TL;DR: After 25 customer meetings in three days, Quest’s Glenda O’Keefe found the same five patterns everywhere: data products are stalling on delivery, governance behaves like a people problem, AI readiness needs data and people readiness, the modern stack is settled but unmanaged, and agentic AI is closer than roadmaps suggest.
Recently, I had 25 customer meetings over three days about AI-ready data products, governance, and AI readiness, across financial services, manufacturing, pharma, consumer goods, aviation, health systems, and retail. And despite the industry differences, the same five themes kept surfacing.
I’m sharing these themes because they reflect the reality of what data and technology leaders are struggling with right now. Whether you’re a CTO, a data leader, or someone trying to understand why AI isn’t scaling as quickly as promised, everyone is facing the same issues regardless of industry or level within your organization.
A materials technology company needs a three‑week sprint to produce a single data product… when things go well.
1. AI-ready data products are now the universal language, but delivery is breaking down
Every organization I spoke with is either building AI-ready data products, planning to, or frustrated that they aren’t doing it fast enough.
The good news is that, for the first time, the definition has stabilized. Across banking, aviation, pharma, and consumer goods, leaders described a data product in almost identical terms: “A packaged, purposeful unit of data, with business meaning, ownership, lineage, quality signals, and discoverability.”
A bank articulated it perfectly: when someone requests a data product, they should receive not just the data, but metadata, lineage, glossary terms, owners, and downloadable assets. Three other organizations echoed that framing almost word for word.
Again, the good news: the vision is clear. Here’s the catch: The delivery is not.
- A materials technology company needs a three‑week sprint to produce a single data product… when things go well
- A European bank rated its logical modeling maturity 5/10 and its physical delivery maturity 1/10
- A pharma company said everyone wants “lots of data products,” but nobody agrees on where they come from or what makes one good
Across aviation, consumer goods, banking, and manufacturing, the competitive edge is no longer the platform. It’s the cycle time from a business request to a trusted, reusable AI-ready data product.
And right now, that cycle time is the bottleneck.
Organizations have cycled through governance tools, and the same problems follow them from tool to tool. One insurance leader said it plainly: “New tool, same behavior.”
2. Governance looks like technology — but it behaves like a people program
This theme came up again and again. Organizations have cycled through governance tools, and the same problems follow them from tool to tool.
One insurance leader said it plainly: “New tool, same behavior.”
Another described workflows so burdensome that people entered incorrect values just to avoid being penalized. The governance system was actively incentivizing bad data.
The patterns were identical across industries: tools bought centrally with no input from the teams who run them, data owners disengaging when a single field change requires approvals they don’t understand, and processes quietly bypassed as data gets reused without structure.
A pharma company working on master data for “product” captured the root issue of four different semantic meanings in four different departments:
- Manufacturing: “product” means the physical box
- Commercial: “product” means the trade name
- Medical: “product” means the compound
- Regulatory: “product” means the submission entity
Until humans agree on the word, no governance tool solves the problem.
The organizations making real progress treat governance as a change management program, led by business owners and domain leads, with tooling as an enabler, not the solution.
Governance fails when it’s treated as software. It succeeds when it’s treated as behavior change.
An aviation leader offered the framing that unlocked the conversation: “AI is not about fewer people. It’s about more effective people.”
3. AI readiness is being blocked by two things nobody wants to talk about
Every organization is somewhere between “we want AI” and “we have AI working at scale.” The ones furthest along addressed the foundation before they began. One bank’s AI team spends most of its time on data readiness, prompt engineering, context engineering, and building the structured data needed for AI. AI amplifies whatever is in your data estate, including the mess.
But the second blocker is harder to put in a roadmap: people readiness.
Across regulated sectors, I heard fear of regulatory exposure, job displacement, ongoing re-orgs, and unclear roles.
An aviation leader offered the framing that unlocked the conversation: “AI is not about fewer people. It’s about more effective people.”
That reframing removes the threat and opens the door to discussing what changes, not what disappears.
One consumer goods company is already exploring “synthetic co‑workers,” AI agents embedded into team workflows over a 3‑to‑7 year maturity curve. They were clear‑eyed: the governance and model setup is the hard part, not the concept.
That honesty felt more credible than the organizations still promising AI transformation in 12 months.
4. The modern stack is settled — but the hard work isn’t
Across these 25 conversations, the technology choices were remarkably consistent. Most organizations have now standardized on a familiar pattern with a cloud warehouse or lakehouse at the core, modern transformation tooling, and a BI layer on top. The debate about which platform to use is largely over.
But the migration to this modern stack, and the ability to actually operate on it, is far from complete.
A gaming operator is running two parallel data streams while launching in a new market, integrating a new CRM, and stabilizing a new backend. The business expects trusted reporting, but the underlying data platform isn’t mature enough yet.
A financial services organization is mid‑migration to a cloud warehouse with modern transformation tooling, but critical risk models are still on‑prem, creating a split‑brain architecture.
A consumer goods spin‑off moved directly to a modern ERP, modeled initially in a virtualization layer, and is now migrating those models into a lakehouse because the original approach created a black‑box dependency on external consultants. Private equity ownership means the volume and urgency of data questions keeps increasing.
The insight is simple and universal: Legacy does not retire cleanly. Running two stacks simultaneously is expensive, not in platform cost, but in team capacity, reporting inconsistencies, and business trust.
And this is exactly where organizations realize the platform was never the hard part. The hard part is modeling, governance, lineage, quality, ownership, and delivery — the layers that make the stack usable, trustworthy, and ready for AI.
5. Agentic AI is closer than roadmaps suggest, if the foundation exists
This was the most energizing theme — and the one that produced the clearest split.
At one bank, their audit department is working toward industrialized, automated data quality controls, not because an auditor remembered to run them, but because the system does it.
Today, they match auditor SQL skills to SQL databases manually, in Excel. The gap between where they are and where they want to be is not technology. It’s data quality and lineage.
On the other side of that gap, a supply chain leader is piloting AI‑powered order intake where agents read orders from email, PDF, and EDI, handle standard cases autonomously, and escalate exceptions to humans.
The challenge they described, distinguishing an account number from a purchase order number in a poorly formatted PDF, is exactly the kind of problem that sounds trivial but is extraordinarily hard. They’re solving it with an agentic layer, not a rule engine.
The question I found myself asking in almost every meeting: “If an agent could answer any operational data question in real time — without a BI analyst — what would that unlock?”
Every room had an answer. Very few had the foundation to make it possible today.
The competitive edge is no longer the platform. It’s the cycle time from a business request to a trusted, reusable AI-ready data product.
What I’m carrying forward
Three days. Twenty‑five meetings. Dozens of industries and architectures. Here’s the synthesis:
- AI-ready data products first, AI second. Every organization furthest along in AI had first resolved ownership, quality, and discoverability.
- Governance is not a technology project. It is a cultural and operating model shift that uses technology, not the other way around.
- The delivery gap is the real problem. The vision for data products is shared. The ability to deliver them quickly, at scale, with trust signals is where organizations are losing time and confidence.
- Agentic AI is 18–24 months away for most. The decisions being made today on architecture, governance, and quality determine whether autonomous agents will eventually be possible or remain a pipedream.
I left these conversations more energized than when I arrived. Not because everything is solved – clearly it isn’t – but because the leaders wrestling with these problems are asking the right questions.
And they are closer than they think.
