Designing a Labor Intelligence System
A labor budgeting and forecasting experience embedded directly into scheduling, enabling enterprise teams to make confident labor decisions in real time.
Enterprise workforce management and administrative platforms
Discovery, synthesis, UX strategy, interaction design, IA, annotated specs, edge case coverage, trust and governance patterns
STACK




CONTEXT
Labor budgeting was built for reporting and configuration. Managers needed it for decisions.
Labor is one of the largest controllable cost, but most tools reveal budget coverages after schedules are published. That time lag causes a predictable failure: managers make high stakes staffing decisions without cost visibility, then clean up the damage later.
PROBLEM STATEMENT
How might we help operators make staffing decisions with confidence on the scheduling grid, with numbers they trust?
WHAT WAS ALREADY IN PLACE
The labor budgeting model was powerful but manual and disconnected from scheduling.
This structure supported planning, but it did not reliably influence the minute by minute scheduling decisions.
COMPETITOR RESEARCH
I reviewed how adjacent workforce tools connect forecasting, budget thresholds, and scheduling behavior.
Decision guidance belongs in the scheduling grid with consistent totals and preserved context across views. Agents propose and explain actions, but execution remains human-controlled through review and permissions.
Solution
AI Setup
Making real-time budget logic usable for frontline managers.
Labor budgeting is data-dense and runs in real time, but many end users were early-career managers with limited budgeting context. A major source of failures was importing existing budgets without a knowledgeable admin or finance lead validating assumptions, mappings, or edge cases. Bad inputs looked “successful” on upload, then surfaced later as broken totals, misallocated hours, and reactive cleanup.



More manual configuration and deeper drilldowns would have increased iteration cycles and made imports harder to recover.
That constraint pushed us toward agentic guidance: lightweight inputs up front, stronger validation during import, and proactive notifications that help managers correct issues before budgets cascade into schedules.To shift labor budgeting forward in time, the system needed to start with intelligence, not inputs. This reframed setup as a conversation rather than a form.


Ideation
Keeping budget context across time views
Labor decisions rarely live at a single level of time. Budgets are set monthly, schedules are built weekly, and staffing decisions happen in minutes. The challenge was not defining more tasks. It was preserving context as users moved across time without losing confidence in the numbers. I explored how a manager might start with a high-level budget and move step by step toward granular detail while always understanding how each decision connected back to the whole.
We originally planned a 7-layer drilldown for a manual, reactive workflow. Once we shifted to an ML-driven, proactive model where agents surface budget adjustments via notifications, the UI no longer needed to expose every layer. We used existing calendar filters to create custom views and capped depth at three layers, balancing visibility with simplicity while agents handled deeper complexity.


Integration
Integrating Into Scheduling

Labor budgeting was embedded directly into scheduling so cost impact was visible as shifts were created, not after schedules were published. As managers built or adjusted shifts, real-time budget signals surfaced inline, showing spend and risk before changes were committed. This removed the need to switch tools or reconcile costs later.
Reflection
Some Takeaways…
Designing this system was a highly collaborative effort, shaped through close partnership with product managers, engineers, and ML stakeholders. Working in short, focused sprints allowed the team to pressure-test ideas early, align on constraints, and iterate quickly as complexity surfaced. Many of the strongest design decisions emerged directly from working sessions across disciplines, ensuring the system was both usable and technically sound.
View More







