The Future of Managed Learning Services in the Age of AI

An AI tool can write a training module, tailor it by role, and answer a dozen follow-up questions in the amount of time it takes a human to write a paragraph. If you've been to a demo like that recently, you know it sells well enough that someone on your leadership team is almost certainly wondering aloud, "If AI can do this, why are we still paying a managed learning services partner?" Run that experiment for a few months, and the answer usually reveals itself. Content drifts out of date. No one verifies what the AI recommends. It never gets beyond the initial use case.

That stall is the tell. Three months in, someone still has to verify whether the AI's recommendations were right, update the model when a compliance rule changes, and figure out what to do when personalization makes a mistake for a particular team. No one was assigned to do that job, so it’s not getting done, and the pilot quietly ends. That task, not the LMS logins or the trainer roster, is what a managed learning services partner is truly being compensated to own today.

Most organizations aren't equipped to answer that internally. Most vendors marketing themselves as AI-ready haven't built the operating muscle to answer it either, even if strategy tops their own pitch deck.

The Question Everyone's Asking Is the Wrong On

"Build or buy" is the framing most companies default to once AI enters the L&D conversation. That framing assumes the constraint is access to technology. It isn't.

Research from The Josh Bersin Company, published in February 2026, found that fewer than 5% of companies have deployed AI-native technology in learning and development, and less than 10% have an actual AI strategy for L&D at all. Every enterprise already has access to AI tools. What's missing is the operating capacity to deploy them responsibly and keep them working.

Set that against LinkedIn's 2025 Workplace Learning Report, which found that 71% of L&D professionals are already experimenting with or integrating AI into their work. Most teams have started something with AI. Almost none have finished it. That gap between starting and finishing is an execution problem, and execution at scale is exactly what managed learning services exists to solve, provided the partner knows how to do it in an AI environment rather than the one that came before it.

What the Data Actually Shows

Here's how big that gap is. The Josh Bersin Company's research, drawn from more than 50 case studies and 800-plus organizations, found that 74% of senior leaders believe their companies lack the skills to compete, despite global training spend sitting at roughly $400 billion. Fewer than 30% are satisfied with how they're developing workforce skills, and only one in four have integrated learning into the flow of work.

The same research identified a small group, about 5%, that reached what it called "dynamic enablement" maturity: AI-enabled, workflow-embedded learning that adapts continuously. Those organizations were 28 times more likely to unlock employee potential, 6 times more likely to exceed financial targets, and 7 times more likely to achieve high productivity than the rest.

That gap between the 5% and everyone else isn't a technology gap. It's an operating model gap, and operating models are exactly what managed learning services was built to change, not just administer around.

Why This Gap Costs More Than It Looks Like

Only a quarter of learning happens inside the actual workflow. For the other three quarters, someone has to find forty-five minutes in a week that doesn't have forty-five minutes to spare, so the module slips to next week, then next month. Nobody's job is to catch that slippage, so it never gets flagged, and the skills gap on next quarter's leadership report looks the same as this quarter's.

The learning teams closing that gap aren't running fancier AI models than everyone else. They have a person to verify on a weekly basis that the AI's suggestions continue to reflect what actually changed in the business the prior week. A platform isn’t able to do that by itself. It needs a partner whose mandate belongs to recognizing when a compliance update or a reorg shifts what the AI should be suggesting, and one who can make those updates prior to the next review cycle, as opposed to the one after.

What Separates Top Managed Learning Services Providers From Everyone Else Selling the Same Pitch

Pull up an actual RFP scorecard for a managed learning services deal. You'll find a column for seat licenses, a column for cost per learner, a column for catalog size. You won't find a column for who checks the AI's recommendations six months after go-live. AI didn't create that blank column, it just made it show up faster. A few things separate the providers who can actually fill it in from the ones still filling out the rest of the sheet.

They can show you the data pipeline, not just the demo: Ask where the data comes from, not for another demo. Name the system that's supposed to be feeding your AI's skill recommendations, then ask who owns it and how often anyone checks it for accuracy. Most vendors get through the first question fine. Few have an answer for the second, because nobody's actually assigned to that job yet.

They treat content as a living system, not a delivery schedule: Inquire about the last time the compliance module was tested against the real policy, not when it was last published. Most providers will tell you the publish date. Considerably less can tell you if anyone ever re-read it in the light of a policy change three months ago, or if that gulf is currently bridging a quarterly refresh gap.

They own governance as a deliverable, not an afterthought: Someone has to be accountable for what an AI-personalized learning path recommends, whether it's accurate, and whether it holds up across employee populations. If a provider can't describe their governance process in specific terms, they don't have one yet.

They redesign the operating model instead of staffing the old one: The highest-value engagements aren't about adding headcount to run existing programs. They're about rebuilding how the learning function makes decisions, so the AI has good data to work with and a governance layer to sit inside.

One Way This Plays Out

Six months into a manufacturing engagement, a plant supervisor had just completed a module refresher on a machine guarding procedure. The rewritten procedure was twice rewritten since this module was filmed and once after a near miss, once after a new machine came online. But no one had returned to make sure the video was still aligned with the current procedure, because conducting that review was no one’s job. That gap, not the LMS login count or the trainer roster, was the real work the contract was supposed to cover.

One standing rule: whenever the safety team filed a revision, the learning team was handed a ticket, and that particular module was taken offline until it was re-shot. After that policy was instituted, the manufacturer quit inviting the partner to the quarterly reviews of the training calendar, and began inviting them to the safety team’s revision meetings.

The Real Question

Go back to that pilot from the start of this piece. The AI did not hallucinate; it wrote the module as promised. What never did get implemented was an analog of that ticket rule from the production narrative: something that detects a policy change and pulls a module back for a rewrite. There's no one in most companies to do that. Nor most vendors, no matter what their strategy deck's slide says.

So the next time someone asks whether AI has made a managed learning services partner unnecessary, ask them just one thing: show me the ticket rule. If they can't show you some mechanism for detecting a policy change and pulling a module before a supervisor is forced to sit through out-of-date training one more time, the answer's already there.

The Question Everyone's Asking Is the Wrong On

"Build or buy" is the framing most companies default to once AI enters the L&D conversation. That framing assumes the constraint is access to technology. It isn't.