The Enterprise AI Truth Nobody Wants to Admit: Your Model Is Only as Good as Your Plumbing

Every enterprise wants an AI transformation story. Very few want to hear that the real work isn’t the model — it’s the four-year slog of rebuilding trust, restructuring how technologists report to business units, and rewiring a 108-year-old company’s data supply chain before a single pilot becomes production. Yet that’s exactly what Cushman & Wakefield’s Chief Digital and Information Officer Sal Companieh described in a recent CIO.com interview, and it’s what separates enterprises delivering AI value from those still running disconnected experiments.

Why The Pilot Craze Was A Trap Most Enterprises Fell Into

According to Companieh’s CIO.com interview, while most organizations were running scattered AI pilots, Cushman & Wakefield deliberately maintained a top-down focus — what she calls the “Cushman Way” — attacking the largest go-to-market and employee experience transformations first across 53,000 colleagues worldwide. Her line lands hard: “Most companies were running pilots. We were building the foundation that would make every pilot worth something.”

Pilot proliferation without a shared data spine is how enterprises accumulate technical debt disguised as innovation. Each business unit spins up its own vendor, its own dataset, its own governance model — and eighteen months later, leadership discovers that none of it composes into an enterprise capability. The cost isn’t just wasted budget; it’s the trust deficit that forms when business leaders stop believing IT can deliver.

If you’re a Fortune 1000 CIO looking at a portfolio of 40 disconnected AI proofs-of-concept right now, the practical move isn’t to launch a 41st — it’s to freeze new experiments until an ownership model and data foundation exist. Our prediction: within the next 18 months, boards will start asking CIOs to justify pilot count reductions the same way they once asked about SaaS sprawl.

The Operating Model Change That Actually Made AI Work

The most underreported detail in the interview is structural, not technical. Four years ago, Companieh embedded technologists directly into every business unit and — critically — gave them accountability for revenue and EBITDA. Outside of cyber and infrastructure, every technology capital investment had to be co-created and co-presented with a business leader.

That shifts the staffing question entirely. When technologists carry P&L accountability, the incentive to build shelfware evaporates. They become co-owners of business outcomes rather than order-takers for feature requests. It also solves the perennial problem of shadow IT: when the central platform is co-created, business units have less reason to route around it.

Imagine you run a regional operations division and want a forecasting model. Under the Cushman approach, you’re not filing a ticket — you’re sitting next to a technologist whose bonus depends on your division’s numbers. That’s a very different conversation than a vendor demo. For leaders weighing whether to build AI-integrated software solutions in-house or contract them out, this operating model is the prerequisite either way. Companieh’s admission that she has “successfully matured our operating model three times” tells you this isn’t a one-time reorg — it’s a discipline.

How Databricks Became A Strategy Anchor, Not A Vendor

Companieh’s framing of the Databricks relationship is worth quoting: “I believe the word ‘partner’ should mean something.” She evaluated the partnership on three dimensions — leadership and co-creation culture, product roadmap alignment with Cushman’s forward capabilities, and actual feature functionality. Notice the order: culture first, roadmap second, features last.

That sequence is the inverse of how most enterprise procurement works, and it explains why so many six-figure platform deals underdeliver. When you buy on features, you’re optimizing for today’s requirements against a roadmap you don’t control. When you buy on roadmap alignment, you’re betting that the vendor’s next three years of investment will match your next three years of ambition.

A concrete example from the interview: Cushman uses Databricks Genie to let business users run natural language queries against enterprise data — identifying missing records, validating quality across systems, reviewing governance policies — without deep technical expertise. That’s not a feature you buy; it’s a capability that only works if the underlying data foundation, governance model, and business-unit trust have already been built. Teams weighing the custom AI versus off-the-shelf SaaS AI decision should note that Cushman didn’t pick one — they picked a platform partner and then built custom capability layers on top.

Our take: transactional vendor relationships are losing ground in AI-adjacent categories. Buyers will increasingly demand roadmap co-investment clauses, and vendors who can’t credibly co-create will lose to those who can.

Why The Real Metric Is Time From Idea To Outcome

Companieh’s measurable outcomes skip dollars entirely and land on cycle time: idea-to-outcome has gone from months to days, and client and acquisition onboarding time has “materially reduced.” She calls out that a question that once took “five phone calls, three emails, and two Teams chats” is now at leaders’ fingertips.

Most enterprise AI ROI conversations are still stuck on cost takeout — how many FTEs a chatbot replaces, how many tickets get deflected. Cushman is measuring something more strategic: the velocity of the decision loop itself. When leaders can answer questions in seconds instead of days, that advantage compounds across every strategic decision they make.

If your organization is currently running an AI business case built entirely on labor savings, you’re likely under-selling the actual value. The bigger prize is the orchestration of enterprise data pipelines that turn tribal knowledge into queryable intelligence. Our prediction: within two years, “decision cycle time” will replace “cost per query” as the dominant enterprise AI KPI on quarterly business reviews.

FAQ

Q: What is an enterprise AI core? A: An enterprise AI core is a centralized foundation of unified data, shared platform standards, and governance that supports AI capabilities across every business unit — rather than isolated pilots owned by individual teams. Cushman & Wakefield’s version combines a product operating model, co-invested capital allocation, and a Databricks-based data and intelligence layer that flexes to business-unit needs.

Q: Why do most enterprise AI pilots fail to scale? A: According to the pattern Companieh describes, they fail because the underlying data foundation, operating model, and business-unit trust weren’t built first. Pilots that run on siloed data with no shared governance can prove a concept, but they can’t compose into an enterprise capability — so scaling requires rebuilding the foundation the pilot skipped.

Q: How long does building an enterprise AI foundation actually take? A: Companieh’s own account is four years and three separate operating model maturations. She’s clear that the technology surge in the market was “a pace accelerator and not a pivot in strategy” — meaning the foundational work predates the current AI wave, and organizations starting today should plan for multi-year timelines, not multi-quarter ones.

Key Takeaways

• Freeze new AI pilots until you have a data foundation and ownership model that can actually absorb their outputs — proliferation without a spine creates technical debt disguised as innovation.
• Give technologists P&L accountability inside business units; without shared economic incentive, co-creation becomes theater and shadow IT becomes inevitable.
• Evaluate AI platform vendors on culture and roadmap alignment first, features last — features age out in 18 months, but strategic alignment compounds over years.
• Replace “cost per query” with “decision cycle time” in your AI business cases; the compounding value of faster leadership decisions dwarfs labor-savings math.
• Budget for human change management as a first-class workstream, not an afterthought — Companieh’s core warning is that people, not technology, are the actual bottleneck to enterprise AI scale.