Somewhere in the last two years, most large organizations made the same bet: buy the AI tools first, figure out the people part later. The tools arrived quickly. The people part never quite showed up. Now boards are asking where the return went, and the honest answer is that it went into licenses nobody was trained to use well.
Donna Medeiros, VP of AI and Data Advisory at Data Society, has watched this play out from the advisory seat, sitting across the table from executives trying to explain a gap between spend and outcome that training would have closed if it had been funded at the same pace as the tools themselves.
The ROI Problem Is Real, and It’s Showing Up in the Data
Executives don’t need to be convinced that something is off. They can feel it in their own budget reviews. Donna confirms what the broader research is already showing.
“The research is showing organizations just aren’t getting a return on investment, so they have stopped and halted some of their AI tools purchasing, or they’re burning through their tokens.”
That second half of the sentence is worth sitting with. “Burning through their tokens” isn’t a description of heavy productive use. It’s a description of spend without a clear multiplier attached to it, usage that consumes budget without a corresponding, measurable outcome anyone can point to. Some companies have responded by pulling back on purchasing entirely, which solves the overspend problem but does nothing to solve the underlying capability problem. Pausing the purchase doesn’t retroactively teach anyone how to use what they already bought.
This pattern is well documented in Data Society’s own research into the $632 billion blind spot in AI investment, where the gap between spend and value isn’t a technology failure so much as an execution failure sitting quietly underneath the tool layer: https://datasociety.com/the-632-billion-blind-spot-in-ai/
Training Got Treated as an Afterthought
If tool spending accelerated, training spending should have kept pace. It didn’t, and Donna is candid about the imbalance.
“Training is not being done right, and it’s kind of neglected, right, small part of the budget. So I hope we’re going to see acceleration in the spending for training and upskilling this year.”
Think about the ratio most organizations actually operated under. A significant tool contract, sometimes seven or eight figures at enterprise scale, paired with a training line item that might cover a single onboarding webinar and a shared folder of documentation. That ratio was never going to produce workforce transformation. It was going to produce exactly what it did: a workforce with access to powerful tools and no structured path to using them well.
The fix isn’t complicated to state, even if it’s uncomfortable to fund. Training and upskilling need to scale proportionally with tool investment, not trail behind it as an afterthought squeezed into whatever budget is left. Data Society’s enterprise AI upskilling guide lays out what that proportional investment actually looks like in practice, and the CIO Dive coverage of the retention deficit makes the adjacent point that upskilling isn’t just a productivity lever, it’s a retention strategy for the technical talent every company is competing for: https://datasociety.com/enterprise-ai-upskilling-guide-2026/ and https://datasociety.com/cio-dive-retention-deficit-upskilling-gives-tech-workers-a-path-forward/
The Missing Link, Named Directly
Donna doesn’t hedge on what’s actually broken in the workforce transformation story. She names it plainly.
“It’s the missing link right now. It’s that missing connection.”
The connection she’s describing runs between two things that should never have been separated in the first place: the tool an employee has access to, and the structured capability to use that tool inside their specific role. Buying the tool without building the connection is like issuing every employee a company car and never checking whether they have a license. The car doesn’t create the capability. It just creates liability and unused potential sitting in the parking lot.
This is precisely the gap Data Society’s advisory work was built to close, formalized in the launch of AI advisory services designed around responsible, outcomes-driven adoption rather than tool deployment for its own sake: https://datasociety.com/data-society-launches-ai-advisory-services-to-support-responsible-outcomes-driven-ai-adoption/
Why Employees Adopt Tools Informally Anyway
Here’s the part of the story that surprises fewer executives than it should: employees aren’t waiting around for formal training before they start using AI. Donna points out the social pressure driving that behavior.
“Everybody else is already talking about using those tools, whatever they are, Claude or Gemini or whatever they are, and they want to not be left behind.”
This is a genuinely important insight for workforce transformation strategy, because it reframes the risk. The risk isn’t that employees will refuse to adopt AI without permission. The risk is that they’ll adopt it anyway, informally, without guidance, without governance, and without any connection to the KPIs the organization actually cares about, purely because the social cost of being seen as behind feels worse than the risk of using a tool incorrectly.
Leaders who respond to that dynamic with restriction policies are fighting the wrong battle. The response that actually works is formal, role-mapped training that gets ahead of the informal adoption already happening, channeling the same motivation, not wanting to be left behind, into structured capability instead of unmanaged shadow usage. That’s the same logic behind treating upskilling and implementation as one connected motion rather than assuming implementation alone will produce capable users: https://datasociety.com/ai-upskilling-vs-ai-implementation/
Most Organizations Skip the Use-Case-Level Discipline
Even organizations that do invest in training often miss the layer of specificity that makes training actually work. Donna is direct about how rare true precision is.
“Most won’t take a use case approach to see, okay, for this specific instance of how we’re using AI in this specific role, how are we going to track getting to our success metric.”
Generic AI literacy training, the kind that explains what a large language model is and shows a few prompting tips, has a place, but it isn’t workforce transformation. Transformation happens at the level of a specific role doing a specific task, with a success metric attached to that exact combination. A customer service rep using AI for ticket triage needs different training, tied to a different metric, than a financial analyst using AI for variance analysis. Treating both with the same generic course misses the entire point of what training is supposed to accomplish.
This is the same discipline Data Society applies in its AI use case prioritization framework: specificity isn’t a nice-to-have layered on top of a good training program. It’s the mechanism that makes the training program measurable at all: https://datasociety.com/ai-use-case-prioritization-framework-2026/
Closing the Gap
For an executive trying to fix this before the next budget cycle, the pattern Donna describes points to a few concrete corrections:
Stop treating training as a rounding error in the AI budget. Fund it proportionally to tool spend, not as leftover.
Assume informal adoption is already happening. Employees are not waiting for permission, so build formal training that gets ahead of that behavior rather than restricting it.
Map training to specific roles and specific use cases, each with its own success metric, instead of one generic literacy course for the whole company.
Treat the training-to-tool connection as the actual deliverable of a workforce transformation initiative, not a side benefit of buying the tool.
Workforce transformation was never going to come from the tool alone. It comes from the connection between the tool and the people trained to use it inside their actual roles, which is exactly the link most organizations left out.
Frequently Asked Questions
AI workforce transformation is the process of building an organization’s capability to use AI tools effectively within specific roles, connecting tool access to structured, role-mapped training and measurable success metrics rather than relying on tool deployment alone.
Buying tools without proportional investment in training creates a gap between access and capability. Employees have licenses but no structured path to using them well, which is why research shows many organizations aren’t seeing returns despite heavy tool spending.
Employees adopt AI tools informally because they see peers already using them and don’t want to be left behind professionally. This social pressure drives adoption faster than most companies roll out formal training programs.
Effective AI training is mapped to specific roles and specific use cases, each tied to its own success metric, rather than delivered as one generic AI literacy course applied across the entire organization.
The biggest mistake is treating training as a small, neglected line item compared to tool spending. Training needs to scale proportionally with tool investment, and it needs to be specific to roles and use cases to actually close the ROI gap.
Close the Training Gap Before the Next Budget Cycle
If your organization has scaled AI tool spending faster than training spending, that mismatch is very likely where your ROI is disappearing. Book time with Donna to map a role-based training plan that closes the missing link between your tools and your workforce: https://meetings.hubspot.com/donna-medeiros/meet-with-data-societys-ai-and-data-advisor
