Stop Having an Opinion About AI
There are plenty of opinions about AI in L&D. It’s exciting. It’s overhyped. It’s going to transform the way we work. It’s creating more work than it saves.
The more useful question is what AI can do for your team, your learners, and your organization right now. That’s something you can test.
Former Chief Learning Officer. Built skills and AI programs for Google, Novartis, Microsoft, Accenture, and Oracle.
Founder and CEO of Torrance Learning. A background in business processes and data shapes how she approaches learning design.
Neither looks at AI as one big decision to adopt or reject. They break it into smaller questions: Where could it help? What task would it support? What data would it need? How would we know whether it worked?
This guide will help you do the same.
Decide Which Kind of AI Conversation You’re Having
Part of what makes AI feel overwhelming is that we use one term to describe several very different things. Megan Torrance recommends separating it into three areas because each one brings different opportunities, questions, and considerations.
In type one, your organization’s AI policies still apply, but this is often an approachable place for L&D teams to begin experimenting and learning. In type two, people should understand when AI is involved, what it is helping them do, and where they can turn if an answer doesn’t seem right—context that helps the organization build trust as it introduces something new.
In type three, L&D’s role is broader than creating training. Learning teams can help identify which tasks are changing, what employees need to do differently, and where managers may need support.
You may have projects in all three zones at once. Naming the zone simply helps you ask the right questions for each one.
Stop Talking About Skills for a Minute
L&D teams naturally think in terms of skills, competencies, behaviors, and proficiency. AI becomes easier to evaluate when you move one level deeper and look at the tasks that make up those skills.
Consider a skill such as closing deals effectively. It may involve researching a prospect, preparing questions, personalizing a proposal, responding to objections, and writing a follow-up. AI may be helpful for some of those tasks and less appropriate for others.
An illustration of the exercise, not a recommendation—the sort depends on your role, your tools, and your policies.
Instead of asking whether AI can support an entire skill, ask which parts of that skill it could make easier, faster, or more consistent—and which parts benefit most from human context.
You don’t need to map your entire competency model before you begin. One role, one skill, and one useful task can teach you a great deal.
See Where the Gaps Actually Are
Bridge maps skills to roles so you can see proficiency by team before deciding which task to test.
Clean Up What You Have Before Creating More
It’s easy to begin an AI conversation by asking what the technology could help you create. But one of its most useful applications may be helping you better understand the content you already have.
Most learning libraries contain a mix of highly useful resources, outdated material, overlapping courses, and content that employees rarely apply. Because many L&D teams manage large catalogs with limited time, it can be difficult to identify what should stay, what needs attention, and what may no longer serve a clear purpose.
AI can help with that first pass. By comparing learning content with a shared skills taxonomy, an AI-enabled tool can surface likely connections and highlight content that may need a closer look.
The results still need human review. But instead of opening every asset individually, your team can begin with a more focused list of questions and priorities.
The goal isn’t to let AI decide what stays or goes. It’s to give your team a clearer view of the library so you can make those decisions more confidently.
Stop Debating. Run a Pilot.
AI claims can be difficult to evaluate in the abstract. A feature that produces strong results for one organization may have a very different impact in another.
That doesn’t mean you need to wait for complete certainty. It means you need a focused question and a practical way to answer it. As Megan Torrance explains:
“If I offer value to the business, that’s a scientific question, not ‘I want to prove that I am.’ You’re leading there, and it causes you to make all sorts of strange mistakes.”
Megan Torrance, Torrance Learning
A useful pilot isn’t designed to prove that an AI tool works. It’s designed to find out whether the tool improves a particular outcome in your environment.
That may involve comparing a group using the new feature with one that isn’t, choosing a few measures before the test begins, and setting a date to review the results. If the pilot shows little or no improvement, that’s still useful. You’ve learned something before committing more time or budget.
A specific business or performance problem, not a technology goal.
“We believe this AI feature will improve this outcome.”
Pick the evidence up front, and a group that isn’t using the feature where practical.
Record what worked, what didn’t, and what you’d change in the next test.
A pilot doesn’t have to settle your entire AI strategy. It only has to help you make the next decision with better information.
Fix the Data Before Blaming the AI
Some of the most promising AI use cases bring together information that already exists across the business.
For example, a CRM may show how a salesperson works with prospects and customers. An LMS like Bridge can show what the employee has learned, the skills they built, and how they performed. Together, that information may reveal useful connections between learning, behavior, and business outcomes.
Connected, current records are what let reporting answer questions about learning, behavior, and results—rather than just activity.
If information is incomplete, inconsistently structured, outdated, or difficult to connect, AI will have less to work with. In Marc’s words:
“If we don’t have a common understanding of [skills], then that means that we’re going to be measuring the adherence as well as the growth of that skill in different ways, particularly related to the learning management systems [and] even on the HR side.”
Marc Ramos, former Chief Learning Officer
This doesn’t mean the organization needs a perfect data environment before it can begin. It does mean data readiness should be part of the AI conversation from the start.
Sometimes the first step in an AI initiative isn’t choosing the AI. It’s understanding the information you already have and how it can be used responsibly.
Start Smaller Than Your AI Strategy
You don’t need to solve every AI question at once. Start with one useful opportunity, learn from it, and build from there.
Start With Your Own Team
Choose a low-risk internal task, such as brainstorming, drafting, or creating a first version of learning content.
Keep a person involved to review the work, apply context, and decide what is ready to use.
Support Learners Thoughtfully
Be clear when learners are interacting with AI and help them understand what the tool can and cannot do.
Invite feedback so learners can flag answers or experiences that don’t feel accurate or useful.
Join Business Conversations Earlier
Ask where AI is already changing roles, workflows, and expectations across the organization.
Work with business leaders and managers to identify the support employees will need as those changes take shape.
Think in Tasks
Break one high-value skill into the individual tasks employees perform.
Test AI on a frequent, clearly defined task while keeping people involved where context and judgment matter most.
Use What You Already Have
Review your current content and data before investing in something new.
Run a small pilot, define success in advance, and use the results to decide what comes next.
