Diagnosis: where your team stands today

78% of employees already use generative AI in some form. The problem: only 21% do so with company-approved tools. The rest use ChatGPT with their personal accounts, upload confidential documents to free platforms and make decisions based on outputs no one verifies.

Before designing any talent strategy, you need to answer three questions:

1. Who uses AI today in your organization? Not who says they use it. Who actually uses it. An anonymous 5-question survey: which tool, how often, for what task, with what type of data, with what result.
2. What AI skills does your current team have? Most employees can write a basic prompt. Very few know how to design a workflow, evaluate the quality of an output or integrate AI into an existing process.
3. What skills will you need in 12 months? It depends on your strategy (CX01). If you plan to optimize processes, you need advanced users. If you plan to build AI products, you need engineers and data scientists.

Hiring vs reskilling: the decision that defines everything

The temptation is to hire. An "AI expert" who solves everything. The reality: an external expert without business context takes 6-12 months to become productive. A current employee with AI training takes 4-8 weeks to apply it to their daily work.

When to hire

• You need to build proprietary AI products. Model fine-tuning, data pipelines, ML infrastructure. This requires specialized engineers you cannot train internally in months.
• Your industry requires specific certifications. AI in medical devices, autonomous vehicles, critical systems. Regulation that demands demonstrable expertise.
• You want a CAIO (Chief AI Officer). From maturity level 3 onwards (CX01), someone must lead AI strategy at the executive level. This can be internal or external, but requires real authority.

When to train (reskilling)

• You want the entire team to use AI. Mass training is faster, cheaper and builds culture. A 40-hour program converts a basic user into an advanced user.
• Your competitive advantage is business knowledge. A financial analyst who learns advanced prompting generates more value than a prompt engineer without financial knowledge.
• You need results in under 3 months. Training scales. Hiring does not. You can train 50 people in parallel. You cannot hire 50 AI experts.

The hybrid option (recommended)

Train your current team (80% of the effort). Hire 1-3 specialized profiles to lead the transformation (20% of the effort). Insiders provide context, outsiders provide technical expertise. The result is greater than the sum of its parts.

Cost comparison for a 200-employee company:

AI competencies by role

Not everyone needs to know the same things. AI competencies depend on the role. This framework defines three levels for each business function:

Level 1: User (all employees)

• Basic prompting: formulate clear instructions, provide context, iterate
• Output evaluation: detect errors, verify data, identify bias
• Approved tools: know how to use the company's 2-3 official tools
• Basic security: what data can be shared with AI and what cannot
• Ethics: when to trust the output and when to request human review

Level 2: Power User (middle managers, analysts, creatives)

• Advanced prompting: chain-of-thought, personas, few-shot, reusable templates
• Workflow design: integrate AI into existing processes, automate sequential tasks
• Quality evaluation: accuracy metrics, model comparison, A/B testing of prompts
• Data management: prepare data for AI, clean inputs, structure outputs
• Team training: teach others, create documentation, share best practices

Level 3: Specialist (technical team, CAIO, champions)

• AI solution architecture: choose models, design pipelines, evaluate providers
• System integration: APIs, automations, connectors with existing tools
• Model evaluation: benchmarks, costs, latency, accuracy by use case
• Compliance and governance: AI Act, GDPR, internal policies, audits
• Strategy: AI roadmap, business case, ROI, communication with leadership

The resistance spectrum

When you introduce AI into an organization, your team distributes across four profiles. Understanding each one is the difference between successful adoption and failure.

Enthusiasts (10-15%)

Already using AI on their own. They try new tools every week. They share tricks with colleagues. They are your greatest asset but also a risk: they may use unapproved tools, share sensitive data or create unrealistic expectations.

Action: channel them. Give them official tools, turn them into champions, ask for their feedback. Do not hold them back, but give them structure.

Pragmatists (40-50%)

Open to AI but need to see concrete results. They do not experiment out of curiosity. They ask "what does this save me?" and "does it work with my actual tasks?". They are the silent majority and the key to mass adoption.

Action: give them concrete use cases from their department. Practical workshops, not theoretical ones. Show them before/after with real metrics. If they see a colleague saving 2 hours a day, they adopt.

Skeptics (25-30%)

They distrust AI. They think it generates errors, that it is a fad, that it will take their jobs. They are not irrational: they have seen technology fads that did not deliver on promises. Blockchain, metaverse, Big Data without results.

Action: listen to them. Their skepticism often flags real risks that enthusiasts overlook. Invite them to controlled pilots where they can verify for themselves. Do not impose, demonstrate.

Resisters (5-10%)

They actively reject AI. It may be out of fear (of change, of unemployment), out of principle (ethics, privacy) or out of inertia ("I have always done it this way"). Some have valid reasons. Others simply do not want to change.

Action: do not force direct adoption. Offer training without pressure. Assign tasks where AI is optional, not mandatory. If after 6 months they still do not use it, assess whether their role can remain productive without AI. In most cases, it can.

The distribution is not fixed. A skeptic who sees concrete results becomes a pragmatist. A pragmatist who discovers a transformative use case becomes an enthusiast. Your job as an executive is to create the conditions for people to move toward the left of the spectrum.

Identifying and developing champions

Champions are the most important piece of AI adoption. They are not technical experts. They are people who combine three characteristics: they know the business, they are motivated about AI, and they have influence on their team.

How to identify them

1. Observe who already experiments. Who uses AI without being asked? Who shares tricks in the team chat? Who asks "can we do X with AI?"
2. Look for profiles with credibility. Not the most enthusiastic junior. The middle manager the team respects. The senior analyst whose opinions carry weight. The champion must be credible to pragmatists and skeptics.
3. Prioritize department diversity. One champion per functional area: sales, finance, operations, HR, marketing. Adoption is contagious within departments, not between them.

How to develop them

• Intensive training: 40-60 hours of advanced training (Level 2-3 from the competency framework). Practical workshops with real cases from their department.
• Privileged access: tools, budget for experimentation, direct access to the technical team or the CAIO.
• Protected time: 4-8 hours per week dedicated exclusively to AI. If they do not have the time, they cannot be champions. This is an investment, not an expense.
• Internal community: monthly meetings among champions to share learnings, problems and solutions. Create a dedicated internal channel.

Their role

The champion does not implement technical solutions. Their role is to translate: they translate business needs to the technical team, and AI capabilities to the business team. They are the bridge between "this would be useful" and "this is possible". They are also the first line of support: when a colleague has an AI problem, they turn to the champion before IT.

Building AI culture without forcing it

AI culture is not decreed. It does not work with a CEO email saying "from now on we use AI". It works through habits, incentives and examples. These are the 6 principles that differentiate organic adoption from imposition:

1. Start with problems, not technology

Do not ask "how do we use AI?". Ask "what problem costs us the most time/money?". If the answer is "writing reports", AI is the obvious solution. If the answer is "lack of leadership", AI does not help. Starting from the problem avoids the "solution looking for a problem" syndrome.

2. Celebrate time savings, not the technology

When someone automates a report that took 3 hours and now takes 20 minutes, celebrate the 2 hours and 40 minutes freed up. Do not celebrate "how well they use ChatGPT". The message is: we value efficiency, not the tool.

3. Make the invisible visible

Create an internal channel where people share their "wins" with AI. A weekly "AI in action" post with real examples. A dashboard with adoption metrics. What is visible gets imitated.

4. Allow failure

AI makes mistakes. Prompts fail. Workflows break. If the culture penalizes error, nobody experiments. Create an environment where trying and failing with AI is acceptable. The only rule: do not use sensitive data in tests.

5. Lead by example

If the CEO uses AI and communicates it, the message is clear. If the CEO delegates AI to IT and never mentions it, the message is also clear. Executives should use AI visibly: in presentations, in meetings, in emails. Not because it is mandatory, but because it demonstrates that it works.

6. Do not mandate, but set expectations

Do not force anyone to use AI. But do establish expectations. "We expect each department to identify 3 processes where AI saves time in the next quarter." The difference between "you must use AI" and "we expect efficiency results" is subtle but critical.

Change management for executives

AI is a profound organizational change. It alters processes, roles, power dynamics and even the professional identity of individuals. Managing it poorly generates resistance, turnover and failure.

The 4 real fears

1. Fear of unemployment. "AI will take my job." Honest response: AI will eliminate some tasks, not entire positions. But it will change what is expected of each role. Communicate this with radical transparency.
2. Fear of incompetence. "I don't know how to use this and I'll look obsolete." Response: universal training without judgment. Nobody is born knowing prompting. Create an environment where asking questions is not a weakness.
3. Fear of loss of value. "If AI does my work in 10 minutes, my work is worth nothing." Response: your value is not the task, it is the judgment. AI generates a draft. You decide whether it is correct, relevant and useful.
4. Fear of surveillance. "They are monitoring me with AI." Response: clear privacy policy. AI measures process productivity, not individuals. Never use AI for individual surveillance.

The communication framework

Every significant change needs three communications:

• Before (30 days): what will change, why, what is expected of each person, what support they will receive.
• During (day 1-90): training, support, questions channel, monthly checkpoint to adjust.
• After (day 90+): measured results, public recognition, next-phase plan.

Each communication must answer the implicit question of every employee: "what does this mean for me, personally?"

Measuring adoption (what is not measured does not exist)

Without metrics, AI adoption is an anecdote. With metrics, it is a program with measurable returns. These are the 8 metrics every executive should track:

Activity metrics

• Adoption rate: percentage of employees using approved AI tools at least once per week. Target: >60% at 6 months.
• Usage frequency: average number of AI interactions per user per week. Target: >10 weekly interactions for power users.
• Usage diversity: number of distinct use cases per department. More diversity = deeper real integration.

Impact metrics

• Time saved: hours freed per employee per week thanks to AI. Measure with quarterly surveys and validate with productivity data.
• Quality: error rate before vs after AI in measurable processes (reports, analysis, customer responses).
• Satisfaction: internal NPS of AI tools. If the team hates the tool, adoption dies.

Maturity metrics

• Autonomy level: percentage of tasks where AI generates the final output without human review. This grows with trust.
• Innovation: number of new use cases proposed by employees (not IT). This indicates genuine AI culture.

AI competency framework

An AI competency framework defines what each person can do in relation to AI, and what they should be able to do. It serves three purposes: plan training, evaluate progress and make hiring decisions.

Framework structure

How to use it

1. Initial assessment: each employee self-assesses across the 5 competencies. Their manager validates. Result: a competency heat map by department.
2. Gap analysis: compare current level with target level for each role. The gap defines the training plan.
3. Individual plan: each person has a 6-month competency target. Champions help design the plan.
4. Periodic evaluation: every 6 months, reassess. Link progress to performance objectives (not to direct compensation, at least initially).

The framework is not rigid. Each company adapts it to its context. What matters is that it exists, that it is explicit and that it is reviewed.

Practical exercise

Puntos clave