| name | daily-action-queue |
| description | Compute "today's top N accounts" for a K-12 EdTech PSM as a ranked list with next-best-action + confidence + rationale. Weighted-signal formula (lifecycle-aware), with RICE / ICE as alternate framings. Every action carries an explicit rationale string naming the dominant signal + the threshold crossed + the prescribed play. Designed so the PSM can answer "why this account today?" in one sentence. |
| last_reviewed | 2026-06-04T00:00:00.000Z |
| confidence | high |
| primary_agent | edtech-partner-success-manager |
| secondary_agents | learning-analytics-analyst, success-playbook-designer |
Daily Action Queue — Skill
What this skill does. Reads a PSM's portfolio + signal table + lifecycle stages + the day's date, and returns a ranked list of accounts with {score, top_3_signals, recommended_action, rationale_string, confidence}. The output is the data behind the Daily Action Center widget in the PSM home page (see ../../knowledge/psm-dashboard-canon-2026.md §2).
What this skill is NOT. It's not a generic "top customers" ranker — it's the math behind the next-best-action surface that names the play. A bare ranking without a rationale string is a regression, not a feature.
1. Contract
1.1 Input
psm_id: "dana.l"
portfolio:
- account_id: "riverside-usd"
lifecycle_stage: "renewal-build"
tier: 1
arr_usd: 240000
renewal_date: "2026-11-15"
segment: "k12"
flags:
esser_funded: true
superintendent_change_12mo: false
state_testing_window_active: false
signal_table:
- account_id: "riverside-usd"
signals:
teacher_adoption: { value: 84, leading: true, half_life_days: 7, last_event: "2026-06-03" }
admin_engagement: { value: 88, leading: true, half_life_days: 14, last_event: "2026-06-02" }
decision_maker_touch_recency: { , , , }
1.2 Output
queue:
- rank: 1
account_id: "pinecrest-isd"
score: 87.3
top_3_signals:
- { signal: "champion_status", value: 0, delta_30d: -100, threshold: "champion departure" }
- { signal: "decision_maker_touch_recency", value: 38, threshold: ">21d zero touchpoints" }
- { signal: "esser_funded", value: 1, threshold: "structural renewal risk" }
recommended_action: "Run Recovery — Sponsor Re-Anchor play"
rationale_string: "Pinecrest ISD: named champion departed + 21d zero meaningful touchpoints + ESSER-funded with Oct renewal → Recovery — Sponsor Re-Anchor play, owner Marcus T., target meeting by 2026-06-09."
confidence: 0.86
play_id: "recovery-sponsor-re-anchor"
suppressed_reason: null
- rank: 2
The rationale_string is the differentiator. Without it, this skill is bare top-K ranking; with it, it's an NBA surface (Inogic — NBA in CRM, accessed 2026-06-04).
2. The weighted-signal formula (default)
2.1 Core scaffold
Account priority score = Σᵢ (weightᵢ × signalᵢ × recency_decayᵢ) × tier_multiplier × lifecycle_multiplier × k12_overlay
Signals are normalized 0-1. Weights sum to 1.0. Convergent scaffold from 5 independent sources (Customers.ai — Recency-Weighted Scoring; Vitally — 4 Metrics; Gainsight — Customer Health Scores — SERP summary; Typewise — Prioritizing Support Tickets; Heap — Lagging to Leading Indicators, all accessed 2026-06-04).
2.2 Recency decay
Per ../../knowledge/health-score-v2-extension.md:
recency_decay(signal) = 2^(-Δdays / half_life_days)
Per-signal-class half-lives (defaults; configurable):
- Sentiment: 7d — fast-decay, surveys + verbal pulses (Customers.ai, accessed 2026-06-04; convergent with Velaris — CS Health Scores).
- Touchpoint: 14d — meeting recency (Factors.ai — Time Decay, accessed 2026-06-04).
- Support: 30d — ticket impact lingers but isn't permanent.
- Adoption / outcome: 90d — delivered outcomes shouldn't decay fast.
- Financial / renewal posture: 180d — slow-moving.
Note: the research lists adoption / outcome at 90d half-life — this overrides the 7d default seen in marketing-attribution contexts because partner adoption isn't a click-event signal (Heap — Leading vs Lagging, accessed 2026-06-04).
2.3 Lifecycle-aware weight vectors
"Early in the journey, usage and onboarding signals may be weighted more heavily, while later, outcomes and adoption patterns become stronger predictors" (Heap, accessed 2026-06-04). The weight vector is a function of lifecycle stage, not a single fixed vector.
| Lifecycle stage | Adoption | Touchpoint | Outcome | Sentiment | Champion | Support | Financial |
|---|
| Onboarding (Day 1-90) | 35% | 25% | 5% | 10% | 10% | 15% | 0% |
| Early-adoption (Day 91-365) | 30% | 20% | 15% | 10% | 10% | 10% | 5% |
| Steady-state | 20% | 15% | 25% | 10% | 10% | 10% | 10% |
| Renewal-build (T-180 to T-90) | 15% | 15% | 25% | 10% | 15% | 5% | 15% |
| Renewal-active (T-90 to T-0) | 10% | 20% | 20% | 15% | 20% | 5% | 10% |
| Recovery | 25% | 25% | 15% | 5% | 20% | 10% | 0% |
These are seed weights; tune against renewal outcomes per partner-health-score-drift.md Step 2 "Hold-out cohort."
2.4 Tier multiplier
Common-tier model (convergent across Velaris — Account Prioritization; Customer Imperative — Portfolio Segmentation, accessed 2026-06-04):
| Tier | Multiplier | Description |
|---|
| 1 | ×1.3 | Top 10-15% ARR — monthly check-ins |
| 2 | ×1.0 | 25-30% — quarterly + automation |
| 3 | ×0.7 | 55-65% — self-serve + reactive |
2.5 K-12 overlay multipliers
From k12-signal-taxonomy.md:
- ESSER-funded account: ×1.3 (structural renewal risk through 2027 [verify-at-use — 2026-06-04]).
- Superintendent / CTO / curriculum-director change in last 12 months: ×1.2 (each, multiplicative — capped at ×1.5 total to prevent runaway).
- State-testing-window-active: suppression — usage signals contribute 0 to score during window; recovery rate post-window participates in next-day score.
- Funding-source unconfirmed AND in renewal-build stage: ×1.4.
3. Recommended-action mapping
The rationale string names the play. The play mapping is a lookup, owned by the success-playbook-designer agent:
| Dominant signal class | Lifecycle stage | Play |
|---|
| Champion departure | any | Recovery — Sponsor Re-Anchor |
| Admin disengagement + teacher usage holding | renewal-build / renewal-active | Recovery — Leadership-Watch Outreach |
| Roster sync errors active | any | Implementation — Rostering Diagnostic |
| State-testing-window post-recovery <50% | early-adoption / steady-state | Recovery — Post-Testing Re-Engagement |
| License claim rate <50% by Day 14 (Activation Watch) | onboarding | Onboarding — Phase 2 Recovery |
| Decision-maker touch recency >30d AND renewal-build | renewal-build | Renewal — Sponsor Confirmation Sync |
| Sentiment NPS drop >20 over 60d, admin persona | any | Account 360 — Diagnose persona |
| Outcome milestone met + sentiment up + tier-1 | steady-state | Expansion — Value-Trigger Outreach |
| Health composite stable, tier-3, low touch | steady-state | Maintain — async check-in |
Plays are defined under the success-playbook-designer agent's owned skills (renewal-play-design, expansion-play-design, recovery-play-design).
4. The rationale string — anatomy
Every entry in the queue produces a rationale string that names:
- The account (named, not "Account #4")
- The dominant signal + the threshold crossed (with numbers)
- The lifecycle context (so the PSM knows why now)
- The prescribed play + owner + date target
4.1 Good rationale strings
"Westfield USD: usage -34% over 14d in Phase 2 onboarding (target: ≥75% weekly active by Day 30 — currently 41%) → run Phase 2 Recovery play, owner Dana L., target meeting by 2026-06-09."
"Granite State University: NPS 'we're evaluating alternatives' + 2 consecutive support escalations to leadership tier in renewal-active stage (T-57) → Recovery — Sponsor Re-Anchor, owner Priya R., target exec sponsor call by 2026-06-07."
"Northshore Academy: outcome milestone 'family activation >85%' met + sentiment +12 over 60d + tier-1 in steady-state → Expansion — Value-Trigger Outreach, owner Priya R., propose tier upgrade at next QBR 2026-06-25."
4.2 Bad rationale strings (regressions to top-K ranking)
"Pinecrest ISD: low health score." ← no signal named, no threshold, no play.
"Cedar Valley is trending down." ← no number, no lifecycle context, no action.
"High-priority account." ← reveals nothing; the PSM has to do the work themselves.
The hook in ../../hooks/flag-psm-anti-patterns.sh should be extended to flag rationale strings that omit any of (signal name | numeric threshold | play | owner | date).
5. NBA confidence — what to report
Every recommended action carries confidence: 0.0-1.0 per the cross-plugin NBA convention (Inogic — NBA in CRM, accessed 2026-06-04: "Each potential action … is given a score based on expected impact … Recommendations surface with priority levels and confidence percentages — for example, Recommendation A at high priority with 92% confidence").
5.1 Confidence calculation
confidence = min(
signal_freshness_score, # 1.0 if all top-3 signals < half-life; decays otherwise
play_match_strength, # 1.0 for exact lookup match; 0.6 if best-match fallback
rationale_completeness # 1.0 if all 4 anatomy components present; -0.2 per missing
)
5.2 Confidence bands
- ≥ 0.85 — present in queue, recommended for action today, auto-triggers Cited-Adjudicator review per
../../../ravenclaude-core/rules/agent-collaboration.md.
- 0.70 - 0.84 — present in queue, surfaced for review; PSM judgment required.
- < 0.70 — held back; surfaced as "low-confidence candidates" in a separate small-multiples panel, not the main queue. Avoids alarm fatigue (Activu — SOC Best Practices, accessed 2026-06-04).
6. Alternate frameworks — RICE / ICE
Sophisticated teams sometimes apply PM-prioritization frameworks to daily-queue ranking. Expose as alternates via framework: rice|ice.
6.1 RICE (Reach × Impact × Confidence ÷ Effort)
Borrowed from product roadmap prioritization (ProductPlan — RICE; Intercom — RICE Prioritization, accessed 2026-06-04).
PSM-adapted:
- Reach — number of stakeholders / classrooms / schools affected by the play (e.g., a district-wide rostering fix has high reach; a single-champion re-anchor has low reach).
- Impact — ARR at risk × probability the play addresses it (0.25 / 0.5 / 1 / 2 / 3 scale).
- Confidence — PSM's confidence the play succeeds (50% / 80% / 100%).
- Effort — PSM-hours to execute (1 / 2 / 5 / 8 scale).
Useful when the queue contains items of very different sizes — a 1-hour async check-in shouldn't outrank an 8-hour escalation prep just because the weighted-signal score is higher.
6.2 ICE (Impact × Confidence × Ease)
Lighter-weight; suitable for action-by-action triage (Kaizenko — ICE / RICE / Weighted Scoring, accessed 2026-06-04). Used as the default fallback when Effort estimation isn't available.
6.3 When to use which
| Framework | Use when |
|---|
weighted (default) | Day-to-day operational queue. The PSM has a defined book; signals are the source of truth. |
rice | Quarterly book-rebalance; deciding which plays to invest meaningful effort in over a longer horizon. |
ice | Triage when many small actions compete and effort isn't easily quantified. |
The weighted-signal default is correct for ~90% of PSM operational use. RICE / ICE exist for the planning-mode cases the manager and PSM together review.
7. Worked examples
7.1 Example A — Recovery account (high-confidence)
Input:
- Account: Pinecrest ISD, K-12 district, ARR $180K, renewal 2026-10-01.
- Lifecycle: renewal-build (T-119).
- Tier: 2.
- Flags: ESSER-funded = true; superintendent_change_12mo = true.
- Signals (most recent):
champion_status: 0 (departed 2026-05-08) — leading, half-life 30d → recency_decay = 2^(-27/30) = 0.54decision_maker_touch_recency: 38 (last touch 2026-05-08 = 27 days ago) — leading, half-life 30d → 0.54teacher_adoption: 40 (last event 2026-06-02) — leading, half-life 7d → 2^(-2/7) = 0.82admin_engagement: 38 — leading, half-life 14d → 0.91sentiment: 52 — leading, half-life 7d → 0.82outcome: 50 — lagging, half-life 90d → 0.99
Calculation (renewal-build weights):
base = 0.15×40×0.82 + 0.15×38×0.91 + 0.25×50×0.99 + 0.10×52×0.82
+ 0.15×0×0.54 + 0.05×ticket_score + 0.15×financial_score
≈ 0.15×32.8 + 0.15×34.6 + 0.25×49.5 + 0.10×42.6 + 0.15×0 + ...
≈ 4.9 + 5.2 + 12.4 + 4.3 + 0 + ...
≈ 26.8 (raw composite, 0-100, lower = sicker)
Convert to urgency: urgency = 100 - composite = 73.2.
Apply multipliers:
- Tier 2: ×1.0
- Lifecycle renewal-build: weights already applied
- K-12 overlay: ESSER ×1.3 × superintendent_change ×1.2 = ×1.56 (capped at ×1.5)
Final score: 73.2 × 1.5 = 109.8 → clamp to top of band; this account is rank 1.
Top 3 signals (by leading-indicator-weighted contribution to deficit):
champion_status (0; threshold "champion departure" crossed)decision_maker_touch_recency (38; threshold ">21d zero touchpoints" crossed)esser_funded flag (priority multiplier active)
Recommended action: Recovery — Sponsor Re-Anchor (mapped via §3 table; "Champion departure" → this play).
Rationale string: "Pinecrest ISD: named champion departed 2026-05-08 + 27d zero meaningful touchpoints + ESSER-funded with 2026-10-01 renewal (T-119, renewal-build) → Recovery — Sponsor Re-Anchor play, owner Marcus T., target sponsor call by 2026-06-09."
Confidence:
- signal_freshness_score = 0.82 (top signals' decay weighted)
- play_match_strength = 1.0 (exact lookup match)
- rationale_completeness = 1.0 (all 4 anatomy components present)
- confidence = min(0.82, 1.0, 1.0) = 0.82
7.2 Example B — Onboarding Phase 2 (mid-confidence)
Input:
- Account: Westfield USD, K-12 district, ARR $95K, contract started 2026-04-15.
- Lifecycle: onboarding (Day 50).
- Tier: 2.
- Signals:
teacher_adoption: 41 (target ≥75% by Day 30); license_claim_rate: 62; roster_sync_error_count: 3 active.
Top 3 signals:
teacher_adoption (41; threshold "≥75% weekly active by Day 30" crossed downward — actually a -34% drop over 14d)license_claim_rate (62; threshold "≥80% by Day 14" missed)roster_sync_error_count (3 active; threshold "0 active errors" crossed)
Recommended action: Onboarding — Phase 2 Recovery.
Rationale string: "Westfield USD: usage -34% over 14d in Phase 2 onboarding (Day 50, target ≥75% weekly active by Day 30 — currently 41%) + 3 active roster errors → Onboarding — Phase 2 Recovery play, owner Dana L., diagnostic call + rostering escalation by 2026-06-09."
Confidence:
- signal_freshness_score = 0.95 (all signals < half-life)
- play_match_strength = 1.0 (exact match)
- rationale_completeness = 1.0
- confidence = 0.95
7.3 Example C — Suppression case (dead-zone)
Input:
- Account: Mesa Community College, higher-ed, contract steady-state.
- Date: 2026-12-23 (winter break).
- Signal:
teacher_adoption: 18 (would normally trigger yellow).
Computation:
state_testing_window_active = false but calendar dead-zone = "winter break" per k12-psm-operating-cadence.md §2.- For higher-ed: finals week + winter break suppression active.
Output:
- rank: null
account_id: "mesa-cc"
score: 0
suppressed_reason: "winter-break dead zone active 2026-12-22 through 2026-01-02; resume signal evaluation 2026-01-03"
recommended_action: null
rationale_string: "Mesa Community College: signal suppressed — winter-break dead zone (2026-12-22 → 2026-01-02). Resume at 2026-01-03."
The suppressed item appears in a "Suppressed today" collapsible section, not the main queue. Avoids alarm fatigue (Activu — SOC Best Practices, accessed 2026-06-04).
8. Trigger thresholds (practitioner-validated)
From Lyniro — 15-Play Playbook and Planhat — Churn & Retention (accessed 2026-06-04):
| Trigger | Threshold |
|---|
| Onboarding risk | <50% license utilization in first 30 days; 2-week decline in WAU during onboarding; no client activity for 10+ days; 2+ blocked tasks simultaneously |
| Renewal | health score enters defined risk band within 90 days of renewal; outreach 90 days before renewal for annual contracts (K-12 overlay: 120-180 days per renewal-pricing-conversations-edtech.md) |
| Expansion | customers hitting plan limits; regularly engaging with advanced features |
| Churn | product usage drops below baseline for 2 consecutive weeks; health score <threshold; 2+ churn signals simultaneously |
These are the calibration anchors for the threshold language in rationale strings ("threshold X crossed" — point to one of these).
9. Anti-patterns this skill must flag
- Rationale strings without a named signal, threshold, play, owner, AND date (all 4 required — see §4).
- Confidence reported without the three sub-scores (signal_freshness / play_match / rationale_completeness).
- A queue item firing during a dead zone without
suppressed_reason populated.
- A score that places a tier-3 self-serve account above a tier-1 high-ARR account without the tier multiplier having been applied.
- An ESSER-funded account in renewal-build stage that doesn't carry the priority-multiplier flag.
- A weighted-signal output where weights don't sum to 1.0 (rounding to 0.99-1.01 OK).
- A persona-aggregated signal where the per-persona breakdown contradicts the aggregate (e.g., teacher_adoption holding at 84 but admin_engagement at 18 → the aggregate "adoption: 51" hides the real signal; surface both).
- A renewal-build / renewal-active item where decision-maker touch recency >30d AND no Recovery — Sponsor Confirmation Sync play has fired.
10. Output Contract (skill-level)
The skill returns the queue, and (per the plugin's Output Contract in ../../CLAUDE.md §6) the calling agent emits:
Status: ✅ | ⚠️ partial | ❌ blocked
Files changed: <relative paths or "none">
Partner / segment context: PSM portfolio (segment K-12 / higher-ed / corp-ld / mixed)
Signals cited: (per-item, named in the rationale strings)
Followups: top-N items with owners + dates (from rationale_string)
Open questions: any low-confidence items (<0.70) held back
Grounding checks performed: confidence sub-scores computed; dead-zone suppression rules applied
Plus the cross-plugin Structured Output Protocol JSON block.
11. References (existing plugin artifacts)
12. Refresh triggers
- The convergent-weight-vector consensus shifts (e.g., new academic research on CS health-score validity).
- A new lifecycle stage emerges in the PSM's book (e.g., post-merger consolidation as a distinct stage).
- A play in §3's mapping table is retired or replaced.
- Confidence calibration shows confidence bands don't predict play-success rates (per
partner-health-score-drift.md Step 5 quarterly check-in).
- Vendor NBA conventions evolve away from the [confidence + rationale] standard.
