AI Assist and Optimizer: Generative AI and Mathematical Scheduling in MAS 9

Who this is for: Maximo administrators, functional leads, and IT architects evaluating the AI and optimization capabilities in MAS 9 — whether you are planning a pilot, building a business case, or trying to understand what these applications actually do under the hood.

Estimated read time: 10 minutes

The Two Sides of Intelligent Maintenance

Every maintenance organization faces two fundamental challenges. First, getting the right information to the right person at the right time. Second, getting the right person to the right place at the right time. MAS 9 addresses both with two purpose-built applications: AI Assist for the information problem, and Optimizer for the scheduling problem.

These are not incremental improvements to what Maximo 7.6 offered. The 7.6 Scheduler module provided a graphical Gantt chart where dispatchers manually dragged and dropped assignments. There was no natural language interface, no AI-powered field recommendations, no mathematical optimization. MAS 9 introduces capabilities that simply did not exist in the previous generation.

We will also cover two additional applications in this part — Civil Infrastructure and Parts Identifier — that round out the MAS suite for specialized use cases. But the bulk of this discussion belongs to AI Assist and Optimizer, because they represent the most transformative additions to the suite for general maintenance operations.

AI Assist: Generative AI Comes to Maximo

The watsonx.ai-Powered Conversational Interface

Maximo AI Assist — and the broader Maximo Assistant capability — brings generative AI directly into the maintenance workflow. Powered by IBM watsonx.ai foundation models, it replaces complex search filters and multi-screen navigation with a conversational interface.

Instead of building a query in the List tab with six WHERE clauses, a planner types:

"Show me all open emergency work orders for Building A pumps"

Instead of navigating to Asset, searching by location, clicking the work order tab, and scrolling through history, a technician asks:

"What is the maintenance history for Pump P-1001?"

Instead of calling a senior colleague or searching through binders of documentation, a field tech asks:

"Pump P-1001 is vibrating excessively. What should I check?"

The AI translates natural language into Maximo queries, retrieves results, and presents them in context. This is not a chatbot bolted onto the side of the application. It is an AI layer woven into the Manage interface that understands Maximo's data model, your asset hierarchy, and your historical work patterns.

Natural Language Work Order Creation

This is where AI Assist moves from convenience to genuine productivity transformation. Users can create work orders through plain English descriptions, and the AI drafts the entire record:

User: "The HVAC unit on the 3rd floor of Building B is making a loud noise
       and not cooling properly. It needs to be looked at today."

AI Assistant:
  - Asset: HVAC-B3-001 (identified from location description)
  - Work Type: CM (Corrective Maintenance)
  - Priority: 2 (High - inferred from "today")
  - Description: "HVAC unit making loud noise, not cooling.
                   Requires same-day attention."
  - Failure Class: HVAC
  - Problem Code: NOISE
  - Location: B3-MECH (Building B, 3rd Floor, Mechanical)

  "I've drafted this work order. Would you like me to submit it?"

Think about what the AI is doing here. It identified the asset from a location description ("3rd floor of Building B"). It inferred the work type from the problem description. It set priority based on the urgency language ("today"). It selected the appropriate failure class and problem code from the organization's configured failure hierarchy. It resolved the location code. All from one sentence of plain English.

For organizations where work order quality is a persistent problem — missing failure codes, wrong priorities, incomplete descriptions — this is a direct solution. The AI drafts consistently structured records every time.

AI-Powered Asset Search

Beyond work order creation, AI Assist transforms how users find information:

  • "Find all critical pumps that have had more than 3 breakdowns this year"
  • "Show me assets in Building C that are past their expected life"
  • "Which motors were last maintained more than 6 months ago?"

The AI translates natural language into Maximo query syntax and returns results in standard list views. Users can refine queries conversationally, narrowing results without rebuilding filters from scratch.

Failure Code Identification

When technicians report problems, the AI suggests appropriate failure codes by analyzing the problem description text, recommending the failure class, problem code, cause code, and remedy code based on historical patterns for similar assets and similar descriptions. This directly improves failure code consistency across the organization — a problem that has plagued maintenance data quality for decades.

PM Optimization Recommendations

AI Assist analyzes preventive maintenance history to surface actionable recommendations:

  • Which PMs are generating little value — no defects found in the last N executions
  • Which PMs need shorter intervals — failures occurring between scheduled PMs
  • Which PMs could be extended — consistently finding assets in good condition
  • Optimal PM timing based on asset health scores and prediction data

This is the AI closing the loop between execution data and maintenance strategy. Instead of a reliability engineer manually reviewing PM effectiveness once a year, the AI continuously monitors and recommends.

Knowledge Base Search

AI Assist integrates with document repositories to provide contextual technical knowledge during troubleshooting:

  • Equipment manuals and technical documentation
  • Standard operating procedures (SOPs)
  • Historical repair narratives from work order long descriptions
  • Safety procedures and regulatory requirements

When a technician asks "What is the procedure for replacing a mechanical seal on a centrifugal pump?", the AI searches across all connected knowledge sources and returns the relevant procedure — not just a link to a document, but the specific content the technician needs.

Field Technician Assistance

For field technicians working on complex equipment, AI Assist provides:

  • Step-by-step troubleshooting guidance based on the asset type and reported symptoms
  • Integration with Visual Inspection for visual diagnosis — the technician can capture an image and get AI analysis
  • Access to the complete repair history for the specific asset they are working on
  • Collaboration tools for connecting with subject matter experts when the AI's guidance is not sufficient

This is particularly valuable for organizations facing the skilled trades shortage. Junior technicians can leverage the collective knowledge embedded in years of work order history and documentation without needing a senior tech standing next to them.

MAS 9.1: AI Gets Smarter

Field Value Recommendations with Confidence Scores

MAS 9.1 introduces AI-native field value recommendations directly within the Manage interface. When users are filling out records — work orders, service requests, assets — the AI recommends values for key fields:

Field — How AI Recommends

Priority — Based on asset criticality, description text, historical patterns

Work Type — Based on description text and failure codes

Failure Class — Based on asset type and problem description

Problem Code — Based on description text and historical failures for similar assets

Cause Code — Based on problem code and asset type patterns

Remedy Code — Based on problem/cause combination history

Classification — Based on description and asset attributes

Owner Group — Based on asset location, work type, and classification

Craft — Based on work type and historical assignments

Each recommendation includes a confidence score. A recommendation with 92% confidence is treated differently than one at 61%. Users can accept, modify, or reject recommendations, and this feedback feeds back into the model to improve future accuracy. This feedback loop is critical — the AI gets better the more your organization uses it.

Similar Record Detection

When creating a new work order or service request, the AI identifies similar historical records:

  • "3 similar work orders found for this asset in the past 12 months"
  • Shows resolution details from previous similar work
  • Helps avoid duplicate work orders
  • Enables knowledge transfer from experienced technicians to new staff

This is a practical solution to the duplicate work order problem that every large organization faces. It also surfaces institutional knowledge — when the AI shows that the last three times this pump had this symptom, the root cause was a bearing failure, a new technician immediately has context that would otherwise take years to accumulate.

Multi-Language Support

AI recommendations work across all supported languages in MAS 9.1. A technician entering a work order description in Spanish receives the same quality of field recommendations as one working in English. This matters for global operations where maintenance teams operate in multiple languages but share the same Maximo instance.

The Maximo AI Service Architecture

Understanding how AI Assist works under the hood matters for planning and deployment. The Maximo AI Service is the central integration hub that connects watsonx.ai to MAS applications:

+-------------------+     +-------------------+     +-------------------+
| watsonx.ai        |     | Maximo AI Service |     | MAS Applications  |
|                   |     |                   |     |                   |
| Foundation Models |<--->| Model Management  |<--->| Manage            |
| Custom Models     |     | Feature Store     |     | Mobile            |
| Training Pipeline |     | Inference Engine  |     | Health             |
|                   |     | Feedback Loop     |     | Predict            |
+-------------------+     +-------------------+     +-------------------+

The AI Service consists of several components:

Component — Role

AI Service Operator — Deploys and manages the AI Service on OpenShift

Model Management — Configure, train, and retrain AI models against your data

watsonx.ai Connector — Connects to watsonx.ai foundation models (SaaS or on-premises)

Feature Store — Manages the features used for AI recommendations

Inference Engine — Serves AI model predictions to MAS applications in real time

Feedback Loop — Captures user acceptance/rejection of recommendations for retraining

The feedback loop deserves emphasis. Every time a user accepts or rejects a field recommendation, that signal flows back into the model. Over weeks and months, the AI learns your organization's specific patterns — your failure code hierarchy, your priority conventions, your craft assignment preferences. A generic model becomes your model.

AI Assist Pilot: What It Takes

For organizations evaluating AI Assist, here is the realistic scope:

Phase — Effort — People

watsonx.ai deployment decision (SaaS vs. on-prem) — 8-16 hours — Architecture team

AI Service operator deployment — 4-8 hours — OpenShift admin

watsonx.ai connection configuration — 4-8 hours — AI Service admin

Training data preparation (2+ years of WOs with failure codes) — 16-24 hours — Data team

Initial model training — 8-16 hours — AI/ML team

Field recommendation testing — 8-16 hours — Functional testers

Natural language query testing — 4-8 hours — Functional testers

User feedback and accuracy assessment — 8-16 hours — End users

Total: 68-128 hours, 2-3 people, targeting a 60%+ recommendation acceptance rate. That acceptance rate target matters — it is the threshold where the AI is adding value rather than creating noise. Most organizations exceed this target within the first retraining cycle because the feedback loop rapidly adapts the model to organizational patterns.

Maximo Optimizer: Mathematical Scheduling at Scale

The Problem Optimizer Solves

Scheduling maintenance work is an NP-hard optimization problem. When you have 50 technicians, 200 work orders, varying skills, different tools, geographic spread, priority constraints, SLA deadlines, parts availability windows, and predecessor/successor dependencies — the number of possible schedule combinations is astronomical. No human dispatcher, no matter how experienced, can find the optimal solution manually.

In Maximo 7.6, the Scheduler module gave dispatchers a Gantt chart to drag and drop assignments. It was a visualization tool, not an optimization engine. Optimizer replaces manual scheduling with constraint programming and mixed-integer programming algorithms that find mathematically optimal solutions.

Constraint-Based Scheduling

Optimizer considers every relevant constraint simultaneously:

Constraint — What It Means

Work order priority — Higher priority work orders get scheduled first

Required crafts/skills — Technician skills must match work requirements

Required tools — Tools must be available at the scheduled time

Asset availability — Work scheduled within asset downtime windows

Technician availability — Working hours, shifts, vacation, training days

Travel time — Minimize travel between work locations

Work dependencies — Predecessor/successor WO relationships respected

SLA deadlines — Work completed before service level agreement deadlines

Parts availability — Work only scheduled when required parts are in stock

This is not a priority queue. Optimizer is solving a multi-dimensional constraint satisfaction problem where changing one assignment ripples through the entire schedule.

Crew Scheduling

Optimizer handles crew dynamics that manual scheduling struggles with:

  • Assigning work orders to crews — groups of technicians rather than individuals
  • Balancing workload across crew members
  • Respecting crew composition requirements (lead technician paired with apprentice)
  • Handling overtime constraints and shift boundaries

Route Optimization with ArcGIS

For field service and geographically distributed work, Optimizer integrates with Esri ArcGIS to minimize travel:

  • Real road network routing between work locations — not straight-line distance
  • Traffic-aware travel time estimates that account for time-of-day congestion
  • Multi-stop route optimization that sequences work locations efficiently
  • Service territory definitions that keep technicians in their assigned areas
  • Map-based visualization of optimized routes in the Dispatching Dashboard

Without ArcGIS, Optimizer still calculates travel using straight-line distance estimates. With ArcGIS, it uses actual road networks and real-time traffic data. The difference matters — straight-line distance between two points in a metropolitan area can underestimate actual drive time by 2-3x due to road layouts, bridges, and traffic patterns.

Multiple Objective Functions

Optimizer supports multiple optimization objectives that can be weighted to match your priorities:

Objective — What It Optimizes

Minimize total cost — Labor cost + travel cost + penalty for late work

Maximize work completed — Complete as many work orders as possible within the window

Minimize travel — Reduce total travel time and distance

Balance workload — Even distribution of work across technicians

Minimize tardiness — Reduce late completions relative to target dates

You can weight these objectives. A utility company after a storm might set "maximize work completed" to the highest weight. A facility management company with strict SLAs might weight "minimize tardiness" highest. A field service organization might weight "minimize travel" because fuel and drive time are their largest variable costs.

Configurable Optimization Parameters

Optimizer exposes configuration parameters that let you tune the engine for your environment:

Parameter — Description — Typical Values

Planning horizon — Time window to optimize — 1 day to 2 weeks

Optimization time limit — Maximum solver runtime — 30 seconds to 30 minutes

Priority weights — Relative importance by priority level — P1=100, P2=50, P3=20, P4=10, P5=5

Travel speed — Average speed for distance calculations — 30 mph (urban), 60 mph (highway)

Work buffer — Buffer time between assignments — 15-30 minutes

Overtime allowance — Whether optimizer can schedule overtime — Yes/No, max hours per tech

Skill matching — Strict match vs. closest available match — Strict or Flexible

The optimization time limit is worth understanding. A 30-second solver run produces a good schedule quickly — useful for real-time re-optimization when an emergency work order arrives mid-day. A 30-minute run explores more of the solution space and produces a better schedule — appropriate for overnight batch optimization of the next day's work.

Integration with Graphical Assignment and Dispatching

Optimizer results feed directly into the Manage Graphical Assignment (Gantt chart) and the Dispatching Dashboard:

  • Optimized schedule appears on the Gantt chart for dispatcher review
  • Dispatchers can manually adjust assignments after optimization
  • Manual changes are validated against constraints — the system warns if you break a rule
  • Re-optimization can be triggered after manual changes to re-optimize around the dispatcher's overrides
  • The Dispatching Dashboard shows a real-time map view of technician locations and assigned work

This human-in-the-loop design is important. Optimizer does not replace dispatchers. It gives dispatchers a mathematically optimized starting point that they can then adjust based on knowledge the system does not have — a technician who is having a bad day, a customer who specifically requested a different tech, a road closure that is not yet in the GIS system.

Real-World Use Cases

Field Service Optimization:

  • 50 field technicians covering a metropolitan area
  • 200+ work orders per day across 500+ locations
  • Optimizer creates daily routes minimizing travel while respecting priorities and SLAs
  • Result: 20-30% reduction in travel time, 15-25% more work orders completed per day

Shutdown Planning:

  • Annual plant shutdown with 500+ work orders in a 14-day window
  • Multiple crafts, tools, and contractors to coordinate
  • Complex predecessor/successor dependencies between work orders
  • Optimizer creates a feasible schedule, identifies resource conflicts, suggests solutions
  • Result: 10-15% reduction in shutdown duration

Emergency Response:

  • Storm damage creates 1,000+ emergency work orders overnight
  • Optimizer rapidly schedules crews to highest-priority locations
  • Re-optimizes continuously as new work orders arrive throughout the day
  • Result: Faster restoration, equitable workload distribution

Optimizer Pilot: What It Takes

Phase — Effort

Review current scheduling process, identify pain points — 8-16 hours

Verify craft/skill data quality in Manage — 8-16 hours

Verify work location coordinate data (lat/long) — 4-8 hours

Deploy Optimizer operator on OpenShift — 4-8 hours

Configure optimization model with current constraints — 8-16 hours

Run first optimization on sample work order set — 4-8 hours

Compare optimized vs. manual schedule — 4-8 hours

Test with larger volumes and iterate parameters — 8-16 hours

Evaluate ArcGIS integration (if applicable) — 8-16 hours

Total: 60-120 hours, 2-3 people. The critical prerequisite is data quality — technician skills must be accurately recorded, and work locations need coordinates for travel optimization to work. Most organizations underestimate the data cleanup effort.

Civil Infrastructure: Regulatory-Compliant Inspection

Maximo Civil Infrastructure is a purpose-built application for organizations that manage bridges, roads, tunnels, retaining walls, and other transportation assets. If your organization is a state DOT, county public works department, toll authority, or transit agency, this section is directly relevant. For everyone else, skim it and move on.

Bridge Inspection (NBI Compliance)

Civil Infrastructure provides FHWA-compliant bridge inspection with:

  • NBI element-based inspection per the AASHTO Manual for Bridge Element Inspection methodology
  • Condition ratings on the standard 0-9 NBI scale for deck, superstructure, and substructure
  • Element condition state tracking — quantity in each condition state per element
  • Automated inspection scheduling per FHWA biennial (2-year) cycle requirements
  • NBI data export in the format required for federal reporting
  • Load rating tracking and scour assessment for scour-critical bridges

Road and Tunnel Inspection

For roads, Civil Infrastructure supports:

  • Pavement Condition Index (PCI) scoring
  • International Roughness Index (IRI) ride quality metrics
  • Distress identification and treatment recommendation

For tunnels:

  • NTIS (National Tunnel Inspection Standards) compliance
  • Element-based tunnel inspection
  • Fire and life safety system tracking

Regulatory Compliance Framework

Standard — Coverage

FHWA NBIS — National Bridge Inspection Standards — program management

AASHTO Manual — Element-based inspection methodology

23 CFR 650 — Federal regulations for bridge inspection frequency and reporting

NTIS — National Tunnel Inspection Standards

MAP-21 / FAST Act — Federal transportation legislation — performance-based planning

GASB 34 — Government accounting for infrastructure asset valuation

Visual Inspection Integration

Civil Infrastructure integrates with Maximo Visual Inspection for AI-augmented inspections. Drone-captured imagery of bridge decks, piers, and beams can be analyzed by MVI models trained on civil infrastructure defects. MAS 9.1 introduces Large Vision Models (LVMs) specifically trained for civil infrastructure — detecting cracks, spalling, corrosion, and delamination from aerial imagery and linking visual evidence directly to element-level inspection records.

Parts Identifier: Computer Vision for the Field

Maximo Parts Identifier solves a specific but common problem: a field technician encounters a part they cannot identify. The label is worn. The part number is not visible. The equipment is unfamiliar. In Maximo 7.6, the technician would call the shop, describe the part over the phone, and hope someone could identify it. Parts Identifier replaces that workflow with AI.

How It Works

The technician takes a photo of the unknown part with their mobile device. The AI model analyzes the image and matches it against a trained parts catalog. Results come back with the part number, description, manufacturer, current stock levels in nearby storerooms, and a link to initiate a purchase requisition if the part is not in stock.

The system also suggests alternative or substitute parts when the exact match is not available, and it learns from confirmed identifications over time — the more your team uses it, the more accurate it becomes.

Use Cases

  • Field technicians identifying parts on unfamiliar equipment during emergency repairs
  • Receiving clerks identifying parts that arrive without proper documentation
  • Inventory staff cataloging unlabeled stock during warehouse cleanup
  • Apprentice technicians learning parts identification as part of their training
  • Emergency repairs when subject matter experts are unavailable and a part needs to be identified immediately

Key Takeaways

  • AI Assist is the most significant change in how users interact with Maximo since the web interface — natural language queries, work order creation from plain English, and AI-powered field recommendations change the daily experience for planners, technicians, and supervisors
  • The feedback loop is the key to AI value — every accepted or rejected recommendation makes the model more accurate for your organization. Plan for the retraining cycle, not just the initial deployment
  • Optimizer replaces manual scheduling with mathematical optimization — constraint programming that considers skills, tools, travel, priorities, parts, SLAs, and dependencies simultaneously
  • ArcGIS integration transforms travel optimization from straight-line estimates to real road network routing with traffic awareness
  • Civil Infrastructure addresses a compliance gap that general-purpose EAM cannot fill — FHWA, AASHTO, NTIS, and GASB 34 requirements built into the inspection workflow
  • Parts Identifier solves the identification problem with computer vision — photo to part number to stock level to purchase requisition in seconds

References

Series Navigation:

Previous: Part 13 -- Visual Inspection and Maximo Mobile Deep Dive
Next: Part 15 -- AppPoints Licensing and Application Roadmap

View the full MAS FEATURES series index

Part 14 of the "MAS FEATURES" series | Published by TheMaximoGuys

AI Assist and Optimizer represent the two halves of intelligent maintenance — getting the right information to the right person, and getting the right person to the right place. Together with Civil Infrastructure and Parts Identifier, they complete the MAS 9 application suite picture. In Part 15, we will cover AppPoints licensing and how to build a practical implementation roadmap across all suite applications.