The L&D model is broken

Companies spend an average of 1,200 EUR per employee per year on training (CIPD, 2025). 90% of that training is forgotten in 30 days (Ebbinghaus forgetting curve). Only 12% of employees say they apply what they learned in training to their daily work (McKinsey). We are throwing away 88% of the training budget.

The problem is not the content. It is the delivery model. 8-hour sessions in a classroom (physical or virtual), the same course for everyone, no follow-up, no connection to real work. It is like trying to learn to swim by watching a PowerPoint.

AI changes the L&D model in four dimensions:

1. Personalization: each employee receives the training they need, when they need it, at the right level.
2. Continuity: instead of one-off events, continuous training in short capsules integrated into the workflow.
3. Practice: simulations, adaptive exercises and immediate feedback, not just theory.
4. Measurement: real data on what works and what does not, not just "course rating" (the famous satisfaction survey everyone fills with 4/5 to leave early).

Personalized learning paths

A personalized learning path is a sequence of training content designed to take a specific employee from where they are to where they need to be. It is not a course catalog. It is a path with a start, destination and intermediate stops.

How AI builds the path

• Input 1: current profile. What does the employee know? Assessed skills, prior training, experience, assessment results.
• Input 2: target profile. What do they need to know for their current role or next career step? Required skills, role competencies, department objectives.
• Input 3: available catalog. What training resources do you have? Internal courses, external platforms (LinkedIn Learning, Coursera, Udemy Business), internal documentation, mentors.
• Output: personalized path. Ordered sequence of content with estimated duration, format (video, reading, practice, mentoring), and completion criterion.

Prompt to generate a learning path

"Generate a learning path for a [position] who needs to develop these competencies: [list]. Their current level in each: [basic/intermediate/advanced]. Timeline: [X months]. Available hours for training: [X hours/week]. Structure: for each competency, 3-5 learning resources ordered from basic to advanced. For each resource: descriptive title, type (video, reading, practice, online course, shadowing), duration, and how to verify the competency has been acquired. Include 1 integrating practical project at the end."

Microlearning: training that works

Microlearning is training in 5 to 15-minute capsules, designed to be consumed within the workflow. It does not replace deep training (a 20-hour negotiation course still makes sense). It complements with continuous reinforcement and just-in-time learning.

Why microlearning works

• Fights the forgetting curve. Capsules spaced over time (spaced repetition) improve retention from 10% to 80% according to Ebbinghaus studies applied to corporate learning.
• Integrates into daily life. 10 minutes while waiting for a meeting, 15 minutes at the start of the day. No need to block an entire day.
• Lowers the entry barrier. "Complete a 20-hour course" generates procrastination. "Read this 5-minute capsule" generates action.
• Enables fast feedback. Quiz at the end of the capsule, immediate result, correction on the spot.

AI as a microlearning generator

AI can transform any long content (manual, course, presentation, technical documentation) into microlearning capsules. A 100-page manual becomes 20 five-minute capsules with quiz included.

Prompt to create microlearning capsules

"Convert this content [paste text or summarize topic] into 5 microlearning capsules. Each capsule: engaging title, explanatory text (300-400 words), 1 practical example from the work context, 1 immediately applicable tip, 3 quiz questions (multiple choice, with correct answer and explanation). Tone: direct, practical, no unnecessary jargon. Level: [basic/intermediate/advanced]."

Skill gap analysis with AI

Skill gap analysis answers a fundamental question: what skills does my organization have and which ones is it missing? Without this answer, training is a shot in the dark.

The process with AI

1. Current skills inventory. AI can analyze CVs, performance evaluations, certifications and completed projects to generate a skills map per person and per team.
2. Required skills definition. Based on the company's strategic objectives, job descriptions and industry trends, AI defines which skills are critical at 12, 24 and 36 months.
3. Gap calculation. The difference between what you have and what you need, quantified by area, team and person.
4. Prioritization. Not all gaps are equal. AI prioritizes by business impact, urgency and acquisition difficulty (can I train or do I need to hire?).

Prompt for skill gap analysis

"Analyze the skill gap of my [department] team. Current team skills: [list with levels]. Department objectives for the next 12 months: [list]. Relevant industry trends: [list]. Generate: 1) Current skills map (table: skill, current level, required level, gap). 2) Top 5 gaps by business impact. 3) For each gap: recommendation (train, hire, or outsource) with justification. 4) Prioritized training plan for gaps that can be resolved through training."

AI as a personalized tutor

Generative AI can act as a one-on-one tutor for each employee. Not as a replacement for the human mentor, but as a 24/7 complement that answers questions, explains concepts, proposes exercises and gives feedback.

Tutoring use cases with AI

• Post-training reinforcement. After a course, the employee can ask questions anytime: "I am not clear on how to apply technique X in context Y". AI explains with examples adapted to their reality.
• Soft skills practice. Simulating difficult conversations (giving negative feedback, negotiating a salary, managing a conflict). AI plays the role of the counterpart and gives feedback on the approach.
• Tool learning. "How do I make a pivot table in Excel that crosses sales by region and product?" AI not only explains but generates the step by step.
• Certification preparation. AI generates exam questions, explains incorrect answers, and adapts difficulty to the employee's level.

Prompt for soft skills tutor

"Simulate a conversation where I am a manager who has to give negative feedback to an employee about their punctuality. You play the role of the employee. React realistically (somewhat defensive at first, but open if the feedback is delivered well). After the simulation, give me feedback on my approach: what I did well, what I could improve, and a specific tip for next time."

Creating training content with AI

Creating a quality internal course requires weeks of work. AI reduces that time to days (or hours for simple content). It does not generate perfect content, but it generates a first draft that an expert can review and improve in a fraction of the time.

What AI can generate

• Video scripts: from a topic, generates the script for a 5-10 minute video with narrative structure, examples and call to action.
• Presentations: slide outline with content, presenter notes and interactive exercises.
• Procedure manuals: step by step with screenshots (textual), decision points and troubleshooting.
• Case studies: realistic scenarios based on the company's context for employees to practice decision-making.
• Assessments: quizzes, case studies, evaluation rubrics, with model answers and grading criteria.

Prompt to create a case study

"Create a training case study for the [department] department on [topic]. The case should: 1) Present a realistic situation in the context of a [sector/size] company. 2) Include 3 decisions the participant must make. 3) For each decision: 3 options with different consequences. 4) Sufficient data to support decisions (numbers, deadlines, stakeholders). 5) Facilitator's guide with the recommended solution and discussion points. Duration: 45 minutes of group work."

Measuring training ROI

Training ROI is the holy grail of L&D. Everyone wants to measure it. Almost nobody does it well. The Kirkpatrick model (1959) remains the most widely used framework:

Kirkpatrick's 4 levels

1. Level 1: Reaction. Did they like the course? (The satisfaction survey.) Useful but insufficient. Liking a course does not mean you learned.
2. Level 2: Learning. Did they acquire the knowledge? (Pre and post training tests.) AI can generate and evaluate these tests automatically.
3. Level 3: Behavior. Do they apply what they learned at work? (Observation, manager feedback, performance metrics.) It is the hardest level to measure and the most valuable.
4. Level 4: Results. Did it impact business KPIs? (Sales, productivity, errors, customer satisfaction.) Requires correlating training with results, something AI can do with enough data.

How AI improves measurement

• Automatic adaptive tests: AI generates personalized pre and post training assessments, evaluates them, and calculates the learning delta per person and per group.
• Behavior tracking: AI can analyze whether employees are using the taught tools or techniques (for example, if the sales team is using the new sales script after the sales course).
• KPI correlation: AI crosses training data with performance data to identify which programs have real impact and which do not.
• Improvement recommendations: based on the data, AI suggests adjustments to program content, format or duration.

AI upskilling for the entire organization

AI upskilling is the most urgent training program for any company in 2026. Not just for IT. For all departments. AI affects marketing, sales, finance, legal, operations, and of course HR.

AI upskilling framework by levels

• Level 1: Awareness (all employees). What is AI, what it can do, what it cannot do, ethical implications. Duration: 2-4 hours. Format: e-learning + case studies.
• Level 2: Basic use (80% of employees). Using LLMs for daily tasks: writing, analysis, summarization, search. Duration: 8-16 hours. Format: practical workshops.
• Level 3: Advanced use (20-30% of employees). Workflow automation, data analysis with AI, advanced prompt engineering. Duration: 20-40 hours. Format: bootcamp + real project.
• Level 4: Building (5-10% of employees). Developing custom AI solutions, integrating APIs, fine-tuning. Duration: 80+ hours. Format: specialized technical training.

Common L&D mistakes with AI

1. Generating content without reviewing. AI generates drafts, not final content. A domain expert must validate every piece before publishing. An error in training material multiplies by every person who consumes it.
2. Replacing human interaction. AI is a complementary tutor, not a substitute for the mentor, coach or trainer. Soft skills training especially needs real human interaction.
3. Measuring only satisfaction. "95% of participants rated the course 4 or above." Great. Now tell me: have they changed their behavior? Have their KPIs improved? If you cannot answer, you are not measuring anything useful.
4. Mandatory training without context. "Everyone must complete the AI course by Friday." Without explaining why, without connecting to their daily work, without follow-up. That is compliance, not training.
5. Ignoring generational differences. A 55-year-old employee needs a different approach than a 25-year-old to learn AI. Not in content, but in format, pace and support.

Ejercicio practico

Puntos clave