weighting manager-initiated one-on-one cancellations as a leading indicator, not a "watch." Second, we recommend a sweep for other ungoverned agents in Demand Gen and across Marketing, since the Sigil pilot suggests deployment is happening outside the governance process. Both await your sign-off.
System health
Three retention calls (Idris, Marcus, Zoë) rest on the same historical pre-departure pattern, which has matched 3 prior departures with a 47-day mean. That is real but thin, so treat confidence as directional, not precise. Known blind spot: we do not see compensation history or recent comp conversations, which directly affects the Marcus and Zoë calls. Bias check: retention flags cluster in Engineering, which is partly real load (91% utilization) and partly that Engineering is our most instrumented department, so absence of signal elsewhere is not absence of risk. No data anomalies this week.