The Partner Program backbone.
I took over a partner integration program that had real revenue coming in but nothing operational underneath it. So I built the cadence, the source of truth, and the reporting layer it needed. It now sits behind a ~$20M ARR partner-sourced motion.
The Partner Program launches and scales integrations with warehouse and order management system partners that refer merchants into Shippo. Each integration is a distribution channel. Done well, it compounds. Done poorly, it leaks pipeline, frustrates partners, and creates support overhead with no offsetting revenue.
Real partner traction, but nothing operational underneath. Reporting was hand-keyed every week. Partner status was kind of a vibes call. Decisions sat in DMs. Dashboards drifted from each other. Onboarding a new partner was a one-off every time. My job was to turn all of that into something repeatable.
The four pieces that mattered most
Inside the program, these four pieces of work made the biggest difference. In each case I tried to replace the manual work with a system, not just speed up doing it by hand.
Leadership Reporting System Redesign
The bi-weekly leadership update and a separate leadership snapshot page were doing 80% the same job. Updates were 2,000+ words. Asks were buried below tables. The two surfaces drifted on partner status, RAG flags, and OAuth state.
The fix. I collapsed everything into a 3-layer system. A 2-minute leadership email with R/Y/G partner status and the decisions needed pinned to the top. A single always-current Confluence hub as the deeper layer. A dated cycle archive for history. I also standardized a subject-line formula so leadership could triage without even opening the email.
Outcome. Weekly maintenance time went from about 55 minutes to about 35, a 35%+ cut. One source of truth across email and hub. The decisions live in the first 5 lines now instead of being buried.
Confluence Hub Consolidation
32-page hub across 4 folders. Duplicated tables, stale workstream pages, no defined RAG criteria, last refresh date over 30 days old. The hub was simultaneously trying to be a leadership snapshot, an operational tracker, a partner portfolio, and a link directory.
The fix. I cut it down to about 18 active pages (a 44% reduction), organized around audience: leadership, working group, and ops. Standardized a partner page template so any team member can pick up coverage. Replaced hand-keyed tables with embedded live dashboards. Wrote a documented update ritual with cadence and DRIs.
Outcome. One bookmarkable URL for any leadership reader. The manual copy-paste on the highest-traffic data tables is gone. The update ritual now takes a fraction of the time it used to.
Partner Health Dashboard (Databricks)
No partner-specific deep dive existed. Looker dashboards covered program-wide metrics but couldn't show per-partner merchant churn, label-to-rating ratios, or at-risk merchant trends. Partner managers were running ad-hoc SQL or asking the data team for one-offs.
The fix. A 6-page Lakeview dashboard. An overview plus dedicated deep dives for the top 5 partners by label volume. Per-partner KPIs: labels in period, rating calls, label-to-rating ratio, active merchants, new merchants, at-risk merchants over 30 days, and a merchant churn risk table. All parameterized by date range. 27 datasets total. I built it through a reproducible Python build script so the whole thing can regenerate end to end.
Outcome. Partner managers can answer "is this partner growing or stalling?" in under 60 seconds. The ad-hoc SQL pulls are gone. And the build script makes it pretty trivial to add new partners or new metric panels later.
AI-Powered Workflow Adoption
Manual operational lift was eating 40%+ of weekly time. Slack signal triage, partner deep-dives, dashboard regeneration, and update drafting were all hand-keyed. The team kept treating AI as an assist tool when it could replace the work outright.
The fix. I built MCP-based workflows that scan partner Slack channels for signals, generate first-pass leadership updates, and regenerate dashboards from code. Then introduced reusable skills and templates so repeat tasks could become one-shot prompts.
Outcome. Reclaimed multiple hours a week across reporting, signal triage, and dashboard maintenance. Set the standard for AI tool adoption inside BizOps, and adjacent teams are now borrowing the patterns. This work also fed directly into the AI Tool Governance pilot (see the interactive case study).
The principles underneath
01 · Start scrappy, prove it works, then make it nice
- The leadership update started as a Google Doc
- The dedup tool started as a Streamlit script
- The polish came after the value was already proven on real cycles
02 · Replace the work, don't just speed it up
- Manual reporting cycles → systems, not just faster manual work
- Ad-hoc SQL pulls → reproducible dashboards
- Hand-keyed reconciliation → SOPs that get written the second time something happens
03 · Lead with the decision needed
- Leadership emails: R/Y/G status and asks pinned to the top
- Dashboards: "is this growing?" answered in 60 seconds
- AI governance: tier check decides sign-off depth before scoping
04 · Codify the second time it happens
- Second carrier dispute → 5-phase SOP
- Second AI-tool collision → registry and intake
- Second partner deep-dive → Lakeview dashboard with 27 datasets