How Excalibur's Decision Trees Were Designed

An AI wins-and-corrections assistant built on a persona-mapping and decision-tree design process -- stakeholder research, workflow automation, and rule-based conversation design -- that replaced a manual, siloed reporting process with a single assistant that answers the same question differently depending on who's asking.

600+ hrs

annual manual burden identified

300 hrs

annual hours saved by automation

$22,500

annual direct savings

stakeholder personas mapped into decision trees

Design Process, at a Glance

Stakeholders submit a win or correction of error through a short Microsoft Form. A Power Automate flow instantly moves that response into a governed, centralized SharePoint list, replacing manual compiling. A Copilot Studio agent, Excalibur, sits on top of that list. Before answering, Excalibur detects who is asking -- Transportation Manager, Replenishment Manager, Event Manager, or Executive -- and routes to a decision branch that changes both which information it prioritizes and the language it uses, while attaching a confidence level to every answer based on how well it is cross-validated across sources.

A simplified view of the intake-automation pipeline and persona-adaptive decision tree behind Excalibur. Exact survey links, team names, and record-level detail are intentionally generalized for this public one-pager.

How It Was Designed, Behind the Scenes

The starting point was not a chatbot -- it was a stakeholder map. Before a single agent instruction was written, the real users of this knowledge base were interviewed and documented as full personas: Transportation Managers, Replenishment Managers, Event Managers, and Executives, each with distinct goals, pain points, and vocabulary. That research became a decision tree: detect who is asking (or the shape of what they're asking), route to that role's information priorities, and answer in that role's language, rather than returning one generic summary to everyone. A second, parallel decision tree governed submission: instead of one long freeform form, a short branching sequence -- which of three event types does this fall under? -- routed each submitter straight to the matching intake form. Every answer the agent gives also carries a confidence label, because a number that's been validated across two sources is not the same as one that's been estimated from a single survey response.

Technology Behind the Tool

Microsoft Copilot Studio (agent instruction and conversation design)
Power Automate (event-triggered workflow orchestration)
SharePoint list (governed, structured data layer)
Microsoft Forms (branching, guided intake)
Persona-adaptive decision-tree logic (rule-based routing by stakeholder type)
Confidence-scored, multi-source validation logic
Fiscal-calendar-aware temporal query logic

Capabilities Demonstrated

Stakeholder persona research
Decision-tree / conversation design
Cross-functional facilitation
Workflow automation
AI product design
Business case development
Enablement & rollout planning