This post was shaped by ongoing conversations we're having across the ServiceNow ecosystem about where AI initiatives succeed and where they stall. If you want to go deeper on this, our Chief AI Officer Ben Durham-Kilcullen covers the full picture — measurement frameworks, use case selection, governance, and what real AI adoption actually takes — in the latest issue of Fortified Quarterly.
Every morning, people wake up and use tools that actually work. They ask a voice assistant for the weather, get a personalized playlist without browsing, split a dinner bill in seconds, and easily book a flight without a travel agent. The experience is fast, frictionless, and almost invisible.
Then they go to work.
They open a ticket in one system. Search for a policy in another. Wait two days for an approval that could have taken two minutes. Navigate a portal built around how the organization is structured, not around how they actually think. At some point, they stop looking for the right answer in the right place and just improvise with a workaround, a Slack message to someone who knows, or a tab left open for later.
This isn't a technology gap. Every organization these employees work for already has powerful platforms in place. The gap is something else: the distance between what people now expect from technology and what enterprise tools are actually delivering.
The Tolerance Has Run Out
For a long time, there was an unspoken trade-off in enterprise software. The tools were complicated but they were powerful. Workers learned the system. They memorized the taxonomy. They internalized which department owned which request, even when the distinction made no sense to anyone but the org chart.
That trade-off is gone.
Employees have spent the last two years using consumer AI that can draft an email from a single sentence, summarize a 40-page document in 30 seconds, and route a complex request without a single dropdown menu. They use these tools at home, on their phones, in their free time. Then they sit down at work and ask — reasonably — why their employer's systems don't work the same way.
The expectation has shifted. It's no longer "help me find the right form." It's "just handle this for me."
That's not an unreasonable ask. But it's a fundamentally different product requirement, and most organizations haven't caught up to it yet. The ones that don't close this gap aren't managing a user experience problem anymore. They're managing an adoption problem. A retention problem. A credibility problem with the people they're asking to drive the business forward.
Three Places AI Initiatives Break Before They Start
A lot of organizations have recognized this moment and responded with urgency. AI initiatives have launched. Licenses have been purchased. Demos have been impressive. And yet, real impact has been slower to arrive than the investment would suggest.
The reasons tend to cluster around three failure points.
The first is the absence of a measurement framework before anything gets built. Teams launch AI initiatives without defining what success actually looks like. No baseline. No target metric. No defined business outcome tied to the investment. The work gets done, and then when it's time to prove value there's nothing concrete to point to. Before any use case is selected, the question should have a clear answer: what KPI does this move, by how much, and over what timeframe?
The second failure point is use case selection. Organizations often gravitate toward what sounds impressive rather than what employees actually do every day. If the AI is automating something someone does twice a month, the return on adoption effort will be minimal. The question isn't "what's the most interesting thing AI could do here?" It's "what are people doing constantly, and where is real effort being wasted?"
The third is infrastructure. Organizations want AI that can take action, but the underlying systems often aren't ready for it. Data is fragmented and systems aren't connected. The permissions model was built for human decision-making, not for automation. Without that foundation, AI ends up as a sophisticated search function. It can answer questions but can't close a ticket, approve a request, or update a record. For what those costs, it should be able to do a lot more.
The Platform Is Only Half the Answer
One thing that gets lost in the focus on AI tooling: the platform shift matters, but it's not sufficient.
ServiceNow has historically been a system of record, where work gets tracked. What AI makes possible is something different: a system of action, where the platform completes steps on behalf of employees within defined guardrails. That's a meaningful distinction. Finding information still puts the burden of interpretation and execution on the person. Getting work done means the system carries that burden.
But realizing that shift requires more than selecting the right platform. It requires thinking carefully about governance such as what decisions an AI can make autonomously, which ones need human review, and how you audit what happened. This is where the real implementation work is. And it's where most organizations underestimate the effort.
It also requires addressing the human side of adoption before the technology goes live. The question "will this replace me?" is present in every room where AI gets introduced to employees. If leadership doesn't answer it directly and honestly, adoption will suffer. Not because people can't use the tools, but because they won't trust them. They'll perform for the demo and quietly go back to doing things the way they always have.
Shadow AI, which is employees using personal AI tools outside approved systems to do work the enterprise tools can't do, is already a real pattern. It's the AI equivalent of shadow IT, with the same compliance and security risks. The way to prevent it isn't restriction. It's building enterprise experiences good enough that employees don't feel the need to work around them.
What This Moment Actually Requires
The organizations getting this right share a few traits. They define success before they build anything. They pick use cases based on frequency and employee effort, not on what demos well. They invest in the data and integration work that makes AI action possible — not just AI answers. And their leadership uses AI visibly and talks about it openly, so employees understand that adoption isn't optional, and isn't something to fear.
The real question isn't whether AI belongs in the enterprise. It belongs. The question is whether the work to deploy it well is being taken seriously, or whether organizations are buying licenses and hoping for the best.
