Muhammad Saghir — Senior Product Designer

Muh.

Avatar Nexus

Project Overview

Modern supply chain planning relies on multiple scenarios to evaluate different strategies, but comparing outcomes and identifying issues across them quickly becomes complex.

Designing agentic AI for supply chain planning — enabling Elixum planners to understand, compare, and resolve complex scenarios through conversational intelligence and autonomous agent groups.

Product: Avatar AI Agents & Agent Groups
Domain: AI-powered Supply Chain Planning
Platform: Enterprise Web Application
Role: UX Designer & UX Strategist
Project Type: Conceptual / Product Enhancement

01 — Background & Problem

Elixum gave planners power.

Avatar Nexus gives them intelligence.

🏗️

Built on Elixum

Elixum introduced modular supply chain planning — strategic network planning, demand management, S&OE, and production scheduling — on a composable, infrastructure-agnostic platform. Avatar Nexus is the AI intelligence layer on top of this foundation, extending the same personas into an agentic planning experience.

Modern supply chain planning relies heavily on scenario-based planning. Elixum planners work with multiple scenarios simultaneously — Scenario A, B, C — evaluating strategies for demand, supply, capacity, and risk. Powerful in theory. Overwhelming in practice.

Avatar AI Agents were introduced as an intelligent layer on top — helping those same Elixum users analyze, reason, and act across scenarios through conversational AI and autonomous agents.

The challenge wasn't adding AI. It was designing for agentic UX — where the system doesn't just respond but proactively monitors, reasons, and executes on behalf of the user without sacrificing their sense of control.

Every interaction had to balance autonomy with trust, especially in high-stakes enterprise supply chain decisions where a wrong recommendation costs millions.

"How might we enable Elixum planners to understand, compare, and resolve scenario-based planning issues using AI — while maintaining control, trust, and transparency?"

User Pain Points (surfaced from Elixum research)

🔄

Scenario Overload

Managing 5+ planning scenarios simultaneously caused cognitive fatigue — users toggled between tabs, losing context across every switch.

🔍

Root Cause Blindness

Identifying why issues appeared across scenarios required manual cross-referencing of KPI tables — error-prone and hours-consuming.

⚙️

Repetitive Manual Analysis

Planners ran identical analytical sequences daily — ideal for automation — yet no intelligent layer existed to offload this cognitive labour.

🤖

AI Trust Deficit

Early AI recommendations in Elixum's beta had no explanation layer. Users ignored them entirely, defaulting to manual decisions.

02 — Design Process

Design Thinking Framework

An iterative, non-linear process used to understand users, challenge assumptions, redefine problems, and create innovative solutions.

Core Agentic UX Principles Applied

Human-in-the-Loop

Every autonomous agent action that affects planning data requires explicit user approval. The system acts, but humans decide.

Explainable Decisions

Every AI insight surfaces its data source, assumptions, and confidence level — turning the black box into a glass box.

Progressive Trust Building

Agents start with recommendations, earn trust through accuracy, then unlock automation capabilities incrementally.

Understand

Explore

Materialize

Empathize

Research users' needs

Define

State users' problems

Ideate

Challenge assumptions

Prototype

Create solutions

Test

Try solutions out

Implement

Introduce to markets

03 — User Personas

From Elixum Planners → Avatar Nexus Power Users

These four personas were first established during Elixum's foundational UX research — 1-on-1 interviews with supply chain account managers. Avatar Nexus was designed to serve these same users, now with agentic AI capabilities layered on top of their existing Elixum workflows.

🔗

Persona continuity: These are not new users. Avatar Nexus extends the Elixum user base — the same people who struggled with scenario overload in Elixum now gain agentic AI capabilities to solve those exact problems.

Forecasting

ML Models

Daily User

Elixum Context

Uses Elixum's demand management module daily. Runs statistical + ML forecasting workflows. Manually tracks 3–5 scenarios per product family.

↑ Avatar Nexus Upgrade

Avatar Nexus gives them a Demand Agent that automatically detects demand spikes, surfaces root causes, and answers natural-language queries like "Why did Scenario A spike last week?"

Before

Spends 40% of day on repetitive analysis

After

Delegates analysis to Demand Agent

Risk Monitoring

KPI Tracking

Multi-Scenario

Elixum Context

Monitors inventory buffers and supply constraints across Elixum's network planning module. Manages KPI dashboards across multiple scenarios.

↑ Avatar Nexus Upgrade

Avatar Nexus provides a Supply Agent that monitors inventory risk across all scenarios in parallel and sends proactive alerts before SLAs are breached.

Before

Manual cross-referencing between scenario views

After

Supply Agent proactively flags anomalies

Supplier Coordination

S&OE

Action-Oriented

Elixum Context

Coordinates supplier lead times and capacity commitments using Elixum's S&OE module. Needs to align procurement actions with planning scenario outcomes.

↑ Avatar Nexus Upgrade

Avatar Nexus lets them query cross-scenario supplier impacts in natural language and receive consolidated action recommendations from multi-agent analysis.

Before

No visibility into how scenarios affect suppliers

After

Cross-scenario supplier impact queries

Executive Review

Risk Approval

Strategic

Elixum Context

Reviews all scenario outcomes at a strategic level. Approves corrective actions. Relies on Elixum dashboards for executive planning reviews.

↑ Avatar Nexus Upgrade

Avatar Nexus gives them the Orchestrator agent — a meta-agent that correlates signals across Demand + Supply agents and delivers boardroom-ready consolidated insights and action plans.

Before

Manually synthesises insights from multiple sources

After

Orchestrator delivers consolidated executive insights

Goals & Target Success Metrics

Business Goals

Increase adoption of AI-driven planning features

Reduce manual effort in scenario analysis

Enable scalable automation through agent groups

User Goals

Quickly understand issues across planning scenarios

Delegate repetitive analysis to AI agents

Gain confidence in AI-driven recommendations

30%

Less scenario analysis time

Baseline from Elixum user research

40%

Reduction in Manual Checks

Through intelligent agent automation

↑

Increased AI agent usage

Session frequency & depth tracking

★★★★☆

Trust & satisfaction

Post-launch survey target

04 — The Solution

Avatar Nexus — Two layers of intelligence

Avatar Nexus delivers AI-driven supply chain intelligence through two complementary capabilities — each designed to meet users where they are in their planning workflow.

💬

Basic Capability

Conversational Scenario Analysis

Users create chat sessions linked to specific planning scenarios. Through natural language, they can ask questions, surface anomalies, and explore root causes without leaving their planning context.

"Why is Scenario A showing higher inventory risk?"

"What changed in Scenario B that increased utilization?"

"Compare demand patterns across all active scenarios"

🤖

Advanced Capability

Agent Groups — Multi-Agent Orchestration

Agent Groups combine multiple specialized agents into a coordinated system. Rather than analyzing issues in isolation, they share context and collaborate to identify complex cross-scenario patterns.

→

Demand Agent detects unexpected demand spikes

→

Supply Agent flags supplier delay patterns

→

Orchestrator correlates signals cross-scenario

→

System delivers consolidated root-cause + action plan

User Flow

The user flow maps how Elixum planners move from their existing scenario workspace into Avatar Nexus — through conversational analysis, to agent delegation, to approved action execution. Explicit approval gates appear at every step that modifies planning data.

05 — High-Fidelity Designs

Basic Capabilities — Conversational Scenario Analysis

05-1 — Out of The box

Advanced Ideas

Five screens map the complete journey from landing in the Avatar Nexus home view through initiating a chat, receiving AI analysis, selecting scenarios for comparison, and continuing the conversation with refined queries.

Screen 1 of 5 — Conversational Scenario Analysis Home

1. Home — Avatar Nexus landing view. Scenario cards surface quick KPIs and alert counts. Users can initiate a chat session directly from a scenario card.

Screen 2 of 5 — Chatting with the Avatar Agent

2. Chat initiated — Conversational interface opens in context of the selected scenario. Suggested prompts help novice users get started quickly.

Screen 3 of 5 — AI Analysis Results

3. AI Analysis results — Agent surfaces root-cause insights with supporting data, confidence indicators, and actionable next steps. Every claim is traceable.

Screen 1 of 4 — Agent Groups Home

1. Agent Groups Home — Overview of active agent groups, their status, last run, and aggregated findings. Orchestrator agent is visible in-chat context.

Screen 2 of 4 — Agent Creation Flow

2. Agent Creation — Guided wizard for creating a new agent with defined scope, data access, trigger conditions, and escalation rules.

Screen 3 of 4 — Multi-Agent Orchestration

3. Multi-Agent Handling — Live view of multiple agents running in parallel, with progress indicators, interim findings, and correlation arrows between agents.

Screen 4 of 4 — Overview & Task Summary

4. Overview + Tasks Summary — Consolidated results from the agent group run, with actionable tasks, approval requests, and scenario recommendation cards.

Additional — Calendar & Filter Panel

Screen 1 — Summaries AI Orchestrator

Screen 2 — Conflict Resolution

Screen 3 — Execution

Screen 4 — Trust Layer

Screen 5 — Draft PR

5. Calendar Filter — Time-period selection for agent analysis scope. Supports rolling windows, fiscal periods, and custom date ranges.

05b — Advanced Capabilities

Agent Groups — Multi-Agent Orchestration

Four screens cover the Agent Groups experience: from the overview dashboard through agent creation, multi-agent coordination, and full-screen filtering — giving power users granular control over automated planning workflows.

06 — Responsible AI Design

Trust, Transparency & Control

In enterprise AI, trust is not a feature — it's a prerequisite. Every design decision in Avatar Nexus was filtered through three questions: Does the user understand what the agent did? Can they verify it? Can they override it?

🔍

Explainability by Default

Every insight includes its data sources, the assumptions made, and the confidence level. Users never face a bare conclusion.

📋

Assumption Disclosure

Agents surface their reasoning assumptions before acting — especially for automation scenarios. Users can challenge or adjust.

✅

Approval Gates

Any action that modifies planning data — adjusting parameters, triggering re-plans — requires explicit user approval. No silent mutations.

📖

Auditable Agent Logic

Agent rules can be reviewed, edited, and rolled back. The system is designed to be interrogatable — not just usable.

Accessibility & Ethics

✓

Clear, non-technical language

All AI explanations are written at plain language level — no jargon, no ambiguity.

✓

Avoidance of black-box decisions

No recommendation is surfaced without an accessible explanation of how it was derived.

✓

Human-in-the-loop automation

Automation is opt-in, scoped, and always reversible. Users set boundaries; agents respect them.

✓

Responsible AI principles

Fairness, accountability, and transparency baked into every agent design and capability scope.

07 — Design Impact

Great design removes friction. These numbers prove it.

Design impact was measured across three dimensions: speed, accuracy, and engagement — each directly tied to supply chain planning outcomes.

30%

Reduction in Analysis Time

Streamlined processes that reduce time spent on initial data preparation and scenario configuration.

"Avatar Nexus didn't just add AI to supply chain planning — it redefined how planners relate to their data. From passive consumers to active directors of intelligent systems."

40%

Reduction in Manual Checks

AI-augmented planning reduced manual checks by 40%, enabling faster and more accurate supply chain decisions.

↑↑

Template Adoption Rate

Higher engagement with AI-assisted templates, resulting in more consistent and efficient workflows.

start

Chat with

Avatar

AI analysis

scenarios

Chat Agent

e.g List Available Capacities

Orchestrator

e.g Remove Station Utilization Error

Scenario Creation

Agent Delegation /

goals

Changes in

Scenario

User

Action

Scenario Comparison

e.g Keep or discard

Baseline

Base Scenario

Child scenario 1

···

Child Scenario 2

Child Scenario 3

Child Scenario 4

Child Scenario 5

Child Scenario 6

Avatar AI

Agented AI

Dashboard

Capacity planning

Scheduling board

Buffer status

Heuristic design

Network view

Supply planning

Location prod.

Demand management

DD-MRP

Logistics

Order network

Scenario comp.

SC Network

Widget types

Add view

Tax & Tarif Cost

Tax Cost

Tarif Cost

11-11

11-23

12-5

50%

100%

150%

Station Utilisation

190%

−10%

Station 12

Station Utilisation

190%

−10%

Station 12

Station Utilisation

190%

−10%

Station 12

Open Sales Orders

€698K

stable

Closed Sales

Closed Sales Orders

€698K

stable

Closed Sales

Open Sales Orders

€698K

−2%

Closed Sales

Open Sales Orders

€698K

stable

Closed Sales

Late Orders

400

1+ Month Late

1+ Week

400

On-Time Delivery

400

On Time

Deallocated

Station Utilisation

190%

−10%

Station 12

Station Utilisation

190%

−10%

Station 13

Supply Reliability

€698K

41% at risk

Open Sales

↗ Profile

MSN Lateness

1+ Week Too Early

On Time

1+ Week Late

1+ Month Late

200

400

500

Constrained Signal Integrity

Okay

Warning

Critical

Product A

480

295

195

Product B

390

220

145

Product C

420

260

170

Total

100

200

300

400

500

Constrained Buffer Status

Blue

Green

Yellow

Red

Dark Red

Product A

Product B

Product C

History Base

List Entry 1

11/24/2021 12:00:00

List Entry 2

11/24/2021 12:00:00

List Entry 3

11/24/2021 12:00:00

List Entry 4

11/24/2021 12:00:00

Default

11/24/2021 12:00:00

Default

11/24/2021 12:00:00

Default

11/24/2021 12:00:00

Default

11/24/2021 12:00:00

Product View

REQ./REC. DATE

NUMBER