Before AI comes readiness: People, process, and data

Every function wants AI. Finance wants it. Operations wants it. Sales wants it yesterday. But here’s the part that rarely makes it onto the roadmap: Enterprise data isn’t really ready for any of them.

The data is scattered across operational systems, spread across clouds, and siloed inside teams that don’t always talk to each other. The ambition is real. The data foundation, in most organizations, hasn’t kept pace.

Tom Fletcher, who leads business technology re-engineering at NetApp, learned this firsthand.

“AI doesn’t just optimize the work,” he said. “It exposes all the work you’ve been compensating for all along.”

The broken handoffs, the low-confidence data, the decisions that live in someone’s head instead of in a system. AI surfaces all of it. Once you see it, you have two choices: fix it or ignore it.

NetApp chose to fix it.

Start with the business problem, not the model

The decision that shaped everything else came early.

“We decided to treat AI as a business transformation opportunity first and only then as a technology decision,” Fletcher said. That framing is important. It changes how you train people, how you govern data, how you plan power and cooling in the data center. It means fixing the process before deploying the model and cleaning the data before chasing the use case.

The payoff showed up quickly in the supply chain. A process that once took nine days now takes two, a 78% reduction in cycle time, according to Fletcher. These aren’t projections from a pilot; they’re results from a live operating model, delivering $100 million to $200 million in quarterly opportunity for additional shipments and $20 million in projected savings over two to three years.

The method behind those numbers matters as much as the numbers themselves. NetApp didn’t start by picking a model. It started by mapping the bottleneck, redesigning the workflow, and improving data quality. Through a common configuration strategy, roughly 90% of bill-to-order variance shifted away from bespoke configurations and into a common model framework, creating a durable two-day build-to-ship capability that’s structurally more scalable, not just faster. AI entered the picture only after all of that groundwork was done.

The lesson is repeatable: Find a real bottleneck, confirm the data is ready, apply AI deliberately, and define success before you scale.

The risk of moving too fast

Speed is useful until it creates problems you didn’t budget for. Move before the data is ready, and you expose sensitive information. Move without governance, and trust erodes. Move without a clear architecture, and you end up with a parallel infrastructure stack that nobody asked for and everyone now has to maintain.

That last outcome is more common than it should be. The default approach in many organizations is to build a walled garden: copy all data into a proprietary AI stack, separate from the storage that runs everything else. The result is two data estates, two governance models, and a migration problem that compounds over time. There’s a more practical assumption to start from.

Data readiness is a capability, not a chore

For Fletcher, data readiness isn’t a one-off task you complete before each model deployment. It’s an enterprise capability you build once and apply consistently. Discovery, classification, permissions, protection, governance — when those are applied uniformly across both operational and AI workloads, AI can reach trusted enterprise data without requiring you to duplicate the entire estate.

This is the premise behind the NetApp Platform. Its AI Data Services layer makes data AI-ready in place, with built-in discovery, classification, governance, and transformation, so AI and analytics tools can access current data across on-premises systems and every major cloud without ETL pipelines or extra copies. The same unified storage running SAP, Oracle, and your core databases can serve your AI agents. One foundation, not two.

Protection is built in from the start. The platform operates on zero-trust principles, with autonomous ransomware detection, rapid recovery, and access policies that follow the data. When an AI agent operates across your data estate, it works within the permissions that already exist.

NetApp applies these same principles internally, using ONTAP^® software, Active IQ^® digital advisor, and Data Infrastructure Insights solution to manage where data lives, how it’s placed, and how it performs. The objective isn’t more AI. It’s the right data: governed, protected, and ready to create value when it’s needed.

One note about infrastructure: NetApp currently runs at 12 to 15 kilowatts per rack, with a clear roadmap to 25 and then 40. At those densities, data readiness stops being a purely software consideration. Physical infrastructure planning — power delivery, cooling, rack design — becomes part of the same conversation.

Governance: not a brake, not a gas pedal

NetApp didn’t get governance right on the first try. Fletcher is candid about that. Initially, they under-governed: models were running and people had access, but ownership wasn’t clear and trust wasn’t established. The AI worked. People didn’t use it. Then they overcorrected, approval processes multiplied, decisions slowed, and progress stalled.

The balance they’ve settled on is clear ownership, accountability, and success metrics. Enough structure to build trust, enough flexibility to maintain momentum. And one principle holds throughout: “Model output informs human judgment,” Fletcher said. “It does not replace it.” That boundary is what keeps trust intact as AI agents begin acting across the data estate.

From reports to answers

Once that foundation was in place, AI stopped functioning as a productivity add-on and became part of how work actually gets done. The NetApp conversational interface for supply chain intelligence illustrates the shift. Quarter-end is precisely when the demand for trustworthy operational data is highest and reporting teams are stretched the thinnest. AI-based chat replaces static dashboards with a real-time conversational interface. Teams ask questions and get answers, without waiting for a report to be rebuilt or a dashboard to be reconfigured.

That capability only holds up because the data underneath it is discoverable, governed, and trusted. A well-designed interface on top of poor-quality data doesn’t reduce lag; it just delivers wrong answers faster.

The takeaway for your data estate

The organizations that get real value from AI won’t necessarily be the ones running the most tools. They’ll be the ones with the strongest data foundations, the clearest governance structures, and the discipline to move from insight to action. You don’t need a parallel stack. You need the infrastructure you already have, prepared correctly.

NetApp started its AI journey as a business problem, built a governed data foundation, and had the honesty to address what AI exposed. That’s how nine days became two. The model was never the hard part. The data always was.

See how NetApp makes data AI-ready

Frequently asked questions

What does AI readiness mean for enterprise data?
AI readiness means your data is discoverable, governed, and protected before you deploy a model, not after. It requires treating AI as a business transformation initiative first: fixing broken processes and cleaning data, then applying AI to a confirmed bottleneck.

Does AI readiness require a separate AI infrastructure stack?
No. Data readiness is a capability you build once and apply across operational and AI workloads alike, using the same governed storage foundation rather than duplicating data into a parallel AI stack.

What role does governance play in enterprise AI success?
Governance provides the ownership, accountability, and success metrics that build trust in AI systems. Too little governance leaves ownership unclear and adoption stalls; too much slows decisions. The right balance keeps AI output informing human judgment rather than replacing it.