admin-baked/skills/dispensary/low-performing-promo-diagnosis

Low-Performing Promo Diagnosis

Purpose

Turn campaign performance data into an opinionated root-cause diagnosis so operators don't repeat the same mistake twice — producing a bounded optimization memo with specific fixes to test on the next send.

When to Use

• A campaign was sent and open rate, click rate, or attributed revenue fell below benchmark
• Operator asks why a promo didn't work or what to change next time
• Craig's post-campaign analysis step fires in a playbook
• A/B test results are available and need interpretation

When NOT to Use

• Campaign planning or copy drafting → use craig-campaign for forward-looking work
• Compliance violations in the campaign → route to Deebo; diagnosis comes after compliance is cleared
• Real-time campaign monitoring → this is a post-send retrospective skill, not a live monitor
• Campaign hasn't sent yet → nothing to diagnose; use craig-campaign to strengthen the brief first
• Attribution modeling → BakedBot tracks a send-to-transaction window; do not over-claim causal lift

Required Inputs

• campaign_id — to pull send/open/click/bounce/revenue data
• org_id — for segment and baseline context
• benchmark — optional; defaults to org's prior 90-day campaign averages

Reasoning Approach

Diagnose in layers — each layer has a different fix. Don't jump to copy quality before ruling out delivery and audience issues.

Layer 1: Delivery Did the message reach inboxes/phones?

• Bounce rate > 5% → list hygiene problem, not copy problem
• SMS delivery failures → carrier filtering or opt-out list stale
• Email open rate < 10% → subject line or sender reputation issue (check before assuming body is bad)

Layer 2: Audience Was the right segment targeted?

• Segment size < 50 → too narrow; statistically unreliable
• Segment is wrong tier for goal (e.g., sent a winback offer to already-active customers)
• Send time mismatch (Tuesday morning for a weekend event)

Layer 3: Offer Was the offer compelling?

• Low margin discount without urgency → customers have seen it before
• No clear expiry → no reason to act now
• Offer doesn't match segment need (VIP gets same offer as at-risk → feels generic)
• GM elasticity check: did the offer lose money even with the lift it created?

Layer 4: Copy & creative Only reach this layer if Layers 1–3 are clean.

• Subject line too vague → no reason to open
• CTA buried below the fold or unclear
• SMS too long (>160 chars forces a split message, kills conversions)
• Tone mismatch with brand voice for that audience

Benchmarks for context (NY cannabis, BakedBot orgs):

| Metric | Below Expectation | On Target | Strong | |--------|------------------|-----------|--------| | Email open rate | < 15% | 20–30% | > 35% | | Email click rate | < 1% | 2–4% | > 5% | | SMS conversion | < 2% | 4–8% | > 10% | | Attributed revenue / send | < $0.50 | $1–3 | > $5 |

Output Contract

## Campaign Optimization Memo — [Campaign Name] — [Date]

CAMPAIGN: [name] · Sent: [date] · Channel: [SMS/email] · Audience: [segment, N recipients]
PERFORMANCE: Open X% · Click X% · Conversion X% · Attributed revenue $X (vs $X benchmark)
VERDICT: [Underperformed / On target / Overperformed]

### Root Cause (Primary)
**Layer [1–4]: [Delivery / Audience / Offer / Copy]**
[2–3 sentences: what the data shows, why this is likely the cause]

### Contributing Factors
[Bulleted secondary issues — labeled as confirmed or suspected]

### Priority Fixes for Next Send
| # | Change | Expected Impact | Confidence |
|---|--------|----------------|-----------|
| 1 | [Specific change] | [+X% open rate est.] | High/Medium |

### Test Recommendation
[One specific A/B test to run on the next send to validate the diagnosis]

### Retest or Abandon?
[Clear recommendation: retest with fix / abandon this approach / escalate to super_user]

Edge Cases

• No click tracking available (SMS): Diagnosis limited to delivery + conversion; note the tracking gap
• Too small to be statistically meaningful (< 50 sends): Flag; don't over-interpret 2 conversions vs 3
• Campaign sent during an outage or technical incident: Separate the technical failure from the creative before diagnosing
• High open + low click: Copy or offer is the problem, not delivery — focus Layer 4 analysis
• High open + high click + low conversion: Landing page or checkout problem — outside Craig's domain; flag for Linus

Escalation Rules

• Attributed revenue is negative (discount exceeded lift): Escalate to super_user with margin impact figure before recommending a retest
• Bounce rate > 10%: Halt future sends to this list — route list hygiene fix to Linus before next campaign
• Legal complaint or opt-out spike (>2%): Route to Deebo immediately; do not diagnose as a performance issue

Compliance Notes

• Attribution is a send-to-transaction window estimate — never present as proven causal lift
• Segment data used in diagnosis is internal only — not surfaced in customer-facing communications
• If campaign copy included content flagged by Deebo, note this as a potential delivery suppression cause