From Silos to System:
Building a Single Source of Truth for Upwork's Product Org
From disconnected tools to a single source of truth: redesigning how Upwork's product org plans, tracks, and operates.
Role: Director of Product Operations, Upwork Timeline: August 2025 to present
As Upwork's product organization scaled, a set of structural challenges had become normalized across the org. Teams were executing well within their own lanes, but the tools and processes connecting them were creating compounding friction at every level.
The Problem
Roadmap planning lived in Google Docs and Slides with no consistent format across teams
Roadmap tracking happened in Google Sheets, with separate tabs standing in for team structure
Product development ran through Jira, where every team had built its own custom workflows and field conventions with no standardization across the org
Weekly reporting required each team to produce its own update document, resulting in 70+ docs per week, manually synthesized across verticals into roughly 10 separate executive updates to the CEO each week
With data spread across multiple disconnected tools in inconsistent formats, applying AI-driven automation or intelligence to any of it was not feasible, which was becoming a real strategic blocker
The mandate was clear going in: reduce fragmentation, create consistency, and build the kind of cross-team transparency and trust the current setup was actively working against.
Goals & Vision
Specifically, the org needed to:
Consolidate planning, tracking, and delivery status into as few tools as possible
Establish a consistent structure so that information means the same thing across every team
Eliminate the manual synthesis layer that was creating delay and inconsistency at every reporting level
Build a clean, consolidated foundation that would make AI-driven reporting, insight generation, and automation actually possible
That last point was the north star. The tool consolidation was the prerequisite. The AI intelligence layer was the destination.
Start with discovery, not design. Before any configuration was touched, structured interviews were conducted with scrum teams across Engineering, Product, and Design. The core question on every call: "If this field disappeared, would your workflow break, be annoying, or would you not notice?" That question separated genuine needs from accumulated habit and gave permission to simplify aggressively.
The Approach
What the discovery surfaced:
Teams had built entirely different workflows inside the same tool, using the same field names to mean completely different things
Most custom fields existed because Jira allowed them, not because teams relied on them
The real source of truth was often a Slack thread or a Google Doc; Jira was a mirror, not a driver
77 Jira roadmap fields were mapped, and the vast majority had no Linear equivalent and did not need one
Design the operating model before building anything. Discovery findings became a standards document covering taxonomy, status conventions, required fields, and team structure. Pressure-tested with EPD leaders and the Linear team before any configuration decisions were made.
Make the strategic call nobody asked for. The original brief was to migrate execution tracking. The decision was made to move roadmaps into Linear as well, which is the moment the tool became a genuine single source of truth rather than just another project tracker. VP leadership was aligned first, with the migration framed as a foundational step toward the AI-native org model the company had already committed to.
Roll out in phases with real support. Marketplace went first in December to set the standard. Every business unit followed a structured two-week cycle:
Week 1: PMs onboard and configure 2026 initiatives
Week 2: Engineering migrates and full team access goes live
Each wave included Monday onboarding sessions, Wednesday office hours, and Linear's CS team on standby
Once a team migrated, Jira was set to read-only. No parallel usage. A deliberate, one-way move
Secure the right access. When budget pressure pushed toward limiting Linear to Engineering and Product only, that framing was challenged directly. The value of the migration depended on Linear functioning as the org's operating system, not just an engineering tool. That argument held.
The Results
Phase 1: Foundation (Q4 2025 to Q1 2026)
40+ PMs and 400+ engineers operating on a single, consistent model
Five disconnected tools consolidated into one Linear-based system
Standardized taxonomy, unified roadmaps, and consistent statuses across every team
All business units migrated in waves through Q1, with Jira set to read-only upon migration
Phase 2: Visibility (Q1 to Q2 2026)
Reporting templates unified across all teams inside Linear
A separate Notion pilot was discontinued; equivalent capability built within the existing stack
Teams gained visibility into each other's priorities and progress for the first time
Leadership gained a real-time view of the entire product org without manual synthesis
Phase 3: AI Layer (Q2 2026 and ongoing)
With a single, consistent source of truth in place, the org was positioned to apply an intelligence layer to work that had previously required significant manual effort. Three areas where that came to life:
Customer interview synthesis: Automated the process of synthesizing customer interview transcripts into structured insights, replacing a fully manual research workflow and feeding signals directly into weekly product decisions
Launch governance: Redesigned the launch readiness process with AI-powered automation built natively inside Linear, replacing a manual approval workflow and giving the org real-time visibility into launch readiness across teams
Leadership reporting and hygiene: Ongoing reporting, roadmap summaries, and initiative health are now surfaced through an AI-assisted workflow pulling directly from Linear, replacing a significant portion of the manual synthesis work that previously required hours of coordination each week