The silent engagement crisis

According to the latest State of the Global Workplace report from Gallup (2025), only 23% of employees worldwide are "engaged" (actively committed to their work). 59% are "quiet quitting" (doing the bare minimum) and 18% are "actively disengaged" (demotivated and potentially sabotaging). In Europe, figures are worse: only 13% engagement.

The cost of disengagement is not abstract. Gallup estimates it at 8.8 trillion dollars annually worldwide (equivalent to 9% of global GDP). For a company of 500 employees with an average salary of 35,000 EUR, disengagement costs between 2 and 3 million EUR per year in lost productivity, absenteeism, turnover and errors.

The problem is not that companies do not care about engagement. It is that they measure it poorly: an annual survey of 80 questions nobody wants to fill out, whose results are analyzed 3 months later, and whose conclusions are filed away. By the time you react, the unhappy employee has already left.

AI changes this in three ways: it makes measurement more frequent and less invasive, it analyzes responses in real time, and it connects engagement data with concrete actions.

AI-powered climate surveys

The traditional climate survey has a design problem: it asks too much, too infrequently, and analyzes results too late. AI enables a complete redesign of this process.

Designing surveys with AI

• Adaptive questions. Instead of the same 80 questions for everyone, AI selects the most relevant questions for each employee based on their department, tenure and previous responses. A 10-question survey with well-chosen questions generates more insight than an 80-question generic one.
• Natural language. Traditional questions are formal and cold: "Rate your satisfaction with professional development opportunities on a scale of 1 to 5". AI can rephrase: "Do you feel you are growing professionally here?" Same information, better experience.
• Intelligent open questions. AI is not afraid of open responses because it can process them automatically. Instead of limiting yourself to numerical scales, you can include 2 or 3 open questions: "If you could change one thing about your day-to-day, what would it be?"
• Optimal frequency. AI can determine the ideal survey frequency per team. A team in crisis may need a weekly pulse. A stable team may work with a monthly pulse. Not everyone needs the same.

Prompt to design a climate survey

"Design a 12-question workplace climate survey for the [department] department. Context: [team size, current situation if relevant]. Include: 3 questions about the manager relationship, 3 about professional growth, 2 about workload, 2 about team culture, 2 open-ended. Format: question + response type (scale 1-5, yes/no, open). Approachable but professional language. Estimated response time: < 5 minutes."

Sentiment analysis in open responses

Open responses are the gold mine of climate surveys. Numbers tell you something is wrong. Open responses tell you what is wrong and why. The problem is that manually analyzing 500 open responses is unfeasible. AI does it in minutes.

What sentiment analysis can do

• Automatic classification. Each response is classified as positive, neutral or negative, with a confidence score. This gives you a quick view of the overall tone.
• Topic extraction. AI identifies recurring themes in the responses: "workload", "communication with manager", "lack of training", "obsolete tools". Without you defining them beforehand.
• Urgency detection. Responses indicating imminent flight risk or serious problems: "I am looking at other offers", "I cannot take this anymore", "I feel ignored". These responses need immediate attention.
• Temporal evolution. Compare this month's sentiment with the previous one. The trend is more important than the absolute value: a team that drops from 3.5 to 3.0 in 3 months needs attention, even though 3.0 does not seem like a crisis.

Prompt for sentiment analysis

"Analyze these [N] open responses from a workplace climate survey. For each response: 1) Sentiment (positive/neutral/negative, score 1-5). 2) Main topic. 3) Urgency (low/medium/high). After the individual analysis, generate: 1) Executive summary (3 paragraphs). 2) Top 5 topics by frequency. 3) Top 3 topics by urgency. 4) 3 recommended actions based on the data. Respect anonymity: do NOT cite literal responses that could identify a person."

Turnover prediction: detecting before it is too late

Most managers find out an employee wants to leave when they submit their resignation. By then, it is too late. The employee has made the decision, probably has another offer, and the chances of retaining them are minimal.

Predictive turnover signals

AI can monitor (with anonymized data and consent) a series of indicators that, combined, predict flight risk:

• Changes in engagement patterns: reduced participation in meetings, fewer contributions in communication channels, absence from social events.
• Survey results: significant drop in satisfaction between two consecutive surveys.
• HR data: time without promotion, salary gap with the market, lack of professional development.
• Market context: high demand for their profile, competitors hiring, growing sector.
• Manager data: recent manager change, reported conflict, negative performance review.

The prediction model

You do not need a sophisticated ML model. A simple rule-based scoring can work well:

• More than 2 years without promotion: +2 points
• Salary below the market 25th percentile: +3 points
• Drop of > 1 point in engagement survey: +2 points
• Manager change in the last 6 months: +1 point
• Profile with high market demand: +2 points

Score > 7: high risk. Immediate action (1:1 meeting, salary review, development plan).

Score 4-7: medium risk. Monitoring and prevention (follow-up survey, mentoring).

Score < 4: low risk. Maintenance (regular feedback, recognition).

Identifying engagement drivers

Not all factors affect engagement equally. AI can identify which drivers are most important for your specific company, not what generic studies say.

Universal drivers (according to research)

1. Relationship with the direct manager. It is the number 1 factor in most studies. The saying is true: people do not leave companies, they leave managers.
2. Professional growth. Opportunities to learn, take on new challenges, progress. Lack of development is the second cause of turnover after the manager.
3. Recognition. Not just monetary. A timely "good job", visibility of contributions, mention in team meetings.
4. Purpose. Understanding how your work contributes to the company's objective. Without purpose, work is just an economic transaction.
5. Autonomy. The ability to make decisions about how you do your work. Micromanagement kills engagement.

How AI identifies YOUR drivers

A simple regression analysis on your survey data can reveal which factor has the most impact on overall engagement in YOUR company. It may not be the manager (perhaps your managers are good). It may be workload (because you are in hypergrowth) or tools (because your internal software is terrible).

AI can do this analysis with 2 or 3 survey cycles and give you a ranking of drivers by impact. This lets you prioritize: instead of trying to improve everything at once, you focus on the 2 or 3 factors that move the needle most.

Pulse surveys: the continuous thermometer

A pulse survey is a short survey (3-5 questions) sent at high frequency (weekly or biweekly). It is the equivalent of taking temperature regularly instead of doing a complete medical checkup once a year.

Effective pulse survey design

• 3 to 5 questions. No more. The employee's commitment is to respond in under 2 minutes.
• 1 overall engagement question: "On a scale of 1 to 10, how much would you recommend this company as a place to work?" (eNPS).
• 1 or 2 focus questions: rotating, focused on the topic you want to monitor (manager, workload, development).
• 1 open question: "Anything you want to share?" Optional, no pressure.
• Automated sending: same day and time each week. AI can personalize the optimal moment based on each employee's response pattern.

Automated pulse analysis

AI processes results in real time and generates alerts when it detects:

• Significant drop in a specific team (> 0.5 points in one week)
• Sustained downward trend (3 consecutive weeks declining)
• Divergence between teams (one team at 8.5 and another at 5.0 in the same department)
• High-urgency open response

From data to action: evidence-based plans

Engagement data only has value if it translates into concrete actions. The biggest mistake companies make is not failing to measure. It is measuring and doing nothing with the results.

Action framework

1. Identify: what are the 2 or 3 most critical themes according to the data?
2. Diagnose: why are they low? Data alone does not reveal causes. Complement with focus groups, 1:1 interviews with managers.
3. Prioritize: what has the greatest impact and is fastest to implement? Use an impact/effort matrix.
4. Act: define concrete actions with an owner, deadline and success metric.
5. Communicate: inform employees that you heard their responses and what you are going to do. This alone already improves engagement.
6. Measure: in the next pulse, include a question about the implemented action. Did it work?

Prompt for action plan

"Based on these climate survey results: [data summary], generate an action plan for the next 90 days. For each critical theme: 1) Problem identified (1 sentence). 2) Probable cause. 3) Proposed action (concrete, measurable). 4) Suggested owner (HR, manager, leadership). 5) Deadline. 6) Success metric. 7) Communication to employees (what you tell them and when). Maximum 3 actions. Better 3 executed actions than 10 planned ones."

Privacy and ethics in engagement measurement

Measuring engagement with AI opens powerful doors. It also opens significant ethical risks. These are the red lines:

• Real anonymity. It is not anonymous if the manager can identify the only remote employee on their 3-person team. Rule: do not report results for groups smaller than 5 people.
• Informed consent. Employees must know what data is collected, how it is analyzed, and for what purpose. GDPR requires it. Trust does too.
• Do not monitor communications. Analyzing email or chat sentiment without consent is a privacy violation. Even with consent, it is ethically questionable.
• Do not penalize honesty. If an employee says their manager is terrible, the correct response is to investigate the manager. Not the employee.
• Transparency in algorithms. If you use turnover prediction, employees should know it exists (not the model details, but its existence and purpose).

Recommended tools

• Culture Amp: the reference in engagement analytics. Surveys, pulse surveys, AI analytics, industry benchmarks. From 5 EUR/employee/month.
• Peakon (Workday): integrated into the Workday ecosystem. Sentiment analysis, turnover prediction, action planning. For companies already using Workday.
• Officevibe (Workleap): simple pulse surveys, anonymous feedback, recognition. Good value for SMBs. From 3.5 EUR/employee/month.
• LLM + Google Forms: the DIY option. Google Forms to collect responses, Claude or ChatGPT to analyze results. Cost: just the LLM subscription. Works surprisingly well for small teams.

Ejercicio practico

Puntos clave