The Human Factor: Change Management for AI

Change Management is Critical to successful outcomes

Post #9 of The Path to Autonomous CRM with Agentic AI on ServiceNow Series

TL;DR

Organisations that invest in change management are seven times more likely to succeed with AI. The research is clear: two-thirds of AI implementation challenges are human factors, not technical ones. What works? Champion programs involving at least 7% of your workforce, phased pilots that build confidence before complexity, and role-specific training that connects to actual workflows. What doesn’t? Top-down mandates, generic training, and big-bang launches. Start with something simple like incident summarisation, involve employees in design rather than presenting finished solutions, and treat early scepticism as useful intelligence. The technology is ready. The question is whether your organisation is.

Change Management First

Organisations that invest properly in change management are seven times more likely to succeed with AI than those that don’t. Seven times. Not a marginal improvement. Not a nice-to-have. A fundamental difference between projects that deliver value and projects that quietly get shelved.

This statistic is fascinating because it reframes the entire AI conversation. We spend enormous energy debating model selection, prompt engineering and data architecture, which are all important and necessary, but the research keeps pointing to something far less glamorous. The organisations winning at AI aren’t necessarily the ones with the best technology. They’re the ones who recognised early what this actually is: a people initiative wearing a technology costume.

The real leverage point

When Prosci surveyed over 1,100 professionals in 2025 about their AI implementation challenges, the results told a clear story. Technical issues accounted for only 16% of problems. The rest? User proficiency, organisational adoption, cultural resistance — human factors, all of them. McKinsey’s State of AI report found that high-performing organisations are three times more likely to have fundamentally redesigned workflows and secured committed senior leadership. Neither of those is a technology question.

This isn’t to diminish the technical complexity. ServiceNow’s AI capabilities require proper configuration, clean data, and thoughtful integration. But I’ve seen too many organisations pour resources into technical excellence while treating change management as an afterthought — a box to tick once the “real work” is done. The research suggests this is precisely backwards.

The good news? If two-thirds of the challenge is human factors, then two-thirds of the solution is within your direct control. You don’t need to wait for better models or more features. You can start building organisational readiness today.

What employees actually need

The conversation around AI resistance often frames employees as obstacles to overcome. I think this misses something important. Most resistance isn’t irrational — it’s a reasonable response to legitimate uncertainty.

The EY AI Anxiety Survey found 75% of employees were concerned about job security in an AI-transformed workplace. That’s not technophobia. That’s people asking reasonable questions about their future that deserve honest answers. KPMG’s 2025 global study revealed a paradox: fewer than half of respondents trust AI, yet two-thirds use AI outputs without verifying accuracy. This gap between usage and trust suggests employees are adopting tools they don’t fully understand or believe in — hardly a foundation for sustainable success.

What shifts this dynamic? The research points to three things employees consistently need: involvement in how AI gets implemented, training that connects to their actual workflow, and transparency about what AI means for their role. Not reassurance. Not mandates. Genuine engagement.

A study of 654 employees identified five features of successful human-AI collaboration: active involvement in design, transparency about how AI makes decisions, control over processes, control over outcomes, and what researchers called “reciprocal strength enhancement”; the sense that AI makes their work better rather than replacing it. Organisations that build these elements into implementation see higher adoption, greater ownership, and reduced threat perception.

The counterintuitive finding here involves generational differences. Gen Z shows both the highest AI usage rates and the highest resistance to workplace AI initiatives. Familiarity doesn’t automatically translate to acceptance. Younger workers may be more attuned to AI’s potential to reshape career trajectories, which means their scepticism often contains useful intelligence about implementation risks. The organisations doing this well don’t dismiss that scepticism. They channel it into better design.

The champion model that actually works

McKinsey’s research offers a valuable benchmark: organisations that involve at least 7% of their workforce in transformation initiatives double their chances of success. The highest performers involve 20–30%. This isn’t about creating bureaucracy. It’s about distributed ownership.

Champion programs work because they solve the “frozen middle” problem. Executives mandate AI adoption. Frontline workers are expected to use it. But mid-level managers, positioned between both groups, often become the bottleneck. They face pressure from above to show results while managing teams uncertain about what AI means for them. Champions create a parallel pathway that doesn’t depend on every manager being an AI evangelist.

GitHub documented a three-phase approach worth considering. The first thirty days focus on launching the program and recruiting initial champions. The following ninety days build community and enable champions with resources and support. Beyond that, the focus shifts to operationalising what’s been learned and scaling successful patterns. Morgan Stanley achieved 98% AI adoption among wealth management teams using a champion-led model. McKinsey’s internal AI platform reached 92% usage, with 74% regular use, through a peer-driven rollout.

Who makes a good champion? The research suggests millennial managers, roughly ages 35–44, often hit the sweet spot. They combine technical comfort with organisational credibility. They’re senior enough to influence decisions but close enough to frontline work to understand practical constraints. They bridge executive vision and team-level execution in ways that neither pure enthusiasts nor reluctant adopters can.

The key insight is that champions don’t just train people on tools. They model new behaviours, surface implementation problems early, and create a safe space for colleagues to experiment. That social proof matters more than any training manual.

Starting small to go big

The case for phased pilots over enterprise-wide launches is overwhelming. IDC found that structured pilots achieve 65% faster deployment than unstructured approaches. Gartner’s research shows executive sponsorship during pilots boosts engagement by 50%. But beyond the statistics, there’s a practical logic: concentrated bets allow you to learn before the stakes get high.

ServiceNow’s own Now Assist implementation followed similar logic. Their recommendation is direct: “If you have not implemented anything GenAI, start with incident or case summarisation. These will teach you about UX, value measurement, and change management.” This isn’t a conservative play. It’s strategic sequencing. Summarisation features work reliably out of the box, require minimal configuration, and deliver visible value quickly. They build organisational capability for more complex implementations while generating the early wins that sustain executive sponsorship.

Southeastern Railway deployed Now Assist chat summarisation in three weeks. The result was modest in isolation: 13 seconds saved per customer service handover, but across their 72-person contact centre, that translated to 108 hours annually. More importantly, it gave teams direct experience with AI assistance in a low-risk context. Confidence was built from there.

The counterargument to phased approaches is speed. Competitors are moving fast. Markets don’t wait. There’s real pressure to accelerate. But the research on “big bang” implementations tells a cautionary tale: fewer than 30% of AI pilots successfully transition to production when organisations try to do everything at once. Moving fast on a failed implementation isn’t actually fast. It’s expensive rework dressed up as urgency.

Training that connects to real work

Only 14% of frontline workers have received AI training from their employers, despite 86% knowing they need it. This gap represents both a problem and an opportunity. The organisations closing it are seeing substantial returns.

The distinction that matters is between tool training and workflow integration. Tool training teaches people which buttons to press. Workflow integration helps them understand where AI fits naturally into their task sequences — when to use it, when not to, how to evaluate outputs and how to combine AI assistance with their own expertise. IBM found 48% of employees would use AI more often with formal training. But the training that drives adoption isn’t generic. It’s role-specific and context-rich.

ServiceNow’s internal implementation emphasised upskilling existing staff rather than hiring machine learning specialists. Their platform engineers and data analysts, people who already understood the business context, could execute effectively with domain-specific AI training. This approach has a double benefit: it develops internal capability while signalling to employees that AI augments rather than replaces their roles.

The training investment question often comes down to budget allocation. I’ve seen organisations spend heavily on licensing while treating enablement as an afterthought. The research suggests this ratio is backwards. BCG found that successful AI deployments require roughly two-thirds of the effort to be focused on human factors — culture, training, and workflow redesign. Technology is necessary but not sufficient.

What this means for ServiceNow implementations

ServiceNow’s documented results from their own Now Assist deployment offer useful benchmarks: $5.5 million in annual savings, 54% efficiency improvement and up to 38 minutes saved per user per day. But the implementation approach matters as much as the outcomes.

Their model included a federated GenAI program with domain expertise aligned to business areas such as finance, HR, customer support and legal — rather than centralised technical control. They established governance early through a bi-monthly forum chaired by the CDIO. Cross-functional AI working groups included change management leads alongside security, risk, and platform teams. The quality work on the knowledge base happened before AI deployment, not after.

One detail from their experience stands out. They discovered four different internal definitions of “deflection” across the organisation — a basic metric that meant different things to different teams. Standardising that understanding was a prerequisite work that enabled meaningful measurement.

AI amplifies existing organisational clarity or confusion. Getting the fundamentals right first isn’t optional.

EY’s approach to Now Assist adoption reinforces the theme of patience. They tested for eight months before the production rollout, finding that resolution notes and summarisation features were “useful 70–80% of the time.” Their key learning: “Everyone focuses on teaching AI things, but you also have to un-teach it things when new versions come out.” This suggests that AI adoption isn’t a one-time change event. It’s an ongoing relationship that requires continued attention.

The path forward

Gartner named ServiceNow the sole Leader in the 2025 Magic Quadrant for AI Applications in ITSM for the second consecutive year. Forrester documents a 40–60% reduction in incident resolution times and 30% reduction in ticket volumes from AI-driven automation. The platform capabilities are genuinely impressive.

But capabilities don’t automatically translate to outcomes. The organisations capturing real value are those approaching AI implementation as transformation work — building champions, running focused pilots, investing in role-specific enablement, and involving employees in design rather than presenting them with finished solutions.

This requires different muscles than those needed for technology deployment. It asks leaders to move more slowly than instinct suggests, to invest in activities that don’t feel like “real work,” to treat employee concerns as intelligence rather than obstacles. It’s more complicated than buying software. It’s also where the actual competitive advantage lives.

The seven-times improvement in success rate from proper change management isn’t a magic number. It reflects the compound effect of doing many small things right — engaging the right people, starting with the correct use cases, building confidence before complexity. None of it is particularly glamorous. All of it matters.

Start Monday

Identify two or three potential AI champions in your organisation — people with credibility, curiosity, and connection to frontline work. Have a conversation about what they’re seeing, what concerns they’re hearing and what opportunities they notice. Don’t pitch them on a program yet. Just listen. That intelligence is more valuable than any implementation plan you could write in isolation.

What’s worked in your experience for bringing people along on AI initiatives? I’m genuinely curious — the playbook is still being written.