2025-26
Weedl
AI That Manages, Not Just Measures
weedl.appThe Challenge
Retail operations drowns in dashboards. Every system—POS, ecommerce, loyalty, web analytics—generates its own reports, its own metrics, its own slice of the picture. Operators export spreadsheets, build manual reports, and try to synthesize meaning from disconnected data. The dashboard shows revenue is down. It doesn't tell you why, and it definitely doesn't tell you what to do about it tomorrow.
The missing role isn't another analytics tool. It's an operations manager—someone who looks across all the data, connects the dots, identifies what matters, and surfaces the next action. Someone who remembers what happened last week and uses that context to interpret today. Someone who knows which budtender needs coaching and what specifically to coach them on.
The question: could you build an AI that actually manages operations—that synthesizes, remembers, coaches, and surfaces action—rather than just measuring and displaying?
Core tensions
Dashboards report what happened. They don't synthesize why or prescribe what's next.
Staff performance review happens monthly at best, based on gut feeling, not data. The highest-leverage coaching moments get missed.
Context evaporates. Every Monday starts from zero. The insight from last week's pattern doesn't carry forward.
Managers spend hours building reports that should build themselves.
Approach
Agent, not dashboard
Weedl isn't a place you go to look at charts. It's a system that watches your operations and tells you what needs attention. Daily briefings arrive in your inbox or Slack before you've opened a spreadsheet. The AI has already identified yesterday's anomalies, surfaced the coaching moments, and prioritized today's focus.
Coaching as the output
The core deliverable isn't metrics—it's coaching. Store-level coaching for the GM: what happened, what's trending, what needs your attention. Budtender-level coaching for individual development: where they're improving, where they're struggling, specific actions for tomorrow. Data-backed, context-aware, actually useful.
Memory through state
Most AI tools treat every interaction as a fresh start. Weedl builds compounding context through cascading states. Yesterday's daily state informs today's generation. Last week's weekly summary provides trajectory context for this week. The AI remembers what happened, tracks streaks, identifies patterns, and uses history to interpret new data.
Human-in-the-loop refinement
The AI drafts; humans review. When corrections happen, those corrections become learnings that inform future generations. The system gets smarter over time, calibrated to each store's specific context and preferences.
Technical Implementation
Cascading state architecture
Raw data transforms through a temporal pipeline—Daily → Weekly → Monthly → Quarterly → Yearly—with parallel tracks for store-level and individual budtender states. Each state captures metrics, comparisons, trajectory analysis, and AI-generated coaching. This isn't aggregation; it's synthesis with memory.
The handoff rule
When a period boundary crosses, the first state of the new period reads the closing state of the previous period. First daily of a new week reads the weekly summary. First daily of a new month reads the monthly review. Context cascades forward. The AI never forgets what just happened at the macro level.
State lifecycle
The system never blocks. Draft coaching surfaces immediately with a "Pending Review" badge. Operators see intelligence in real-time; refinement happens asynchronously. When no one reviews for a week, the AI keeps generating. It keeps working.
End of Day → DRAFT (AI generates automatically)
↓
Next Morning → IN_REVIEW (Human reviews, edits if needed)
↓
Anytime → APPROVED (Final version, learnings captured)Learnings capture
When humans correct AI coaching, those corrections become persistent learnings. The AI that suggested a top performer shadow a lower performer after one bad day? That mistake gets captured: "When Joe has a below-average day, do NOT suggest shadowing. He's consistently top performer; bad days are anomalies." Every correction makes future generations smarter.
Coaching schema depth
Store daily states include 15+ metrics with four comparison dimensions (vs yesterday, vs same day last week, vs target, vs MTD pace), trajectory tracking (streak days, trend direction), category and hourly breakdowns, budtender leaderboards, and AI-generated coaching: summary, highlights, concerns, tomorrow's focus. Budtender states mirror this at the individual level with ranking context and personal performance history.
Natural language interface
Beyond automated briefings, operators can query their data conversationally. "Who was my top performer last week?" "Why did Tuesday's revenue dip?" "Compare this week to last week." The AI reads from synthesized states—not re-querying raw data—so responses include trajectory context and historical patterns automatically.
Multi-source correlation
POS transactions, ecommerce orders, web analytics, loyalty data—the platform correlates across sources to answer questions single systems can't. Did the drop in web traffic cause the drop in foot traffic, or was it the other way around? Which online searches are converting to in-store purchases?
Outcome
Deployed and validated
Beta running with active daily briefings, weekly reviews, and budtender coaching. Managers start mornings informed rather than scrambling to build reports.
Staff coaching transformed
Budtenders receive specific, data-backed feedback instead of vague quarterly reviews. Managers know exactly who needs attention and what to focus on. Coaching conversations have context: "You were third in revenue but first in AOV—your attachment rate is strong. Let's talk about volume."
Architecture proven portable
The cascading state pattern, the handoff rule, the draft-review-approved lifecycle, the learnings capture—this entire architecture transferred directly to Ariyah (personal cognitive infrastructure). What works for retail operations works for individual knowledge management. The pattern is domain-agnostic; cannabis was just the proving ground.
The 'ops manager' role materialized
Morning briefings that used to require an hour of spreadsheet work now arrive automatically. The question shifted from "what happened?" to "what do we do about it?"—which is exactly what an operations manager should be answering.
The measure of success is not whether operators can see their data, but whether the intelligence emerging from that data actually changes what they do tomorrow.