Revolutionizing Field Service: How LLMs Are Powering Smarter Knowledge Access for Technicians

Revolutionizing Field Service with LLM-Powered Knowledge Management

Field service technicians need instant access to accurate repair information, but outdated knowledge systems often create frustrating delays. Large Language Models (LLMs) are changing the game—enhancing knowledge retrieval, streamlining troubleshooting, and reducing technician dependency on senior staff.

In this webinar, Seth Earley and industry experts Sanjay Mehta, and Heather Eisenbraun explore how LLMs and Retrieval-Augmented Generation (RAG) are transforming field service operations. Discover how AI-powered knowledge management is improving efficiency, reducing downtime, and elevating service quality.

LLMs for Instant Knowledge Retrieval – How AI-driven search dramatically cuts troubleshooting time.

Structured Data & AI – Why high-quality, organized knowledge is essential for LLM success.

Real-World Implementation – Lessons from deploying LLM-powered knowledge tools in field service.

Business Impact – How AI reduces service delays, optimizes workflows, and enhances technician productivity.

Empower your field service teams with AI-driven knowledge access. Watch the webinar to see how LLMs are revolutionizing service efficiency.

1. www.earley.com
WEBINAR WEBINAR
Accelerating Knowledge Architecture Using LLMs
Media Sponsor
SANJAY MEHTA
PRINCIPAL SOLUTION ARCHITECT
EARLEY INFORMATION SCIENCE
SETH EARLEY
CEO & FOUNDER
EARLEY INFORMATION SCIENCE
HEATHER EISENBRAUN
KNOWLEDGE ARCHITECT
EARLEY INFORMATION SCIENCE

2. www.earley.com
Today’s Panel
[email protected]
https://www.linkedin.com/in/sethearley/
2
Seth Earley
Founder & CEO
Earley Information Science
Sanjay Mehta
Principal Solution Architect
Earley Information Science
[email protected]
https://www.linkedin.com/in/sanjaymehta/
“I do not know of any books that have such
useful and detailed advice on the relationship
between data and successful conversational AI
systems.”
—Tom Davenport, President’s Distinguished
Professor at Babson College, Research Fellow at
MIT Initiative on the Digital Economy, and author of
Only Humans Need Apply and The AI Advantage
Heather Eisenbraun
Knowledge Architect
Earley Information Science
[email protected]
https://www.linkedin.com/in/heisenbraun/

3. www.earley.com
Before We Get Started
WE ARE RECORDING
SESSION WILL BE
50 MINUTES PLUS
10 MINUTES FOR
Q&A
YOUR INPUT IS
VALUED
Link to recording & slides
will be sent by email after
the webinar
Use the Q&A box to
submit questions
Participate in the polls
during the webinar
Feedback survey afterward
(~1.5 minutes)
Thank you to our media partners : CMSWire and VKTR

33

4. www.earley.com
About Earley Information Science
4
Proven methodologies to organize information and data.
SELL MORE
PRODUCT
SERVICE
CUSTOMERS
EFFICIENTLY
INNOVATE
FASTER
1994
YEAR FOUNDED.
Boston
HEADQUARTERED.
50+
SPECIALISTS & GROWING.

5. www.earley.com
Related Articles
5
Upcoming Session at KM & AI World

6. Copyright © 2025 Earley Information Science, Inc. All Rights Reserved.www.earley.com
What You'll Learn
6
Context: The Key to
AI Accuracy
AI needs the right context
to deliver accurate, useful
responses—without it,
results fall short.
Explore how leveraging key
data sources makes AI
smarter, faster, and more
helpful by delivering more
context.
See how AI-generated &
expert-refined information
architecture ensures
accuracy, consistency, and
trust in AI applications
Not all AI is created equal.
Selecting the right AI models
and skills makes all the
difference.
Powering AI with
Business Data
How IAD-RAG
Powers Reliable AI
Choosing AI That
Delivers Results

7. Copyright © 2025 Earley Information Science, Inc. All Rights Reserved.www.earley.com
General Response vs. Contextualized Response
General Response:
•Provides broad, often generic answers.
•Lacks awareness of specific business context or
structured data (requires greater
disambiguation).
•Relies on probabilistic outputs without
validation.
Contextualized Response:
•Draws from structured, curated enterprise
knowledge.
•Aligns with organizational taxonomy,
governance, and compliance.
•Ensures accuracy, relevance, and trust in AI-
generated insights.
AI Without Information Architecture Is Unpredictable
•AI systems need structured taxonomies, metadata, and ontologies to deliver business-ready insights.
•Organizations that fail to invest in information architecture risk hallucinations, misinformation, and inefficiency.

8. Copyright © 2025 Earley Information Science, Inc. All Rights Reserved.www.earley.com
Understanding AI Accuracy in Business Applications
User Behavior Signals
Quick exits, query rephrasing, and follow-up
questions indicate potential AI response
inaccuracies that need addressing.
Success Rate Tracking
Monitoring and analyzing success metrics
help continuously refine and improve AI
response quality.
Company Context Integration
AI systems must be deeply integrated with
company-specific knowledge to deliver truly
valuable responses.
Environmental Intelligence
Leveraging additional data sources and session data
creates more contextually aware and accurate AI
responses.
Temporal Relevance
Conversation history must be managed dynamically,
recognizing when past context becomes outdated or
irrelevant.

9. Copyright © 2024Earley Information Science, Inc. All Rights Reserved.
Expanding AI Accuracy with Additional Data Sources
Data Foundation
Traditional AI models rely on general content, but business-specific data is essential for relevance.
Combining structured and unstructured data ensures better decision-making.
Key Data Sources
Customer Data (profiles, past interactions) → Personalization & intent recognition. Transactional Data
(purchases, account history) → More informed recommendations. Ticketing & Service Data (support
cases, resolutions) → Context-aware responses.
Information Architecture Impact
Poor IA = Inconsistent, incomplete, or inaccurate AI-generated answers.
Well-structured IA = Better AI recall, precision, and usability.
RAG Enhancement
RAG enhances AI by fetching real-time, relevant data instead of relying on outdated training sets. IA
ensures RAG retrieves the right information from the right sources at the right time.

10. Copyright © 2024Earley Information Science, Inc. All Rights Reserved.
Poll
10
1.What’s RAG?
2.Some experiments and PoCs
3.Deployed for internal processes
4.Deployed for external customer facing processes
5.None of the above
What is your experience with RAG?

11. Copyright © 2024Earley Information Science, Inc. All Rights Reserved.
©2025 Earley Information Science,Inc.All rights reserved.
Retrieval Augmented
Generation
Enables large language models to leverage a company's internal
knowledge sources, improving the accuracy and reliability
of responses.
Can be configured to constrain outputs, significantly reducing the
likelihood of hallucinations or fabricated information.
Information Architecture Directed Retrieval Augmented
Generation (IAD-RAG) allows AI systems to draw upon an
organization's proprietary data and content, from ecommerce to
customer support, to provide more trustworthy and contextual
responses. By integrating IAD-RAG, companies can mitigate the
risks of language model hallucinations.
Transactions
Knowledge
Base
Interaction
History
Audiences
…
External Data
Field Data
Governance

12. www.earley.com
Information Retrieval Continuum
BASIC
SEARCH ENGINE
KNOWLEDGE
PORTAL
VIRTUAL
AGENT
RETRIEVAL AUGMENTED
GENERATION
KNOWLEDGE
BASE
Any text
Multiple sources
Keyword or full text
query
None necessary, but
Improves with metadata
Search box,
documents list
Search
Multiple sources, separate
taxonomies and schemas
Full text query or
Faceted exploration
Taxonomies, clustering,
classification
Role-Based
Search, classification,
databases
Domain specific ontologies
Highly curated sources
Query, explore facets
Offers related info
Conversational
NLP, search, classification
Process engines
Knowledge ingested into
vector space
Similarity/semantic search (vector
search)
Augmented with metadata
Conversational, retains context,
personalized
LLM, vector store, NLP, semantic
search, classification
Ontologies, clustering,
classification, NLP
Ontology, knowledge graph,
metadata models, NLP
SEARCH
INTERACTION
INFORMATION
ARCHITECTURE
USER EXPERIENCE
ENABLING
TECHNOLOGY
Increasing functionality

13. Copyright © 2024Earley Information Science, Inc. All Rights Reserved.
Poll
13
1.Manually testing and evaluating relevance
2.Automatically testing and evaluating relevance
3.Fine tuning for our use cases
4.Fine tuning based on our information architecture
5.Experimenting with different large language models
6.Componentizing content and enriching embeddings
7.N/A or none of the above
What is the maturity of your efforts?
(choose all that apply)

14. Copyright © 2024Earley Information Science, Inc. All Rights Reserved.
POCKET TECHNICIAN APP
FOR FIELD SERVICE
TECHNICIANS​
CASE STUDY

15. Copyright © 2025 Earley Information Science, Inc. All Rights Reserved.www.earley.com 15
Day in the Life of a Service Technician
Technicians receive manual work
orders that lack complete job
details and parts lists.
8:00 AM – Delayed
Assignments
Manually searching for
documentation is time-consuming.
9:00 AM – Inefficient
Diagnosis
Senior & Master Technicians are
interrupted for routine guidance.
10:30 AM –
Information Seeking
A part isn't available onsite.
Identifying and ordering parts is
slow and error-prone.
1:30 PM – Parts Issues
Generic job logs lack crucial
repair details.
Service managers must manually
update records.
3:00 PM – Close Work
Order
Limited billable hours completed
due to inefficiencies.
5:00 PM – End of Shift
with a Backlog

16. Copyright © 2024Earley Information Science, Inc. All Rights Reserved.
Revolutionizing Fleet Maintenance with AI
16
Current Maintenance Challenges
•Technicians struggle with
information access
•Inefficient planning and parts
management
•High costs and operational
inefficiencies
AI-Powered Solutions
•Smart knowledge retrieval
•Intelligent work planning
•Predictive parts management
•Automated logging systems
Transformative Results
•50% reduction in search time
•30% decrease in downtime
•$2.3M - $4.4M annual savings
•$1.3M - $2M revenue increase
•Improved technician satisfaction

17. Copyright © 2024Earley Information Science, Inc. All Rights Reserved.
EIS Approach to an AI-Powered RAG Solution
Structuring &
Enriching Content
21
Understanding & Assessing
the Knowledge Landscape
3
AI Integration &
Contextualization
•Content Audit
•Assess Content Challenges
•Audience Analysis & Use Cases
•Develop IA & Metadata Schema
•Identify & Fill Critical Knowledge Gaps
•Chunk Content & Store in CMS
•AI Model Selection & Skill Application
•Capture Semantics & Automate Tagging
•Develop & Optimize Prompts for RAG
VIA – the EIS Virtual Information Architect –
accelerates the process of developing knowledge architectures so
teams can organize and label information faster.
Retrieve accurate information quickly
using the context provided by the IA
Provide pre-defined prompts to users
that leverage the IA to return the most
relevant answer faster
Get the right information to the right
person at the right time &
in the right context
PHASE
ACTIVITIES
GOAL

18. Copyright © 2024Earley Information Science, Inc. All Rights Reserved.18
•Content Inventory
•Content Purpose
•User Needs & Goals
•Context of Use
Analysis
Organizing Principles
•Classification Systems
•Hierarchies
•Taxonomy
•Entities
•Attributes
•Relationships Among Entities
Ontologies
•Content Types
•Fields & Attributes
•Relationships Among Content Types
•Rules
Content Models
Human experts test and refine VIA results to ensure they align with
enterprise system requirements, data strategies, and user needs.
EIS Approach to an AI-Powered RAG Solution (cont.)

19. Copyright © 2024Earley Information Science, Inc. All Rights Reserved.
VIA Supports an AI-Generated, Knowledge Expert-Refined Process
AI Generates or Extracts Content Human Experts Refine and Validate
19
Human experts validates relationships and adds context
or constraints to build a business-specific ontology
AI + Human Knowledge Architect = Robust & Reliable Knowledge Architecture
WHY
HOW
EXAMPLES
AI can process large volumes of data, identify patterns,
extract key entities and relationships & generate initial
versions of ontologies, taxonomies & content models
Human experts curate the AI's output to ensure accuracy,
relevance & alignment with business needs and goals
AI's efficiency and scalability Human's judgment, expertise & contextual understanding
AI extracts potential topics & categories
from a collection of documents
Human experts review AI suggestions to create a
formal taxonomy that's fit-to-purpose
AI identifies relationships between
different concepts

20. Copyright © 2024Earley Information Science, Inc. All Rights Reserved.
UNDERSTANDING THE
KNOWLEDGE LANDSCAPE

21. Copyright © 2025 Earley Information Science, Inc. All Rights Reserved.www.earley.com
Poll
21
1.Lack of knowledge repository
2.Missing understanding of connection of Knowledge
Engineering to AI Success
3.Poorly structured source information
4.Lack of maturity around knowledge processes and
engineering
5.Tendency of tech org to look at this strictly from a
technology perspective
6.Lack of buy in or commitment from the business due to
burden of knowledge curation
7.Something else
What is your biggest concern about Knowledge and AI?
(check all that apply)

22. Copyright © 2024Earley Information Science, Inc. All Rights Reserved.
Understanding the Knowledge Landscape
22
LLM General
Knowledge
Third-Party
Service Manuals
-How a system works
-Common repairs
-Environmental regulations
-Common tools
-Technical specifications
-Alarms
-Troubleshooting steps
-Unit-specific part numbers
General Knowledge Technical Knowledge
Business-Specific Context
Organization-
Specific IP
-Business glossary & terms
-Internal process & knowledge
-Parts inventory
-Service tickets that document
past problems and solutions

23. Copyright © 2024Earley Information Science, Inc. All Rights Reserved.
Content Audit | Third-Party Manuals
•Equipment manufacturer manuals
•Company services a limited
number of products & models, but
they have hundreds of documents
in the repository
•Unstructured PDFs
•Tricky visuals & tables
•Context is implicit in content
23
Conclusions
•Curate the content set.
•Multiple AI models will be needed
to extract meaning.
•Taxonomy and ontology required
to apply explicit context.
•Requires agentic approach to call
different services and models.

24. Copyright © 2024Earley Information Science, Inc. All Rights Reserved.
Content Audit | Organization-Specific IP
•Sparse internal process documents
•No formal knowledge or content management
process or system
•Planned integration with other information
systems to create more context:
•Work orders and jobs
•Parts inventory
24
Conclusions
•Not enough company-specific knowledge
for an AI model to understand their
specific context.
•Context clues needed to support our use
cases must be designed into the
information architecture.
•IA does the heavy lifting
•Used AI to help with IA (VIA)

25. Copyright © 2024Earley Information Science, Inc. All Rights Reserved.
User Research & Information Needs
User Research Activities | Interviewed Service Managers and technical trainers,
surveyed Technicians, Seth attended an on-site training course & more.
25
Narrative-Based
Persona

26. Copyright © 2024Earley Information Science, Inc. All Rights Reserved.
Audience Analysis & Use Cases | AI-Generated, Expert-Refined
26
UX Wireframes

27. Copyright © 2024Earley Information Science, Inc. All Rights Reserved.
STRUCTURING &
ENRICHING CONTENT

28. Copyri ght © 2024Earl ey Information S cience, Inc. A ll Right s Reser ved.
Ontology | AI-Generated, Expert-Refined
28
Key Entities & Relationships
•Truck/Trailer → Has → Refrigeration Unit
•Refrigeration Unit → Contains → Components (Compressor, Evaporator,
Condenser)
•Refrigeration Unit → Uses → Refrigerant
•Refrigeration Unit → Has → Temperature Sensors
•Refrigeration Unit → Experiences → Temperature Fluctuations
•Temperature Fluctuations → Indicate → Potential Issues
•Potential Issues → Require → Diagnostics
•Diagnostics → Uses → Fault Codes
•Diagnostics → Refers to → Troubleshooting Guide
•Service Technician → Performs → Repairs
•Service Technician → Uses → Tools
•Service Technician → Refers to → Repair Manual
•Repair Manual → Contains → Procedures
•Replacement Parts → Are Used in → Repairs

29. Copyri ght © 2024Earl ey Information S cience, Inc. A ll Right s Reser ved.
29
Domains & Taxonomies | AI-Generated, Expert-Refined

30. Copyri ght © 2024Earl ey Information S cience, Inc. A ll Right s Reser ved.
Ontological Relationships | Product Hierarchy
30

31. Copyright © 2024Earley Information Science, Inc. All Rights Reserved.31
Flexible CMS Structure
•Includes fields to support lifecycle
management, workflow, security, and
descriptive metadata .
•Based on the DITA XML standard to leverage
content portability and transformation.
•All content components inherit metadata at
the document level and possess their own
metadata and embeddings, determined by AI
processing and models.
Knowledge Base Automation
•Captures knowledge as questions are asked
and the RAG application responds.
•Uses generated content to fill in knowledge
gaps.
•Integrates with stored conversations.
CMS Content Model| AI-Generated, Expert-Refined

32. Copyright © 2025 Earley Information Science, Inc. All Rights Reserved.www.earley.com
Poll
32
1.Unusable (lack of trust in results, little relevance or
accuracy)
2.Unpredictable (some accurate results, but answers vary
widely)
3.Somewhat trustworthy (many answers are correct but still
contain significant errors)
4.Answers are technically correct but do not help users
complete their task
5.Mostly producing consistent, accurate and helpful results
6.Have not characterized our results yet
7.N/A or none of the above (tell us in Q&A tab)
How would you characterize your RAG results?

33. Copyright © 2024Earley Information Science, Inc. All Rights Reserved.
AI INTEGRATION &
CONTEXTUALIZATION

34. Copyright © 2024Earley Information Science, Inc. All Rights Reserved.
With unstructured content, complex tables, and detailed diagrams,
a variety of AI models and skills were selected and applied.
34
No single AI model can handle every
piece of content effectively.

35. Copyri ght © 2024Earl ey Information S cience, Inc. A ll Right s Reser ved.
IA-Directed Enrichment Using Azure AI
35
Reference
Schema
&
Metadata
(IA)
Layout Analysis to Detect Structures
•Alarm Code tables
•Troubleshooting checklists
•Compatibility, Make & Model
OCR to Extract Text in Images
•Process schematics & diagrams
•Detect and classify parts &
components
•Segment or associate visual
content on repair
Analyze Chat History for Fine-Tuning
•Extract intent & context
•Extract semantics: topics and
concepts

36. Copyright © 2024Earley Information Science, Inc. All Rights Reserved.
Additional Contextualization Opportunities Through System Integrations
36
Work Order System ​
Pre-fill job details & track estimated repair time.
IoT Sensors & Alarms
Auto-detect faults before techs enter anything.
If suction pressure is too low, the assistant can suggest
checking for refrigerant leaks before the technician
types anything.
If a unit is in a high-temperature area, suggest inspecting
the condenser coil for airflow restrictions.
GPS & Geofencing Remote Location
Suggest temperature adjustments or preventive
checks based on the temperature at the service
location.
Maintenance Logs
Pull repair history to suggest related issues.
This unit had a compressor replacement 3 months ago.
Would you like to check if the issue is related to the new
part?
This fuel filter was last replaced 1,200 hours ago.
It may be due for service.
​Parts Management
Suggest & locate needed parts instantly.
The expansion valve appears faulty. There are 3 in stock at
your closest warehouse. Would you like to place an order?
Pocket
Technician
Application

37. Copyright © 2024Earley Information Science, Inc. All Rights Reserved.
MODEL SELECTION &
CONTENT EXAMPLES

38. Copyright © 2024Earley Information Science, Inc. All Rights Reserved.
Choosing the Right AI Skills: Matching Models to Content Needs
With unstructured content, complex tables, and detailed diagrams,
a variety of AI skills were selected and applied.
38
Content Type or RepresentationNecessary AI Skills & Outcomes
Tables & PDFs Document Intelligence for structured data extraction
Images & Visuals Computer Vision to identify, describe, and classify images
Text & Metadata Natural Language Processing (NLP) and Semantic AI to structure, tag, and categorize content
Retrieval & AI-Generated Responses Embeddings & Vector Search for context-aware results
Content Component RelationshipsKnowledge Graph integration
No single AI model can handle every piece of content effectively.

39. Copyright © 2024Earley Information Science, Inc. All Rights Reserved.
Example | Complex Tables
39
PDF AI-Cracked Chunk
Document Intelligence Models

40. Copyright © 2024Earley Information Science, Inc. All Rights Reserved.
Example | Key-Value Pairs
40
PDF AI-Cracked Chunk

41. Copyright © 2024Earley Information Science, Inc. All Rights Reserved.
Example | Key-Value Pairs
41
Document Intelligence Model

42. Copyright © 2024Earley Information Science, Inc. All Rights Reserved.
Example | Detailed Visuals
42
OCR & Computer Vision

43. Copyright © 2024Earley Information Science, Inc. All Rights Reserved.
HOW PREDEFINED PROMPTS
ADD CONTENT TO QUERIES

44. Copyright © 2024Earley Information Science, Inc. All Rights Reserved.
Smart Prompts, Faster Fixes:
Guiding Technicians to the Right Information
Getting to an answer can
involve multiple steps and
careful thought.
Users must ask the right
question with enough context
to get a correct and complete
answer. That puts a lot of
burden on them.
How do we make getting the
right information easier?
44
Predefined prompts for the
most common information
needs, sorted by popularity
A library of predefined
prompts that can grow as
information needs grow
Ask any question

45. Copyright © 2024Earley Information Science, Inc. All Rights Reserved.
Predefined Prompts | Why TheyMatters
DEFINITION | Pre-set, technician-friendly questions that streamline information exchange between
the AI assistant and the human user.
45
Challenges with Manual User Entry Experience with Predefined Prompts
Unclear Intent: AI struggles to understand what the user really
wants.
Goal-Oriented Design: Each prompt is structured for a specific
diagnostic or repair task.
Unexpected Answers: Responses may not align with technician
expectations.
Predictable Responses: Few-shot learning ensures AI replies in a
useful format.
Inefficient Back-and-Forth: Lengthy conversations slow down
troubleshooting.
Faster Input: A simple tap replaces time-consuming typing.
Inconsistent Results: Different technicians may get different
responses.
Built-In Context: Prompts provide relevant details upfront,
reducing unnecessary steps.
Inconsistent Troubleshooting Methods: Different technicians may
take different diagnostic approaches, leading to inconsistent
repairs and varying service quality.
Standardized Troubleshooting: Ensures consistency across all
technicians and service teams.
Manual troubleshooting can result in variation across Service locations. Predefined prompts ensure consistency and reliability.​

46. Copyright © 2024Earley Information Science, Inc. All Rights Reserved.
Characteristics of a Good Prompt
1.Designed for specific, repeatable
actions. Specific enough to guide
the AI model in the right
direction.
2.Guides the AI toward a structured
task or inquiry.
3.Written in clear and concise
language that's easy to
understand.
4.Provides enough context for the
AI model to ask follow-up
questions and generate
meaningful output.
46

47. Copyright © 2025 Earley Information Science, Inc. All Rights Reserved.www.earley.com
Today’s Panel
[email protected]
https://www.linkedin.com/in/sethearley/
47
Seth Earley
Founder & CEO
Earley Information Science
Sanjay Mehta
Principal Solution Architect
Earley Information Science
[email protected]
https://www.linkedin.com/in/sanjaymehta/
“I do not know of any books that have such
useful and detailed advice on the relationship
between data and successful conversational AI
systems.”
—Tom Davenport, President’s Distinguished
Professor at Babson College, Research Fellow at
MIT Initiative on the Digital Economy, and author of
Only Humans Need Apply and The AI Advantage
Heather Eisenbraun
Knowledge Architect
Earley Information Science
[email protected]
https://www.linkedin.com/in/heisenbraun/

48. www.earley.com
48
We Make Information More Useable, Findable, And Valuable
Earley Information Science is a professional services firm headquartered in Boston and founded in 1994. With over
50+ specialists and growing, Earley focuses on architecting and organizing data – making it more findable, usable,
and valuable.
Our proven methodologies are designed to address product data, content assets, customer data, and corporate
knowledge bases. We deliver scalable solutions to the world’s leading brands, driving measurable business results.
48

49. Copyright © 2025 Earley Information Science, Inc. All Rights Reserved.www.earley.com
THE EIS KNOWLEDGE MANAGEMENT AND
AI READINESS ASSESSMENT
HTTPS://WWW.EARLEY.COM/KM-AI-
READINESS-ASSESSMENT
The EIS Knowledge
Management and AI
Readiness Assessment
A comprehensive evaluation to prepare your organization for
the successful deployment of Generative AI solutions.
Save 20% on Assessments
Contact to schedule an assessment before March 1
st

[email protected]
CODE: EISAIWEB20