Ideas for using AI to move HBW outcomes — category visibility, prevention, growth. Not setup polish, actual category leverage.
Every night, automatically check Rx weight loss and ED deal pages to make sure they have the right legal language (FDA-approved vs compounded distinction, FTC subscription disclosure, eligibility callouts). New violations get flagged to Safet automatically.
Rx / GLP-1
You have separate signals per merchant across Rx tracker, LHR sprint, Botox floor, RF Micro, refund review, and Portfolio Adoption. Reps have to look in six places to know if a merchant is healthy. Combine those signals into one grade per merchant — one number a rep can act on.
HBW-wide (every L4)
The Northstar tool now has defensible 30/90-day GP + MVR targets for 415 PDSes. Wire those targets into a nightly check on every new HBW deal launched in the past 30/90 days and flag any deal running below its cohort p25 by a set margin. Reps and AMs get a same-week signal that a launch is fading, not a 90-day retro.
All HBW (starting with L3 High End PDSes)
Pull public prices, package sizes, and unit-of-sale from the booking pages of every live merchant in cosmetic injections and CoolSculpting. Compare each Groupon deal to what the same merchant charges direct, and rank the biggest gaps in both directions. Below-cost and way-above-market deals both refund at higher rates.
Cosmetic Injections + CoolSculpting (L4 Botox, L4 Plastic Surgery)
You have detailed refund-reason clustering for Plastic Surgery, Botox, and Skin Care. Rx (GLP-1, TRT, ED) doesn't have one yet, and the tracker only covers pre-screen — not post-launch refund patterns. Build the refund-driver template for Rx now, on the small live sample, so it's ready when supply doubles.
Medical Weight Loss (GLP-1), Men's Health (TRT), ED
Mother's Day contamination made RF Microneedling and SEV Laser look like decliners when they weren't. Q4 has Black Friday, holidays, and gifting spikes that will distort every L4 the same way. Build a seasonality-adjusted overlay that flags when a YoY read is holiday-driven vs. real category movement, before you interpret the number.
HBW-wide (every L4 in your book)
Half the Q2 injectables refunds trace back to candidacy mismatch — customer bought Sculptra or PDO threads without knowing they weren't a candidate. Build a standard pre-purchase questionnaire (skin type, expectations, contraindications, unit count needed) that reps can drop into any injectables deal page. Same module across every merchant.
Aesthetic Injectables (Botox, Sculptra, PDO threads, fillers)
Northstar showed the same PDS hits very different GP and MVR ranges across the 8 sales regions. Turn that into a per-PDS × per-region view that flags 'this region under-indexes on Facial by 40% — investigate supply mix, pricing, or rep coverage.' Region GMs get a specific list of gaps, not a national average.
All HBW, broken by sales region
Recurring manual work that costs you real hours each year. Each candidate shows what it costs now, what it would cost after, and total annual hours back.
When Al's Revenue Management team hands over quarterly target dashboards, Katelyn currently reverse-engineers the schema (M0/NM header positions, subcategory placement) by hand before she can push back on numbers. This turns that one-off parse into a script and codifies the target-vs-seasonal-history chart as a template. Lives in the quarterly target review ritual with Mark on the thread.
When: Quarterly · 4x/year
Today: 2-3 hours per quarter of manual schema parsing and chart building
After: Feed the dashboard in, get validated L3 numbers plus seasonal comparison chart out
Annual: ~10 hours/year + sharper data-backed pushback
Pattern extension
Enforces that any restricted-tier merchant sitting at Not-Certified auto-routes to HIGH exclude, badges the tab header when screening dates go overdue, and posts a Monday chat digest of merchants past due. Lives in the Rx tracker workflow that feeds the weekly Rx re-screen ritual. Without it, restricted merchants can sit in wrong buckets and overdue re-screens slip past Katelyn's weekly review.
When: Weekly · 50x/year
Today: 15-20 min manually eyeballing bucket assignments and dates
After: Rules enforce placement automatically; overdue merchants surface on Monday without Katelyn scanning
Annual: Prevention-value: $2K-10K (compliance risk from mis-bucketed restricted merchants)
Drift detection
Once the Foundry AI team confirms the supported path (service account, raised query cap, or blessed snapshot pattern), implement it so the HBW Enterprise Tracker no longer depends on Katelyn's personal query allowance. Lives underneath the weekly HBW Enterprise dashboard the sales team reads. Without it, the tracker will silently break again the next time query limits tighten.
When: Weekly · 50x/year
Today: Full outage until Katelyn re-runs manually — dashboard dark for the sales team
After: Weekly pipeline runs from durable infrastructure, no per-user caps in the loop
Annual: Prevention-value: $5K-15K (sales team decision-support downtime)
Recurring fix
Several HBW dashboard refreshes (Rx Weekly, Cosmetic Injections wrappers, HBIC Pulse) share the same Foundry AI query pool and hit rate limits during Monday morning refresh windows, causing partial failures and silent retries. A dedicated service account or reserved slot pool would let those pipelines run cleanly without competing with other users. Blocked on Foundry AI support response.
When: Weekly · 50x/year
Today: 15 min per failed refresh (diagnosis + rerun)
After: Monday morning refreshes finish on the first try instead of needing manual reruns
Annual: Prevention-value: $8,000-15,000 (stops shipped-with-stale-data incidents like Rx dashboard 2026-09-21)
Recurring fix
Watches Katelyn's persistent memory index and triggers the consolidate-memory skill the moment it crosses 200 lines. Lives in the CoS system hygiene layer that feeds every session's opening context. Without it, the index bloats past its load limit and half the memory silently drops out of Claude's context — which is exactly the 236-line drift the last audit caught.
When: Continuous · triggers ~6-10x/year
Today: never done systematically
After: Memory index stays under its load limit automatically; no more silent context truncation at session start.
Annual: Prevention-value: $2,000-4,000 (avoids repeated re-corrections when memory silently drops)
Drift detection
On the PDS Review Page where Katelyn audits engine recommendations, this would auto-flip the ~72 clear pattern misses that the engine already scores as high-confidence, so she only clicks through the genuinely ambiguous ones. It sits inside the PDS review workflow that feeds the taxonomy overlay. Without it, she spends the same clicks on obvious cases as on the judgment calls, which is where her time actually matters.
When: Per PDS review cycle · ~4-6x/year
Today: 30-45 min of clicking per cycle
After: Only the ambiguous ~20% surface for her review; the obvious hits auto-resolve
Annual: ~3 hours/year
Recurring fix
When Legal or Katelyn issues a scope like 'block all prescription-only services' or 'pause all LVG-gated PDS,' she currently reads the Master v4 PDS tracker by hand to build the CSV Tom needs to action the block. A generator would take the scope statement, match it against the tracker, and output a validated CSV with pds_cat_id, L4, and compliance-note flags. Lives in the Rx/Local Flavor triage workflow — cuts each block/pause/audit request from 45 minutes to 5.
When: Monthly · 12x/year
Today: 45 min per occurrence
After: Paste scope statement, get a Tom-ready CSV validated against Master v4 in 5 minutes.
Annual: ~8 hours/year
Recurring fix
The Meeting Vault archives every meeting with structured action items, but there's no single place that pulls out just the items where Katelyn is the owner, or groups meeting context by project (Rx Programs, HBW Supply Gaps, etc.). This would create one 'my open commitments' file that feeds the Friday 5/15 draft, plus per-project rolling state files that feed project prioritization. Wait 2-3 weeks of real vault usage before building so the design matches actual query patterns.
When: Weekly · 50x/year
Today: 20 min per 5/15 draft to hunt through meeting notes for open items
After: 5/15 draft pre-populates with the week's open commitments; project reviews open with fresh context
Annual: ~17 hours/year
Distribution
New skills you could build once and reuse across many projects. Each one compounds — first use justifies it, every reuse after is free.
Adds deal-structure scoring as a permanent step in the cosmetic injections pipeline and builds a Deal Structure Impact tab on the neurotoxin market intelligence dashboard. Lives in the HBW Category Intelligence workspace where Katelyn reviews Botox/injectables performance. Without it, the scoring stays a one-off analysis and can't be used week-over-week to spot which deal structures are driving refunds or margin loss.
When: Weekly · 50x/year
Today: never done systematically
After: Deal-structure scores refresh every week alongside the rest of the neurotoxin dashboard, with a dedicated tab showing which structures win and which drag margin.
Annual: ~40 hours/year
Pattern extension
Adds a permanent 'closed' state to the GLP-1 Weekly Sweep queue so merchants that were escalated-and-resolved or fully paused stop reappearing in the review list forever. Lives in the Monday GLP-1 sweep review that feeds compliance escalations. Without it, the queue silently grows every week with items already handled, and real new problems get buried in noise.
When: Weekly · 52x/year
Today: 10-15 min per week re-triaging already-handled merchants
After: Resolved and paused merchants retire from the queue permanently; only genuinely new or re-flagged deals surface
Annual: ~10 hours/year plus prevented escalation misses
Recurring fix
Once Foundry AI restores access to the deal-content and Salesforce tables, run one clean end-to-end test of the HBW Restricted Services Screening dashboard to confirm nothing broke during the outage. If access drags past a week, build a browser-based scraper as a stopgap so screening doesn't stall. Later extension: add hyaluron pen keywords to the watchlist.
When: One-time verification + optional stopgap
Today: Screening dashboard currently blocked — high-risk deals not being flagged
After: Screening back online with proven end-to-end trust, or scraper fallback covering the gap
Annual: Prevention-value: $50K-200K in avoided compliance exposure per missed high-risk deal
Drift detection
Watches the Foundry AI Agnes query pool every 15 minutes and sends a Google Chat message the moment slots free up after being maxed. Fires during Rx Weekly Performance dashboard refresh failures (Monday morning ritual). Without it, Katelyn either burns cycles poking Agnes manually or misses her WBR window because she didn't know a manual re-fire would work.
When: Weekly · ~50x/year (fires only when slots choke)
Today: ~15-30 min of manual polling per stuck refresh
After: One passive Chat ping tells her exactly when to re-fire
Annual: ~15 hours/year + prevents 1-2 broken WBR dashboards/quarter
Drift detection
Runs a BigQuery scan of live GLP-1 deals and flags merchants that aren't in the Rx pre-screen tracker, catching LegitScript-certified in-clinic clinics the Salesforce-opportunity-based lead sources missed. Lives as Task B in the 9/15 Rx Programs handoff, feeding the in-clinic LegitScript-certified GLP-1 lead pool. Without it, safe-to-pursue clinics stay invisible and the lead pool leans on Salesforce coverage that has known gaps.
When: Monthly · 12x/year
Today: never done systematically
After: A monthly BigQuery pass surfaces LegitScript-certified in-clinic GLP-1 merchants that aren't yet in the tracker, ready to hand to Sales Ops as leads.
Annual: ~24 hours/year + Prevention-value: $50k-150k in missed qualified GLP-1 lead pipeline
Pattern extension
Weekly sweep of open asks and pending build items on the Injectables Market Intelligence dashboard — Colin's margin-by-brand request, Safet's preferred-structure tab, and the remaining composite score v2 additions (refund guardrail, adjacent-market flag, supply flow, dollar ceiling). Lives in Katelyn's Monday dashboard-review ritual before the injectables category standup. Without it, cross-team asks sit open for 9+ days and the composite score ships incomplete.
When: Weekly · 50x/year
Today: 45 min per occurrence
After: One consolidated pass that pulls open asks from email/Asana, updates the dashboard, and drafts the reply-backs in Katelyn's voice — ready to send.
Annual: ~35 hours/year
Coordination
Before the weekly HBW Industry Intel digest posts to Groupon IQ, scan the finished write-up for words that don't belong in an outside-facing industry read — Katelyn's name, MBR references, PDS jargon, supply-audit framing, or anything that sounds like internal Groupon shorthand. If any slip through, block the publish and flag them so the wording gets fixed first. Today this audience check happens by eye, which means internal phrasing leaks through when Katelyn is moving fast.
When: Weekly · 52x/year
Today: 5-10 min per digest (when remembered) — often skipped
After: Pipeline refuses to publish and lists the offending phrases; Katelyn edits once and reruns
Annual: ~6 hours/year plus prevention-value: keeps industry digest from reading like an internal Groupon memo
Drift detection
The duplicate-check logic used in the fai-agnes migration audit currently runs as a one-off. Rolling it into the weekly Lead Framework sweep as an auto-appended section would surface duplicates in every review with the same match rules and per-row escalation helper. Without this, duplicate merchants slip through weekly reviews and get caught later during ad-hoc audits.
When: Weekly · 50x/year
Today: 20 min per occurrence
After: Every weekly LF sweep ends with a duplicate-check section, pre-populated with match rows and escalation prompts.
Annual: ~17 hours/year
Pattern extension
Improvements to your own automation and dashboards. Surfaced automatically from things your past wins have pointed at.
Weekly scheduled check that compares docs/projects.md to projects/ on disk and flags drift — prevents the 38 ghost entries from accumulating again
surfaced from September system audit followup — executed 5 of 6 remaining cleanup items (inner root, registry, deal skills, parent-wrapper, WBR W39 publish) · 2026-W40
A weekly cron that checks MEMORY.md byte-size against the loader cap and gchat-pings when it exceeds — same watchdog pattern as the freshness-sentinel skill.
surfaced from MEMORY.md compression — stripped 22 merged-tails + consolidated 9-way DISCIPLINE bundles line · 2026-W39
Nightly memory-hygiene sweep (referenced in the rewritten feedback_proactive_memory_hygiene.md as the ideal state — currently manual). Would prevent MEMORY.md from creeping back over cap.
surfaced from System audit + memory consolidation — full setup review, 5 skill/rule patches, and 25-merge memory reduction across 5 batches · 2026-W39
L3-level YoY seasonality overlay on the trend chart -- matches Mother's Day contamination issue Katelyn flagged in May 2026 MBR framing
surfaced from HBIC Pulse tab refactor + AI Maturity audit + Dusan extractor fix · 2026-W39