The cost of bad onboarding

A data point that should concern any HR manager: 20% of employee turnover occurs in the first 45 days. According to Gallup, only 12% of employees say their company does a good onboarding. The remaining 88% start their working life in the organization with a mediocre or bad experience.

The cost of replacing an employee who leaves in the first 6 months ranges between 50% and 200% of their annual salary, depending on the role. For a mid-level profile with a 40,000 EUR salary, we are talking about between 20,000 and 80,000 EUR per premature departure.

Traditional onboarding has three structural problems that AI can solve:

1. Information overload. On the first day, the new employee receives 47 documents, 12 access credentials, 8 presentations and is expected to remember everything. It is impossible.
2. People dependency. "Ask Maria, she will explain." But Maria is on vacation, in a meeting, or simply does not have time.
3. Lack of personalization. The same onboarding for a senior developer as for a junior salesperson. Same talks, same videos, same forms.

AI does not solve everything. But it can solve these three problems significantly.

Automatic documentation generation

Every new hire needs documentation: contract, confidentiality agreement, data protection policy, employee handbook, expense policy, internal regulations. In many companies, HR generates these documents manually for each person, changing names, dates and specific details.

How AI automates documentation

• Document generation from templates. AI automatically fills templates with the new employee's data (name, position, department, start date, salary, benefits). What took 30 minutes per employee is done in seconds.
• Profile-based adaptation. A part-time employee receives different clauses than a full-time one. A temporary contract is different from a permanent one. AI selects the correct clauses automatically.
• Automatic translation. For companies with international employees, AI can generate documentation in the employee's language (while respecting the legal version in the official language).
• Executive policy summaries. Instead of forcing the new employee to read 80 pages of internal regulations, AI generates a 2-page summary with key points, organized by relevance to their role.

Prompt to generate a personalized welcome kit

"Generate a welcome kit for [name], joining as [position] in the [department] department on [date]. Include: 1) Personalized welcome message (5 lines, mentioning the team and a relevant project). 2) First day agenda, hour by hour. 3) List of access they will need (email, tools, VPN) with the person responsible for each. 4) 5 key people they should meet in their first week (name, position, why they are important for their role). 5) Resumen of the 3 most relevant internal policies for their role (3 bullets each). 6) FAQ: 10 frequently asked questions from the first days with answers."

Personalized training plans with AI

Generic onboarding does not work. A backend developer needs to learn the microservices architecture, code repositories, style guides. A salesperson needs to learn the product, sales scripts, the CRM. Giving the developer the salesperson's training (and vice versa) wastes both their time.

How AI personalizes training

• Knowledge gap analysis. From the new employee's CV and the role requirements, AI identifies what they know and what they need to learn. A senior with 10 years of experience does not need the same training as a junior.
• Adaptive learning path. AI generates a sequence of training modules ordered by priority and dependency. Content critical to start working goes first. Context content goes later.
• Microlearning. Instead of 4-hour sessions, AI breaks down training into 15-minute capsules. Each capsule covers one concept, includes a practical exercise, and can be consumed at the most convenient time.
• Adaptive quizzes. After each capsule, a 5-question quiz. If the employee masters the topic, they move on. If they have gaps, AI deepens that area.

Prompt for training plan

"Generate a training plan for the first 30 days of [name], [position] in [department]. Their prior experience: [summary]. Tools they already know: [list]. Tools they need to learn: [list]. Structure the plan in 4 weeks. Each week: 3-4 modules. Each module: title, objective (1 sentence), estimated duration (15-60 min), type (reading, video, practice, shadowing), and completion criterion. Prioritize: week 1 = minimum to start contributing. Week 2 = deepen tools. Week 3 = business context. Week 4 = supervised real project."

Internal chatbot for new employees

The most frequent question from a new employee is not about corporate strategy or their career plan. It is: "Where is the cafeteria?" or "How do I request time off?" or "What is the WiFi password?". Simple questions that interrupt colleagues and managers 15 times a day.

What an onboarding chatbot solves

• 24/7 instant FAQ. The 100 most frequent questions from new employees, with updated answers and links to relevant documents. Without waiting for someone to be available.
• Process navigation. "How do I request an advance?" The chatbot not only answers, but guides step by step: "Go to [system], menu [X], option [Y], fill in these fields".
• Intelligent escalation. If the question is not in the knowledge base, the chatbot redirects it to the right person (HR for employment matters, IT for access, manager for team matters).
• Continuous feedback. The chatbot can proactively ask: "You have been here 2 weeks. Do you have all the access you need?" or "Have you had your first 1:1 meeting with your manager?". If the answer is no, it automatically escalates.

Practical implementation

You do not need to build a chatbot from scratch. Options ordered from least to most complex:

1. Slack/Teams channel with AI: a dedicated channel for new employees where a bot answers questions based on internal documentation. Cost: minimal if you already use Slack/Teams.
2. Knowledge base with AI search: Notion AI, Guru, or similar. You upload all onboarding documentation and employees search with natural language.
3. Custom chatbot: connected to your internal knowledge base via RAG (Retrieval Augmented Generation). Higher initial investment, but the experience is superior.

Algorithmic buddy matching

The buddy program (a companion assigned to the new employee) is one of the onboarding interventions with the greatest impact. According to Microsoft (which has a well-studied buddy program), employees with a buddy are 23% more productive after 90 days than those without one.

The problem with manual matching

In most companies, the buddy is assigned by proximity (the person sitting next to them) or availability (whoever is not too busy). This generates suboptimal matches: a buddy from a completely different department, or a buddy who does not have time or motivation.

How AI improves matching

• Profile compatibility. AI analyzes the new employee's profile (position, experience, interests, personality if assessed) and crosses it with available buddies to find the best match.
• Matching criteria: same or related department, complementary experience (not identical), real availability, buddy track record (good buddies repeat).
• Load balancing. AI ensures no employee is a buddy to more than 2 people simultaneously. Buddy fatigue is real and reduces experience quality.
• Automated follow-up. AI schedules check-ins between buddy and new employee, sends reminders, and collects feedback from both to improve future matches.

Prompt for buddy matching

"I have a new employee: [name], [position], [department], prior experience in [sector], professional interests: [list]. These are the available buddies: [list with name, position, department, years at company, interests]. Suggest the best match and the next 2, explaining why. Criteria: 1) Related department. 2) Complementary experience. 3) At least 1 common interest. 4) Minimum 1 year at the company."

The 30-60-90 day plan with AI

The 30-60-90 plan is the most widely used framework for structuring a new employee's first 3 months. It defines clear objectives for each phase and allows the employee and their manager to evaluate progress objectively.

Plan structure

• First 30 days: learn. Get to know the company, team, tools, processes. Objective: understand how everything works. Deliverable: written summary of what was learned.
• Days 31-60: contribute. Start participating in real projects with supervision. Objective: add value with support. Deliverable: first tangible contribution.
• Days 61-90: lead. Take autonomous responsibility for a task or project. Objective: operate independently. Deliverable: measurable result in their area.

How AI personalizes the 30-60-90 plan

AI generates the plan from three inputs: the role profile (requirements, KPIs), the employee profile (experience, gaps), and the team context (active projects, priorities). The result is a plan with SMART objectives for each phase, associated learning resources, and evaluation criteria.

Prompt to generate 30-60-90 plan

"Generate a 30-60-90 day plan for [name], [position] in [department]. Prior experience: [summary]. Role KPIs: [list]. Current team projects: [list]. For each phase (30, 60, 90 days): 3 SMART objectives, 2 key actions, 1 deliverable, evaluation criterion. Include a check-in meeting with the manager at the end of each phase with a suggested agenda."

Administrative task automation

Onboarding has a relational part (welcome, culture, team) and an administrative part (access, signatures, mandatory training). The administrative part is tedious, repetitive and error-prone. It is perfect for automation.

Tasks that can be automated with AI and workflows

• Access requests: corporate email, tools (Slack, Jira, CRM), VPN, physical access. An automated workflow creates requests and notifies IT on the start date.
• Document signing: contract, NDA, data policy, health and safety. Integrated electronic signature (DocuSign, SignaturIT) that triggers automatically.
• System registration: payroll, social security, benefits. AI extracts data from the contract and enters it into the corresponding systems.
• Mandatory training: health and safety, GDPR, code of ethics. Automatically assigned with reminders and completion tracking.
• Physical welcome kit: laptop, access card, office supplies. Automatically requested from the relevant department with the necessary lead time.

Effective onboarding metrics

1. Time-to-productivity: days until the employee reaches 80% of their expected performance. Target: < 90 days for qualified roles.
2. 90-day retention: percentage of new employees who remain after 3 months. Target: > 90%.
3. 12-month retention: percentage remaining after one year. Target: > 80%.
4. Onboarding eNPS: survey at the end of the first 90 days. "Would you recommend a friend to join this company based on your onboarding experience?" Target: > 40.
5. Plan completeness: percentage of 30-60-90 objectives achieved. Target: > 70%.
6. Manager satisfaction: manager's evaluation at the end of 90 days. "Is the employee where you expected them to be at this point?" Scale 1-5. Target: > 3.5.

Common onboarding mistakes with AI

1. Dehumanizing the onboarding. AI handles the administrative part. The human welcome, the coffee with the team, the first lunch together, that is not automated. If the new employee only interacts with bots on their first day, something is wrong.
2. Overloading the first day. AI can generate 50 training modules in a minute. That does not mean the employee should receive them all on day one. Distribute information across weeks, not hours.
3. Not measuring. If you do not know how long it takes an employee to be productive, you do not know if your onboarding works. Measure before automating to have a baseline.
4. Ignoring feedback. The chatbot receives questions it cannot answer. Those questions are gold: they tell you what is missing from your documentation. Review them weekly.
5. One-size-fits-all. A remote worker's onboarding is different from an on-site one's. A senior's is different from a junior's. If you do not personalize, you waste AI's potential.

Ejercicio practico

Puntos clave