Rebuild the Rails: Why AI only works when your core does

AI adoption is real—but value is uneven

Enterprise AI use has exploded. McKinsey’s latest survey shows 71% of organizations now regularly use generative AI in at least one function (up from 33% in 2023). Overall AI usage spans multiple functions for most respondents. (McKinsey & Company) And yet, three in four companies struggle to achieve and scale value from AI initiatives. The leaders are pulling ahead; the rest are piling on pilots. (Boston Consulting Group)

The gap isn’t about demos. It’s about plumbing.

The dealership lesson: features vs. foundations

What I saw at the service desk is a pattern across industries: a shiny AI layer on a creaky core. In auto retail, the Dealer Management System (DMS) is the operational backbone—inventory, service, invoicing, CRM. Industry analyses call DMS modernization the centerpiece of dealership transformation because it centralizes operations, real-time insights, and integrations. When that core fails, everything else grinds. During the 2024 CDK outage, thousands of dealers reverted to manual paperwork to keep operating. That’s not a workflow AI can rescue with a transcript or a chatbot. (Reuters)

Swap “dealership” for your company’s ATS, CRM, ERP, or finance stack—and the lesson holds. If the system of record is fragmented or outdated, adding AI at the edges mostly creates faster noise.

The hidden tax of legacy

Why does it feel so hard to “go all in” on AI? Because legacy consumes oxygen. In some sectors, 70–80% of IT spend goes to keeping old systems alive—leaving too little capacity for reinvention. Banks routinely spend up to 70% of their IT budget maintaining legacy platforms; public-sector watchdogs cite similar numbers in critical agencies. Even when mainframes are modernized, it’s telling that 80% of enterprises changed their mainframe strategy specifically to accommodate AI—proof that core decisions determine AI outcomes. (IT Pro)

Most “transformations” still fail—humans, not models, are the blocker

We’ve known for years that ~70% of transformations miss their goals. BCG shows you can flip the odds—if you address six success factors end-to-end—but the default is failure. (Boston Consulting Group) The biggest variable isn’t the model; it’s people. Prosci’s benchmarks show projects with excellent change management are up to 7× more likely to meet objectives. Conversely, when sponsorship is weak or employees feel excluded, adoption collapses (leaders say 74% involve employees; only 42% of employees agree).

If your workforce experiences AI as “something done to them,” you’ll get shadow processes, workarounds, and stalled ROI—no matter how promising the pilot.

From “speed up the steps” to “rebuild the system”

The wrong question is: How do we use AI to make the current process faster?
The right one: How do we rearchitect the workflow so AI sits at the core—and humans do what only humans can?

Here’s what that looks like in practice:

1. Start with the system of record.
   Consolidate the data layer first. If skills, requisitions, contracts, timesheets, and payments live in five tools, your model is learning from contradictions. In dealerships, the DMS is the control tower; in hiring, your core workforce platform must be that single source of truth. (This is why DMS/ERP modernization sits at the center of real transformation roadmaps.)
2. Give AI the initiative—keep humans in command.
   Let an agent continuously source, screen, and schedule; let recruiters calibrate context, tell the story, and close. The highest-performing orgs treat AI as an always-on teammate and humans as the final arbiters of fit, risk, and ethics. This aligns with what we see broadly: adoption is highest where AI handles volume and people keep judgment.
3. Instrument everything (so it learns).
   Real-time metrics for pass-through, time-to-fill, acceptance, quality-of-hire, DEI mix—piped back into the system weekly. Organizations that wire feedback loops into their AI stacks report double-digit accuracy gains and steadily better candidate experiences.
4. Treat change as a first-class workstream.
   Executive sponsorship, manager enablement, role-based training, and transparent comms are not “nice to have.” They are the difference between proof-of-concept and production. The evidence is consistent: strong change management multiplies success odds; weak sponsorship drops success to ~27%.

A practical rubric: core, flow, value

As a CPO, this is the three-part lens I use before we green-light any AI initiative:

• Core: Is the system of record clean, connected, and extensible? If not, fix that first.
• Flow: Does AI own the high-volume steps end-to-end, with humans in decisive moments? If not, we’re doing feature theater.
• Value: Can we measure lift (speed, quality, cost, experience) weekly—and feed it back? If not, we won’t compound learning.

If a proposal can’t pass those gates, we don’t ship it. We’d rather modernize the rails than launch another glossy edge feature that dies in year two.

Bringing it back to the service bay

My dealer’s AI inspection video wasn’t the problem. It was the island it lived on. Imagine the leverage if the core ran on a modern platform: the intake would auto-populate from history; approvals would be digital; parts availability and pricing would sync in real time; payment and warranties would reconcile without paper. Same staff, same customers—different throughput.

That’s the promise (and the requirement) with AI at work. Add it to a modern core and you transform capacity. Add it to legacy and you accelerate the bottleneck.

How we live this at WorkGenius

We’ve rebuilt those rails at WorkGenius—so AI isn’t just a bolt-on, it’s the backbone. Our AI runs on clean, connected data and a single source of truth, with recruiters in command where judgment matters. That’s how we turn “AI pilot” into real speed, quality, and visibility.

And because every AI system still needs the right humans, we’ve got those too—skilled, screened, and ready when you are. The technology is only half the equation. The other half is people who can wield it well.

Bottom line: AI isn’t everything. Core systems + change are the force multiplier. Get those right, and the “wow” moments won’t stop at a demo—they’ll show up in your throughput, your candidate experience, and your P&L.