AI won’t fix L&D if firms don’t fix behaviour data first
AI is quickly finding its way into corporate learning platforms, promising adaptive content, personalised learning journeys and frictionless development processes. But beneath these promises lies a risk: Ross Dickie, learning experience consultant at Mindtools, warns that if the foundation isn’t solid, AI will only speed up the creation of ineffective – or even damaging – learning.
The problem begins with how most organisations understand employee behaviour. In theory, L&D teams design programmes to address specific capability gaps. In reality, many start with incomplete or biased data. Common sources like self-assessments, 360 feedback and engagement surveys provide useful insights into perceptions, but they are far from objective measures of behaviour.
A manager may confidently rate themselves highly in “giving feedback” - not because they excel at it, but because they enjoy the process. Or because their direct reports, who happen to like them personally, provide overly generous scores. Engagement surveys can be equally skewed by question design or the prevailing mood at the time. None of this means these tools are worthless. Indeed, they can raise awareness, spark conversations and be deployed cheaply and at scale. But they are blunt instruments, and relying on them as a single source of truth sets programmes off on an uncertain footing.
That uncertainty matters. If the behavioural baseline is wrong, the problem definition is wrong, and AI will happily accelerate the wrong solution.
The lure – and trap – of AI efficiency
Much of the enthusiasm for AI in L&D centres on its ability to quickly and cheaply scale content. Faced with time and budget pressures, it’s easy to see the appeal. Feed in your needs analysis, policy documents or stakeholder requests, and AI will generate training content in minutes. For many teams, this feels like a breakthrough.
But content has rarely been L&D’s real bottleneck. The bigger challenge lies in diagnosing the right problem and designing interventions that meaningfully change behaviour. By making development so quick and easy, AI risks bypassing the critical thinking stage. There’s little built-in friction to challenge flawed assumptions or prompt deeper analysis.
It’s easy to spot the danger. Imagine an AI tool trained on learning science research, that can instantly recommend the “optimal” design for training on a specific policy. This might sound impressive, until you realise the tool never asked whether training was the right solution in the first place. Were employees unaware of the policy, unmotivated to follow it, or facing systemic barriers? Without those answers, “optimal” design is irrelevant. AI doesn’t yet have the remit or the incentive to push back, and therefore, it falls on the L&D teams’ shoulders to challenge the request.
For this to happen, there needs to be a cultural shift in L&D: the confidence to question whether training is the right answer. Too often, requests arrive framed as “We need a course on X”, bypassing any discussion of root causes. L&D professionals who want to be seen as strategic partners need to challenge these assumptions without alienating the requester.
What a strong behavioural baseline looks like
Avoiding these issues starts with robust behavioural data. That means moving beyond how people feel about their skills to how they actually perform. It also means using validated measurement frameworks, drawn from academic research, to ensure the data is credible.
For example, qualitative insights from focus groups and interviews can help capture lived experiences and test stakeholder assumptions against employee realities. Organisations can then deploy validated behavioural surveys to measure specific actions rather than confidence levels or preferences, ensuring the focus is on what people actually do. They can also include control groups that are matched to participants outside the intervention, making it possible to isolate the programme’s true impact, while business metrics – from performance data to customer feedback – link learning activity directly to organisational outcomes. Together, these different lenses offer a much clearer and more accurate view of where to target interventions.
Using AI to improve learning
AI’s real potential is not in automating design and development, but in freeing up time for higher-value work. If AI cuts development time by a third, that “saved” time shouldn’t simply be pocketed as a cost reduction. Instead, business leaders should be reinvesting it into activities that deepen understanding and improve results.
For instance, with more time freed up, the more time L&D teams can spend conducting deeper discovery, speaking to employees, shadowing their work and building a more nuanced picture of the challenges they face. They can devote more energy to robust measurement, planning for evaluation from the start and identifying clear behavioural indicators alongside control comparisons. And when a programme ends, they can focus on iterative improvement, refining interventions year-on-year using post-programme data, rather than treating delivery as the finish line.
AI can also help at the analysis stage, for example, transcribing interviews, summarising focus group discussions or identifying recurring themes in unstructured feedback. The key is to keep human expertise in the loop, using AI to accelerate insight gathering rather than replace judgment.
Indeed, AI works best when layered over strong professional judgement. Experienced practitioners know how to spot flawed data, question assumptions and design for behaviour change. The risk lies with individuals who haven’t yet built that depth of expertise. For them, AI’s output can feel authoritative, even when it’s built on shaky inputs. In the wrong hands, the technology can normalise low-quality analysis, leading to a generation of learning professionals who can operate the tools but not evaluate their results.
While AI can strip out repetitive admin, make sense of sprawling data sets and speed up prototyping, its value is entirely dependent on the quality of the problem definition and the behavioural insight that underpins it. Without that, AI will make bad learning faster.
L&D teams that want AI to improve outcomes should invest in the skills, methods and credibility to challenge the brief, gather robust data and measure what matters. In doing so, they’ll not only avoid AI’s biggest trap but also unlock its potential to create learning that is truly personalised, relevant and, most importantly, effective.
