ServiceNow CEO Bill McDermott: Silicon Valley is getting enterprise AI wrong – Fast Company

Enterprise AI has a delivery problem. Not a technology problem, not a talent problem—a delivery problem. The gap between what vendors demo and what actually runs in production is measured in months, millions, and executive patience. Most mid-market companies sitting through AI pitches right now will never see the returns they’re being promised, and it won’t be because the models don’t work.

This piece is for the operations leader or IT head who has been asked to “bring an AI strategy to the leadership team” and suspects the vendor decks aren’t telling the whole story. You’re right. They’re not.

The Demo-to-Production Gap

A recent Fast Company piece featuring ServiceNow’s CEO highlighted a tension that mid-market buyers feel acutely: the enterprise AI conversation has been dominated by consumer-tech thinking. Big models, impressive demos, vague promises about transformation. What gets lost is the boring, expensive work of making AI useful inside existing operations.

Here’s what the gap looks like in practice. A vendor shows you an AI assistant that drafts customer responses in seconds. Impressive. What they don’t show you:

• The three months your team spends mapping edge cases the model gets wrong
• The integration work to connect the assistant to your actual CRM data, which is messier than the demo data
• The workflow redesign required so employees know when to trust the AI and when to override it
• The ongoing cost of monitoring, retraining, and handling the 15% of cases where the AI confidently produces wrong answers

Production typically costs 3–5x what the proof-of-concept cost. That multiplier isn’t a failure of planning—it’s structural. The POC proves the model works. Production proves your organization can absorb it.

Why Consumer AI Thinking Breaks in Enterprise

Consumer AI products can ship fast and iterate publicly. Enterprise AI cannot. The differences matter:

When a consumer chatbot hallucinates, users laugh and try again. When your AI assistant sends a wrong quote to a customer or misclassifies a support ticket, you have a real problem. The organizations that succeed with enterprise AI treat it as an operations project with a technology component, not the reverse.

The Data Foundation Problem

Most mid-market companies underestimate how much work sits between “we have data” and “we have data the AI can use.” Your CRM has ten years of records, but contact roles are inconsistent, deal stages mean different things across regions, and half your product catalog lives in a spreadsheet someone emails around quarterly.

AI doesn’t fix data problems. It amplifies them. A model trained on inconsistent data produces inconsistent outputs at scale. Budget 30–40% of your AI project timeline for data preparation, and expect that number to grow once you start.

The Change Management Problem

Adoption risk is where enterprise AI projects actually die. The pattern: leadership buys a platform, IT implements it, and six months later usage data shows that 80% of the target users either ignore the tool or use it once and stop. The model worked. The organization didn’t absorb it.

Successful deployments invest in three things most projects skip:

• Clear guidance on when to use AI outputs versus human judgment, written for specific job roles
• Workflow integration that makes the AI the path of least resistance, not an extra step
• Visible executive sponsorship that signals this isn’t optional and won’t be abandoned in two quarters

What the Vendor Pitch Leaves Out

Enterprise AI vendors—including the ones now positioning themselves as the “practical” alternative to consumer-tech hype—still have incentives that diverge from yours. They want to close the deal and expand the footprint. You want to capture value without breaking what already works.

Questions to ask that most vendor pitches won’t answer cleanly:

• What does the typical customer spend on integration and customization beyond the license fee?
• What percentage of your customers are in production with this use case versus still in pilot?
• What’s the median time from contract signature to measurable business impact?
• When implementations fail, what’s the most common reason?

If the answers are vague or redirect to “it depends on your organization,” that’s information. It means the vendor either doesn’t track outcomes or doesn’t like what the data shows.

A More Honest Framing

Enterprise AI in 2025 is not magic. It’s industrial automation for knowledge work—useful, valuable, and demanding the same rigor that manufacturing automation demanded a generation ago. The companies that succeed will be the ones that:

• Start with a specific, measurable business problem rather than a technology mandate
• Treat the first deployment as a learning investment, not a proof point for the board
• Staff for integration and change management at the same level they staff for the technology itself
• Build internal capabilities to evaluate, monitor, and iterate rather than outsourcing judgment to the vendor

The technology will continue to improve. What won’t improve automatically is your organization’s ability to absorb it. That capability—the ability to deploy, adopt, and iterate on AI tools—is the actual strategic asset. The model is a commodity. The operating muscle is not.

The shift happening in enterprise AI isn’t about which vendor has the better model or the slicker demo. It’s about which buyers recognize that the hard work lives downstream of the purchase decision. The organizations that pull ahead in the next twelve months won’t be the ones that bought first—they’ll be the ones that deployed deliberately, measured honestly, and built the internal discipline to keep improving after the consultants leave.

The model will work. What matters is everything around it.