PMConnection

AI Adoption: A People-First Change Management Journey for PMO Leaders

AI adoption is not just another tech upgrade – it’s fundamentally a people-focused change. In fact, many organizations find that integrating Artificial Intelligence (AI) into operations succeeds or fails based more on human factors than on the technology itself.

This reality makes AI adoption closer to an Organizational Change Management (OCM) challenge than a straightforward Digital Transformation project. For Program/Project Management Office (PMO) leaders, recognizing this distinction is critical to guiding AI initiatives to success.

Understanding Key Concepts:

AI Adoption – “the strategic integration of artificial intelligence technologies into organizational operations to enhance efficiency, productivity, and innovation,” according to a recent definition. It goes beyond mere installation of AI tools; it entails embedding AI into workflows and daily processes so that employees actually use these tools effectively in their work.

Organizational Change Management (OCM) – “the systematic approach and application of knowledge, tools and resources to deal with change,” involving defining and adopting new strategies, structures, and processes to transition from a current state to a desired future state. In practice, OCM focuses on guiding people through change – aligning stakeholders, communicating the vision, training employees, and mitigating resistance – to ensure changes are embraced and sustained.

Digital Transformation (DT) – “the process of using digital technologies to create new – or modify existing – business processes, culture, and customer experiences to meet changing business and market requirements,” as defined by Salesforce. It is a broad initiative that fundamentally rethinks how an organization delivers value by leveraging technologies (cloud, data, mobile, AI, etc.). Digital transformation often involves enterprise-wide modernization and cultural shifts toward digital-first mindsets.

AI Adoption vs. Digital Transformation: Scope and Focus

Digital Transformation is broad and technology-driven, encompassing end-to-end modernization of processes and business models using various digital tools. For example, a bank’s digital transformation might include moving to cloud infrastructure, launching mobile apps, and using data analytics to improve decision-making. The goal is comprehensive change in how the organization operates and delivers value. While digital transformation does require change management (e.g. rethinking processes and securing company-wide buy-in for new ways of working), its focus is on leveraging technology to drive strategic outcomes.

AI Adoption is narrower in scope but deeper in impact on roles and workflows. It can be seen as one component or offshoot of digital transformation – specifically focusing on deploying AI capabilities (like machine learning models, AI assistants, or predictive analytics) within the business. Unlike general digital upgrades, AI systems often take on decision-making or automated tasks that were previously done by humans. For instance, implementing an AI chatbot in customer service or an AI diagnostic tool in healthcare introduces a tool that learns and makes recommendations, not just a tool users manually operate. This means AI adoption can fundamentally alter job duties, workflows, and how decisions are made, even if the overall business processes remain the same. As a result, the cultural and employee-facing challenges in AI adoption are particularly pronounced.

Crucially, AI adoption doesn’t equate to instant transformation success. Many organizations discover that deploying AI technology is the “easy” part; getting people to use it correctly and confidently is much harder. If treated purely as a tech install, AI projects can falter. Installing an AI system without preparing the workforce often leads to low usage and wasted investment.

In contrast, organizations that approach AI projects with a people-first, change-centric mindset see far better outcomes. In summary, digital transformation provides the strategic umbrella under which new technologies (like AI) are introduced, but AI adoption itself demands intense focus on the human side of change.

Why AI Adoption is Essentially a Change Management Challenge

Adopting AI is about changing people’s work, not just IT systems. Prosci’s research highlights that while “AI implementation” is the technical deployment, “AI adoption…is about people” – it’s the process of ensuring the technology is actually embraced in daily work. This makes AI adoption akin to leading a major organizational change. Key reasons why AI initiatives align closely with OCM principles include:

Behavioral Change Required: AI must become a “natural, effective part of everyday work”. Achieving this means employees must alter their routines, learn new skills, and trust new tools. Such behavior change is exactly what OCM practices are designed to facilitate. Like any large change (e.g. adopting a new process or policy), it requires communication, training, and gradual buy-in.
Human Factors Dominate Success: Studies show human factors are the primary barriers in AI projects – fear of job displacement, lack of training, and mistrust of AI outputs are common issues. Employees may worry an AI system will make their role obsolete or may distrust algorithmic decisions. These concerns can lead to hesitation or resistance. Traditional digital transformations certainly encounter resistance too, but AI touches a sensitive nerve: it challenges the very role of human judgment and expertise in a workflow. Over 60% of organizations report people-related issues as the top challenge in AI initiatives, far outweighing technical hurdles.
Need for Communication and Trust: Organizational change management emphasizes transparent communication and stakeholder engagement. This is vital for AI adoption. Employees need to understand why the AI is being introduced, how it will affect their jobs, and what support they will receive. Clear messaging can dispel myths (“AI is here to help you, not replace you”) and build trust in the new tools. Without deliberate change management, rumors or fears can fester, undermining adoption.
Training and Skill Development: A cornerstone of OCM is equipping people with needed skills for the future state. AI adoption often reveals skill gaps – for example, staff may lack knowledge in interpreting AI recommendations or in data-driven decision-making. In fact, insufficient training in AI tools accounts for about 38% of adoption challenges in enterprises. A robust change management approach will include targeted training programs and upskilling initiatives so that employees feel confident and capable using AI. When people feel competent and empowered, they are far more likely to embrace the change instead of resisting it.
Leadership and Sponsorship: No major change succeeds without leadership buy-in and sponsorship – another parallel between OCM and successful AI projects. Executive support provides vision and resources, and signals to the organization that the change is important. Lack of this support is a known failure factor (about 43% of AI adoption failures are attributed to insufficient executive sponsorship). In practice, PMO leaders should ensure an executive champion visibly backs the AI initiative, communicates its strategic value, and addresses high-level concerns. This top-down reinforcement aligns with change management best practices for securing organizational alignment.

In essence, AI adoption lives or dies by how well people adapt to working with AI. The technology may be cutting-edge, but the implementation plan must address human psychology, team dynamics, and corporate culture. This is why treating AI rollouts as change programs – with change managers or OCM frameworks involved – significantly increases success rates. It’s a lesson many digital transformation efforts learned the hard way: even the smartest technology will underdeliver if the people aren’t on board and proficient.

Learning from Examples

Real-world examples underscore how AI adoption thrives with change management (and struggles without it):

Healthcare: A large hospital implemented an AI-powered diagnostics system (IBM Watson for Oncology) to assist clinicians. Initially, doctors were hesitant to trust the AI’s recommendations. The hospital brought in change management consultants to engage the medical staff in the process. They provided hands-on training, adjusted workflows in collaboration with doctors, and maintained open forums for feedback. As a result, the AI tool became an accepted part of the workflow. The outcomes were striking: diagnostic error rates dropped by ~30%, and report turnaround times were cut in half after AI integration. Without the change management effort, those AI tools might have been ignored or under-utilized; with it, the hospital achieved a significant improvement in care quality and efficiency.
Financial Services: A global bank introduced AI-driven analytics to its lending process. When rolled out, loan officers felt threatened by the “black box” algorithms. Recognizing this as a change management issue, the PMO led an initiative to demystify the AI. They held workshops explaining how the AI model worked, framed it as a decision-support tool (not a decision-maker), and gradually introduced it alongside existing processes. The bank also adjusted performance metrics to reward effective use of AI insights, not just loan volume. Over time, adoption climbed. The AI flagged subtle risk indicators that humans often missed, leading to better loan portfolio performance. The key was humanizing the change – making staff co-owners of the AI adoption journey.
When Tech-Only Approach Fails: Contrast these successes with organizations that took a purely technical approach. In some cases, companies have invested in sophisticated AI platforms that technically function well, but users simply bypass them. One common story is an enterprise that launched an AI knowledge base intended to help employees find information quickly – but without proper change management, most employees continued asking colleagues or using old systems out of habit. The AI tool languished with low uptake, and the expected productivity gains never materialized. This “shelfware” outcome can often be traced back to missing OCM elements: users were not convinced of the tool’s value, not trained adequately, or not supported through the change curve. Such examples reinforce that adoption must be nurtured, not assumed.

Implications for PMO Leaders:

For PMO leaders steering AI projects, the takeaway is clear: manage your AI adoption like a change program. This involves combining classic project management rigor with OCM strategies. Here are key recommendations and best practices:

1. Kickoff with Clear Vision and Sponsorship: Ensure there is a well-defined purpose for the AI initiative tied to business outcomes (e.g. “reduce response time by 50% with an AI assistant” or “improve forecast accuracy using machine learning”). Secure an executive sponsor who will champion this vision. A compelling vision from leadership helps rally stakeholders and gives the project authority and priority. Communicate the “why” of the AI adoption early and often.
2. Integrate Change Management into the Project Plan: Treat “people readiness” as a workstream in the project. Conduct a change impact assessment upfront – identify which roles, processes, and teams will be most affected by the AI introduction. Develop a structured plan to manage that change: this should include a communication plan (what messages, to whom, how frequently), a training plan (skills needed and how to develop them), and a resistance mitigation plan (how to gather feedback and address concerns). Leverage OCM frameworks like ADKAR or Prosci’s methodology to structure these activities. For example, if rolling out an AI analytics tool, plan user training sessions and perhaps a pilot phase where feedback is collected and the approach refined. Make the change management tasks as explicit as technical tasks in your AI project schedule.
3. Engage Stakeholders and End-Users Early: Don’t build the solution in a vacuum. Involve representatives of the end-user community in the AI project from design through implementation. This might mean including business SMEs or front-line employees in requirement definition, prototype reviews, and testing. Early involvement creates ownership – users feel heard and are more likely to support the outcome. It also surfaces practical issues early (e.g., if an AI recommendation interface is confusing, you learn this in pilot and can fix it). Some organizations use “AI champions” or early adopters in each department: these are tech-savvy volunteers who try the AI tool early, give feedback, and later advocate its benefits to peers. This peer influence can significantly smooth adoption.
4. Communicate, Educate, Communicate: You cannot over-communicate during an AI adoption effort. Develop clear, tailored messaging for different groups (executives, managers, end-users) about what the AI tool is, why it’s being introduced, and how it will affect them. Address the elephant in the room – if people might fear job loss or reduction of responsibilities, have leadership acknowledge these fears candidly and explain how the organization plans to handle them (e.g., “Our AI will augment your work, not replace you. We are retraining our team to work alongside these tools.”). Share success stories and quick wins as the project progresses to build confidence. Also, educate users not just on how to use the AI, but on basic AI concepts if needed – for instance, if using a machine learning model, explain its accuracy metrics and limitations to manage expectations and foster trust.
5. Provide Training and Ongoing Support: Training isn’t one-and-done for AI. Offer initial training sessions (workshops, e-learning, hands-on labs) to get users comfortable with the AI tool before it’s fully rolled out. Given that 38% of AI adoption challenges come from lack of proficiency, this is a critical investment. Make training role-specific where possible (so people see how AI applies to their job). After go-live, ensure there is a support structure: help desks, “AI coaches,” or an online forum where users can ask questions as they start using the AI in real scenarios. This echoes the change management principle of reinforcing the change. Celebrate those who embrace the AI and share their positive outcomes – it encourages others to follow.
6. Monitor Adoption and Adapt: PMOs are used to tracking project KPIs; for AI adoption, include adoption metrics as success criteria. For example, measure the usage rate of the AI system, the number of transactions or decisions it’s supporting, or user satisfaction levels with the tool. If adoption is below targets, treat it as an issue to be addressed, not an end-user “failure.” Investigate why – perhaps a specific department didn’t get sufficient training, or the AI insights are not as useful in practice and the model needs tuning. Be ready to adjust the rollout plan or provide additional change interventions (like refresher trainings or tweaking the AI’s integration into workflow). Continuous improvement is part of both agile project management and change management; use feedback loops to ensure the AI genuinely becomes part of the new normal.
7. Balance AI Adoption with Other Initiatives: Many PMO leaders juggle multiple transformation initiatives simultaneously. When introducing AI into a portfolio of projects, ensure it aligns with the broader digital strategy. AI adoption shouldn’t happen in a silo; for instance, if you have a digital transformation roadmap, slot the AI project into that roadmap with clear dependencies and contributions to larger goals. Coordinate resources so that the AI project doesn’t starve other critical projects (or vice versa). It’s about integration, not competition: position AI as enhancing ongoing digital efforts (e.g., AI analytics augmenting an existing data transformation program). Also, be mindful of change saturation – if the organization is already undergoing significant change, time the AI rollout in a way that employees aren’t overwhelmed by too many new tools at once. Sometimes a phased approach is better, or bundling training for multiple new systems together if they are related.
8. Leverage Lessons from Broader Digital Transformation: PMO leaders who have managed digital transformations likely learned valuable lessons about change (for example, the importance of leadership messaging or the pitfalls of under-training). Apply those lessons here. However, also recognize AI might present new challenges – for instance, ethical considerations and policy requirements (data privacy, AI bias) could require additional governance work. Ensure the project plan covers these aspects (e.g., involve legal or compliance teams early when deploying AI that affects customers or sensitive data). By handling such concerns proactively, you prevent last-minute roadblocks and build trust in the AI’s outcomes (people are more willing to use an AI if they know it’s been vetted for fairness and privacy).

In summary, PMO leaders should act as both project strategists and change champions for AI adoption. It’s not enough to deliver an AI system on time and budget; true success is realized only when the organization is utilizing AI and gaining value from it. That outcome lies in the realm of people’s behavior and acceptance. As a PMO leader, treat your AI project’s Go-Live as the beginning of the change, not the end. Plan for the post-implementation adoption phase as diligently as you plan the development phase.

By viewing AI adoption through an OCM lens, PMO leaders can ensure that the technology’s potential is fully realized. When employees are prepared, supported, and bought into the change, AI integration can lead to impressive performance gains – from efficiency improvements to new insights – as evidenced by companies that have done it right. On the other hand, approaching AI purely as a tech rollout risks underutilization. AI adoption is a human journey. For PMOs guiding their organizations into this new era, success will come from leading the change, not just the project. Embrace your role as a change leader, and you will help unlock AI’s true value for your business.

Note:
AI Essentials for Project Managers

How to Create an AI Agent that Functions as a Subject Matter Expert on Your Team

Executive Summary

Artificial Intelligence (AI) is evolving beyond simple chatbots and predictive dashboards into the realm of domain-specialized advisory.

Creating an AI agent that functions as a Subject Matter Expert (SME) on your team can significantly enhance decision-making, improve efficiency, and reduce dependency on human specialists for routine inquiries.

These agents augment human expertise, streamlining workflows and providing real-time, context-aware insights.

While enterprise LLM pilots saw significant growth between 2023-2025, the demand is shifting from generic copilots to domain-grounded agents that can reason over proprietary data and external best practices. Building such an agent requires disciplined knowledge engineering, robust governance, and intentional change management.

1. Defining the AI SME Agent

An AI SME agent is a software system designed to understand, reason, and advise within a narrowly scoped knowledge domain. This could be fields like earned-value management, pharmacovigilance, cloud cost optimization, legal, finance, healthcare, IT, customer support, or engineering. They are conversational or task-based agents trained on domain-specific data.

Key capabilities of an AI SME include:

Natural language dialogue and questioning.
Retrieval-augmented generation (RAG) over curated knowledge bases.
Tool invocation (search, calculations, simulations) to extend reasoning.
Continuous learning loop for feedback and emerging knowledge.
Information Retrieval & Synthesis: Quickly accessing and summarizing vast amounts of data.
Contextual Understanding: Interpreting queries within a specific domain.
Problem-Solving & Analysis: Assisting with diagnostics or identifying patterns.
Decision Support: Providing data-driven insights.
Knowledge Transfer: Disseminating specialized knowledge.

It's crucial to distinguish AI SME agents from general AI assistants due to their specialized training, narrow domain focus, and deep understanding. Before development, you must clarify the domain & scope, use cases, interaction style, and accuracy requirements.

2. Core Components and Architecture

Building an AI SME requires several key components working together:

Knowledge Base & Data Sources: This is the foundation, requiring a curated corpus of policies, procedures, wikis, presentations, recordings, structured data like databases and FAQs, unstructured data like PDFs and case files, and potentially live data feeds. Most organizations underestimate data readiness, expecting 40–60 % of time spent on data wrangling. Data should be tagged and structured for retrieval, ensuring data privacy and compliance. Ontology and knowledge graph development help structure this knowledge.
Language Model: A foundation LLM (e.g., OpenAI GPT-4o, Anthropic Claude 3 Opus) forms the reasoning core. This model is fine-tuned or augmented with domain-specific data using techniques like RAG, few-shot or zero-shot prompting, and custom embeddings. You can choose between hosted LLMs or open-source models for deployment flexibility.
Retrieval-Augmented Generation (RAG): A critical technique to ensure up-to-date, factual, and cited responses. RAG involves using a vector database (e.g., Qdrant, Weaviate, Pinecone) to fetch relevant documents based on the user's query. Fine-grained permissions must be baked into the RAG pipeline to prevent leaks. Chunking documents intentionally and embedding with a model aligned to the target LLM minimizes mismatch.
Interface & Integration: How users interact and how the agent connects to other systems. This can include Slack/Teams bots, web chats, REST APIs for system-to-system calls, and plugins for document search or task automation. Connecting to internal tools (CRM, ERP, BI dashboards, project-management systems like Jira, MS Project) via APIs is essential.
Orchestration Layer: An agent framework (like LangChain Agents, CrewAI) routing user intents to tools or the knowledge base.
Reasoning Layer: The LLM combined with policy-aligned system prompts.

Hard Truth: Your cloud bill scales quadratically with context window abuse. Token discipline is a design requirement, not an afterthought. Domain taxonomies & ontologies are also crucial; if you don’t define the vocabulary, the model will.

3. Development Roadmap

Creating an AI SME agent typically follows a multi-phase roadmap:

Phase 1: Define Scope and Use Cases: Prioritize moments of high cognitive load or knowledge bottleneck. Identify high-impact tasks like onboarding, compliance checks, or technical Q&A. Interview SMEs to understand workflows and pain points. Prioritize based on ROI and feasibility.
Phase 2: Data Collection and Curation: Aggregate and clean internal documents, past queries, and expert responses. Perform a content inventory and gap analysis. Convert unstructured assets (PPT, PDF, video) into machine-readable text and add metadata. Ensure data quality, consistency, and format.
Phase 3: Model Selection and Customization/Training: Select the foundation model evaluating provider roadmaps, latency SLAs, and compliance posture. Decide whether to use hosted or open-source models. Fine-tune or instruct-tune on domain Q&A pairs, avoiding overfitting small datasets. Leverage NLP for understanding queries and generating responses.
Phase 4: Build and Test the Agent: Develop a prototype, integrating with vector databases. Build the RAG pipeline, chunking documents intentionally and using aligned embedding models. Integrate tools and workflows, connecting to systems via APIs and implementing tools like calculators. Conduct user testing with SMEs and iterate.
Phase 5: Deployment and Monitoring/Iteration: Roll out in stages (pilot → team-wide). Run champion pilots and track the deflection of SME inquiries. Monitor performance based on accuracy, latency, and user satisfaction. Implement automated evaluations for factuality and citation accuracy. Continuously update the knowledge base and retrain as needed. Processes for ongoing model training and refinement should be in place.

An important step often layered throughout is Guardrails, Evaluation, and Continuous Learning. This includes Human-in-the-loop (HITL) review for critical responses in early phases and establishing user feedback loops.

4. Challenges and Considerations

Several challenges must be addressed:

Hallucination Risk: Inaccurate responses can lead to wrong decisions and rework. Mitigate this with RAG, citations, post-answer verification chains, and human review.
Data Security & Privacy: Risk of regulatory fines and IP loss. Ensure encryption, access control, and compliance (e.g., GDPR, HIPAA). Implement per-user authorization and redact PII before embedding. On-premise deployment can help with privacy concerns.
Change Management & User Adoption: Resistance and lack of trust can hinder adoption. Train users, manage expectations, and foster adoption. Retrain roles so SMEs become model mentors, not replaced personnel. Provide clear use cases and training sessions.
Data Quality and Availability: The "garbage in, garbage out" principle means poor data leads to poor performance. Expect significant time dedicated to data wrangling.
Bias and Fairness: Potential biases in training data can lead to unfair outputs. Audit model outputs and ensure inclusive data.
Cost Overruns: Can lead to budget blowouts. Implement context-window budgeting, caching, and compression. Remember token discipline is key.
Model Drift: Eroded trust over time. Requires scheduled re-evaluations and rolling fine-tunes.
Risk & Compliance: Hallucination and outdated advice are operational risks. A risk & compliance framework is necessary.
Integration with Existing Systems: Compatibility and interoperability can be challenging.
Human-AI Collaboration: Defining the optimal workflow between humans and the AI agent.
Maintenance and Governance: Ongoing effort is required to keep the agent relevant and effective. Budget for refactoring, allocating ~20% of run-rate for maintenance.

5. Benefits and ROI

Integrating an AI SME agent offers significant benefits:

Enhanced Productivity and Efficiency: Reduced time spent on research and information gathering, faster problem resolution, and automation of routine expert tasks. Case studies show reductions in documentation lookup time and improved onboarding speed.
Improved Decision-Making: Access to more comprehensive and data-driven insights. Faster decision loops have been observed in pilot projects.
Knowledge Scalability and Retention: Preserving institutional knowledge and making expert knowledge accessible 24/7.
Reduced Human Expert Burnout: Freeing up human SMEs to focus on higher-value, complex, or creative tasks. Agents can provide a 30–50 % reduction in direct SME interrupts.
Consistent Information and Advice: Ensuring standardized responses and adherence to best practices.
Accelerated Onboarding and Training: Junior staff can reach mid-level competence sooner.

A Project Management SME agent case study showed over 2,000 advisory sessions with an 87% helpful rating and identified schedule risk weeks earlier than manual review.

6. Implementation Best Practices

To ensure successful implementation:

Start narrow, scale later. Win a clear business outcome with a focused problem or pilot use case before adding domains.
Codify tacit knowledge early. Interview retiring experts and capture rationale.
Own your evaluation pipeline. Do not outsource trust metrics to your vendor.
Budget for refactoring. Model and prompt pairings will require maintenance.
Involve Human SMEs Early and Often. Their input is crucial for data curation and validation. Assemble a cross-functional team including AI/ML, IT, and domain experts.
Prioritize explainability and transparency. Users need to understand how the agent arrived at its answers.
Establish clear governance and oversight. Define roles and responsibilities for managing the agent.
Continuous monitoring and improvement. Regularly evaluate performance and update the agent.
Focus on augmentation, not replacement. Position the AI agent as a tool to empower human teams.
Ensure security and compliance.

7. Conclusion

Building an AI agent that truly behaves as a Subject Matter Expert is less about "sprinkling LLM magic" and more about disciplined knowledge engineering, robust governance, and intentional change management. Organizations that treat the agent as a living system—refined, measured, and mentored—will convert expertise scarcity into a scalable advantage. This strategic initiative can transform how teams access and apply knowledge, creating trusted collaborators that enhance human expertise and drive innovation. The future is a collaborative one where human expertise is amplified by intelligent AI systems.

Note: You may find this helpful:

AI Essentials for Project Managers