Key Takeaways

The points worth keeping

• The real customer service landscape in Brazil in 2026
• The customer service pyramid with AI
• The numbers that justify the investment

What goes here: Complaints, exchanges with exceptions, technical questions, negotiations, unhappy customers who need human empathy.

Goal: 30-40% of tickets. Human agent with AI support.

How it works: AI triages, classifies the ticket, suggests a response, pulls up customer history and a summary of previous interactions. The human agent reviews, adjusts the tone, and sends.

Real gain: An agent who used to spend 8 minutes per ticket now spends 3 minutes. AI did 60% of the work. The human added context and empathy.

Tier 3 -- Human-Only Service

What goes here: Crises, VIP clients, legal situations, complaints filed with consumer protection agencies/review platforms, sensitive escalations, high-value negotiations.

Goal: 10-25% of tickets. Full attention from the best available agent.

How it works: AI identifies that the case is critical (very negative sentiment, high-value customer, legal keywords) and routes it directly to a senior agent, with all context already prepared.

Golden rule: AI never responds alone in these cases. It prepares, organizes, and informs. The human decides and acts.

The Numbers That Justify the Investment

If you need to convince your boss (or yourself), here's the breakdown:

Scenario: company with 500 tickets/month

And that's not even counting the gains in satisfaction, retention, and indirect revenue. A well-served customer buys more, refers more, and complains less.

8 Use Cases with Implementation Guide

Here's what matters: practical implementation. For each use case, you'll find the prompt template, recommended tool, and before/after metrics.

1. Smart Automated Responses (WhatsApp + Email)

The problem: Customers message outside business hours and get no response. Or they receive a generic reply that doesn't solve anything.

The AI solution: Contextual responses based on customer history and your company knowledge base.

Prompt template to configure the chatbot:

You are the virtual assistant for [COMPANY NAME], specializing in [INDUSTRY].

Rules:
1. ALWAYS respond in Portuguese (Brazil), friendly and professional tone.
2. Use the customer's name when available.
3. For order status questions, check the system and provide order number, current status, and estimated delivery.
4. For product/service questions, use ONLY information from the knowledge base below.
5. If you don't know the answer with certainty, say: "I'll forward this to a specialist who will help you within [TIME]. Can I help you with anything else?"
6. NEVER make up information about prices, timelines, or policies.
7. For complaints with negative tone, respond with empathy first, then offer a solution.

Knowledge base:
[INSERT YOUR FAQ, POLICIES, PRODUCT INFORMATION HERE]

Human support hours: [YOUR HOURS]

Recommended tools:

Before/after metrics:

2. Ticket Classification and Routing

The problem: All tickets land in the same place. The manager spends time reading each one to decide who should handle it. Urgent tickets get buried under simple questions.

The AI solution: Automated classification by category, urgency, and sentiment. Real-time routing to the correct agent or queue.

Prompt template:

Analyze the support ticket below and classify it:

TICKET: "[CUSTOMER MESSAGE]"

Return in JSON format:
{
  "category": "[financial | technical | sales | logistics | complaint | compliment | other]",
  "urgency": "[critical | high | medium | low]",
  "sentiment": "[very_negative | negative | neutral | positive | very_positive]",
  "requires_human": true/false,
  "justification": "[1 sentence explaining the classification]",
  "suggested_response": "[suggested response for the agent]",
  "recommended_queue": "[queue name or agent]"
}

Urgency rules:
- CRITICAL: mentions to consumer protection, lawyer, immediate cancellation, risk of VIP customer churn
- HIGH: customer waiting over 48h, financial issue, defective product
- MEDIUM: product questions, standard exchanges, common requests
- LOW: compliments, feedback, informational questions

Recommended tool: ChatGPT API (GPT-4o mini) via Make/Zapier. Cost: ~$0.001 per ticket classified.

Before/after metrics:

3. Self-Updating Knowledge Base

The problem: Your FAQ is outdated. Agents know the right answer, but the official document still has information from 6 months ago. New agents learn the wrong information.

The AI solution: AI analyzes support conversations and identifies when agent responses diverge from the official FAQ. Automatically suggests updates.

Prompt template (run weekly):

Analyze the support conversations from the last week and compare with the current knowledge base.

CURRENT KNOWLEDGE BASE:
[INSERT YOUR CURRENT FAQ]

WEEK'S CONVERSATIONS (summary):
[INSERT SUMMARIES OR EXPORT FROM YOUR SYSTEM]

Identify:
1. Frequently asked questions that are NOT in the FAQ (new questions)
2. FAQ answers that agents are correcting in conversations (outdated information)
3. Topics where agents give different answers to each other (inconsistency)

For each item, suggest:
- Exact question for the FAQ
- Updated and correct answer
- Source (which conversation or agent provided the information)

Recommended tool: Claude or GPT-4o + Google Sheets for tracking.

Before/after metrics:

4. Real-Time Sentiment Analysis

The problem: You only find out a customer is furious after they've already posted on the consumer complaint site. There's no early warning system.

The AI solution: Every message is analyzed in real time. If sentiment drops below a threshold, the system immediately alerts the supervisor.

Prompt template (for each incoming message):

Analyze the sentiment of this customer message:

MESSAGE: "[CUSTOMER TEXT]"
CONTEXT: [customer for X months | has made Y purchases | last ticket was about Z]

Return:
- sentiment: [1-10, where 1 = furious and 10 = delighted]
- primary_emotion: [frustrated | irritated | confused | anxious | neutral | satisfied | grateful]
- churn_risk: [high | medium | low]
- supervisor_alert: true/false
- recommended_response_tone: [1 sentence about how the agent should respond]

If sentiment <= 3 OR churn_risk = high, set supervisor_alert = true.

Practical implementation:

1. Integrate via API into your ticket system (Freshdesk, Zendesk, or even a spreadsheet)
2. Configure webhook to alert on supervisor's Slack/Teams/WhatsApp when supervisor_alert = true
3. Create weekly dashboard with average sentiment by channel, agent, and ticket type

Before/after metrics:

5. Conversation Summaries for Agents

The problem: The customer already talked to 3 different agents and has to repeat everything. Or the agent opens a ticket with 47 messages and doesn't know where to start.

The AI solution: Automatic summary of the entire previous conversation in 3-5 lines. The agent starts already knowing exactly what happened.

Prompt template:

Summarize the support conversation below for the next agent who will take over.

CONVERSATION:
[FULL HISTORY]

Summary format:
1. ORIGINAL PROBLEM: [1 sentence]
2. WHAT'S BEEN DONE: [list of actions taken]
3. CURRENT STATUS: [waiting on X / awaiting Y / partially resolved]
4. CUSTOMER SENTIMENT: [how the customer is feeling]
5. RECOMMENDED NEXT ACTION: [what the agent should do now]

Rules:
- Maximum 5 lines
- Include order numbers, ticket IDs, and dates mentioned
- Highlight any promises made to the customer (timeline, discount, exchange)

Recommended tool: Integration with your ticket system via API. Most modern helpdesks (Freshdesk, Intercom, Zendesk) already have native AI plugins that do this.

Before/after metrics:

6. Automatic Post-Support Follow-Up

The problem: The ticket was resolved, but nobody asked if the customer was satisfied. You're missing the chance to recover dissatisfied customers and collect testimonials from happy ones.

The AI solution: Automated message 24-48 hours after closure, personalized based on the support context.

Prompt template (generate follow-up message):

Generate a follow-up message for the customer after support.

DATA:
- Customer name: [NAME]
- Problem type: [CATEGORY]
- Resolution applied: [WHAT WAS DONE]
- Final conversation sentiment: [POSITIVE/NEUTRAL/NEGATIVE]
- Preferred channel: [WHATSAPP/EMAIL]

RULES:
- If sentiment is POSITIVE: thank them and ask for a review/testimonial
- If sentiment is NEUTRAL: ask if it was actually resolved and offer additional help
- If sentiment is NEGATIVE: show genuine concern, offer direct line to supervisor
- Tone: human, friendly, brief (maximum 3 paragraphs)
- Include review link if sentiment is positive
- NEVER send follow-up for tickets involving legal matters or consumer protection

Output example (positive sentiment, WhatsApp):

Hi Maria! This is [Company]. Just checking in to see if your product exchange arrived okay? We're so happy we could resolve it quickly for you!

If you have a moment, we'd love a quick review here: [link]. It really helps other customers like you.

If you need anything else, just reach out! Have a great day!

Before/after metrics:

7. Dynamic AI-Powered FAQ

The problem: Your FAQ is a static page with 20 questions nobody reads. When a customer has a specific question, they can't find it and message support instead.

The AI solution: Conversational FAQ. The customer types their question in natural language and gets a precise answer extracted from your knowledge base.

Prompt template (for the conversational FAQ):

You are the [COMPANY] FAQ assistant.

KNOWLEDGE BASE:
[YOUR DOCUMENTS, POLICIES, PRODUCT INFORMATION]

RULES:
1. Only answer based on the information above. If the answer isn't in the base, say "This information isn't in my database. Let me connect you with an agent."
2. Cite the source: "According to our exchange policy..."
3. If the question has multiple interpretations, ask for clarification.
4. Offer links to relevant pages when they exist.
5. At the end, ask: "Did this answer your question? Can I help with anything else?"
6. If the customer says it did NOT help, route to human support immediately.

Implementation with Typebot (free):

1. Create account on Typebot (typebot.io)
2. Connect with ChatGPT or Claude API
3. Feed your knowledge base with your documents
4. Insert the widget on your website
5. Monitor which questions the AI can't answer and update the base

Before/after metrics:

8. AI-Powered Customer Onboarding

The problem: The customer purchased, received a generic welcome email, and got lost. They don't know how to use the product, didn't set anything up, and canceled within 30 days.

The AI solution: Personalized onboarding sequence, with messages adapted to the customer's profile and where they are in the process.

Prompt template:

Create a 5-message onboarding sequence for a new customer.

CUSTOMER PROFILE:
- Name: [NAME]
- Plan purchased: [PLAN]
- Industry: [CUSTOMER'S INDUSTRY]
- Main objective: [WHAT THEY WANT TO ACHIEVE]

PRODUCT:
- Name: [YOUR PRODUCT]
- Key features: [LIST]
- Initial setup steps: [LIST]

RULES:
- Message 1 (day 0): Welcome + simplest first step
- Message 2 (day 2): Second step + tip relevant to customer's industry
- Message 3 (day 5): Check if they set up + offer help
- Message 4 (day 10): Advanced feature that creates most value
- Message 5 (day 15): Ask for feedback + offer call if needed

Tone: enthusiastic but not artificial. Like a friend who knows their stuff.
Channel: [WHATSAPP/EMAIL]
Size: maximum 4 sentences per message (WhatsApp) or 2 paragraphs (email)

Before/after metrics:

Channel strategy: where and how to use AI

Each channel has its own rules. Copying and pasting the same strategy across all of them won't work.

WhatsApp

The most important channel for customer service in Brazil. Over 97% of companies use it.

Critical tip: On WhatsApp, customers expect conversation, not forms. Configure your chatbot to sound like a natural chat, not a robotic menu of options.

Email

Still relevant for B2B, complex tickets, and documentation.

Instagram DM

Growing volume, especially for e-commerce and local services.

Phone

Still necessary for certain audiences and critical situations.

Website chat

First point of contact for many visitors.

How to build a functional chatbot in 1 weekend

You don't need a programmer. You don't need 3 months of project work. Here's the plan to go from zero to a working chatbot in 2 days.

Saturday morning: Preparation (3 hours)

Hours 1-2: Build your knowledge base

Open a document and answer these questions:

1. What are the 20 questions your customers ask most often?
2. What is the correct, up-to-date answer for each one?
3. What information should the chatbot NEVER provide? (prices that change, varying deadlines)
4. In which situations should the chatbot transfer to a human?
5. What is your company's voice? (formal, casual, technical)

Hours 2-3: Choose your tool

For most businesses, I recommend Typebot:

• Free and open source
• Visual interface (drag and drop)
• Integrates with ChatGPT, Claude, WhatsApp, website
• No conversation limits

Saturday afternoon: Build (4 hours)

Hours 3-5: Set up the basic flow

1. Create an account on Typebot (typebot.io)
2. Connect the ChatGPT API (GPT-4o mini -- costs just cents)
3. Create the "Welcome" block with an introduction
4. Add the AI block with your prompt and knowledge base
5. Set up the "Transfer to human" block as a fallback

Hours 5-7: Test and adjust

1. Ask 20 real questions (use ones your actual customers ask)
2. For each wrong answer, adjust the prompt or knowledge base
3. Test the human transfer scenarios
4. Test outside business hours

Sunday: Launch (3 hours)

Hours 7-8: Integrate with your channel

• For website: copy the widget code and paste it into your HTML
• For WhatsApp: connect via the WhatsApp Business API (Typebot has native integration)
• For Instagram: use Manychat as a bridge

Hours 8-9: Train your team

• Show them how the chatbot works
• Explain when and how they'll receive transfers
• Define who monitors chatbot conversations during the first week

Hours 9-10: Monitor initial results

• Monitor the first conversations in real time
• Adjust responses that didn't work well
• Document what needs improvement

Want ready-to-use prompts for setting up customer service chatbots? Check out the prompt library at TakeAICourse.com.

Quality Metrics Dashboard

You can't improve what you don't measure. Here are the 7 essential metrics for AI-powered customer service:

The 7 Metrics That Matter

How to Calculate Each Metric

CSAT: After each interaction, ask "On a scale of 1 to 10, how would you rate this interaction?" Average all responses.

NPS: Ask "On a scale of 0 to 10, how likely are you to recommend our company?" Promoters (9-10) - Detractors (0-6) = NPS.

FCR: (Tickets resolved on first contact / Total tickets) x 100

AHT: Sum of all interaction durations / Number of interactions

FRT: Sum of time to first response across all tickets / Number of tickets

SRR: (Tickets resolved by AI without human / Total tickets) x 100

ESR: (Tickets where AI transferred to human / Total tickets started with AI) x 100

Prompt for Generating Weekly Reports

Analyze this week's support data and generate an executive report.

DATA:
- Total tickets: [NUMBER]
- Average CSAT: [NUMBER]
- FCR: [PERCENTAGE]
- AHT: [MINUTES]
- FRT: [MINUTES]
- AI self-resolution rate: [PERCENTAGE]
- Escalation rate: [PERCENTAGE]
- Top 5 ticket categories: [LIST]
- Top 3 complaints: [LIST]

GENERATE:
1. Executive summary (3 sentences)
2. What improved vs last week
3. What worsened vs last week
4. 3 priority actions for next week
5. Forecast: if the trend continues, where will we be in 30 days?

The Human Touch: When NOT to Use AI

AI isn't the answer for everything. There are moments when human presence is irreplaceable. Knowing when to turn off AI is just as important as knowing when to turn it on.

Situations Where AI Must Step Aside Immediately

1. Crises and emergencies: Customer reports a health issue related to the product, an accident, or a dangerous situation. Human. Now.

2. Legal threats: Mention of lawyer, lawsuit, consumer protection agency, or complaint platform with aggressive tone. Human supervisor takes over.

3. Grief or sensitive situations: Customer canceling service due to a family member's death. AI lacks genuine empathy for this.

4. High-value negotiations: Contracts over $10,000. The customer wants to speak with a real person, not a robot.

5. Dissatisfied VIP customer: Your best customer is frustrated. They deserve your best agent, not an automated response.

6. Emotional ambiguity: The customer is confused, can't explain the problem, or is emotionally distressed. AI makes this worse.

7. Strategic feedback: Customer offering valuable product suggestions. A human should listen, take notes, and genuinely thank them.

The Golden Rule of AI-Powered Service

When in doubt, err on the side of humanity. It's cheaper to have an agent handle a case AI could have resolved than to have AI ruin a case that needed a human.

Configure your AI with this mindset. It's better to transfer 5% more to humans than to lose 1 customer due to a robotic response at a sensitive moment.

Case studies: 4 segments, 4 implementations

Case 1: Fashion e-commerce (SME)

Scenario: Online store with 2,000 orders/month, 3 support agents, 400 tickets/month.

Implementation:

• WhatsApp chatbot for order tracking and exchanges (Tier 1)
• AI classifies tickets by urgency (Tier 2)
• Agents use AI summaries for repetitive tickets

Results after 60 days:

Case 2: B2B SaaS (startup)

Scenario: Restaurant management software, 180 clients, 1 support person, 220 tickets/month.

Implementation:

• Conversational FAQ on website (reduces simple tickets)
• Knowledge base that auto-updates from conversations
• Automated onboarding for new clients

Results after 90 days:

Case 3: Dental clinic

Scenario: 3 dentists, 1 receptionist who also handles support, 150 contacts/month via WhatsApp.

Implementation:

• Chatbot for appointment scheduling and confirmations
• Automated responses about accepted insurance plans, evaluation prices, and hours
• Post-visit follow-up requesting Google reviews

Results after 45 days:

Case 4: Restaurant (2 locations)

Scenario: 2 locations, delivery + dine-in, WhatsApp flooded with repetitive questions.

Implementation:

• Chatbot for menu, hours, reservations, and delivery status
• Complaint classification (food, delay, billing)
• Automated follow-up after delivery requesting reviews

Results after 30 days:

7 mistakes that ruin your AI customer service

Implementing AI wrong is worse than not implementing it at all. Here are the 7 most common mistakes and how to avoid each one.

Mistake 1: AI that can't say "I don't know"

What happens: The chatbot makes up answers. The customer receives wrong information about deadlines, prices, or policies. A problem turns into a crisis.

How to avoid: In the prompt, explicitly include: "If you're not sure of the answer, say 'I'll check with my team and get back to you in [TIME]'. NEVER make up information."

Mistake 2: Ignoring brand voice

What happens: A casual, laid-back company, but the chatbot responds like a technical manual. Or a serious B2B company with a chatbot using slang and emojis.

How to avoid: Include 3-5 examples of real responses from your company in the prompt. The AI learns the tone from examples, not abstract instructions.

Mistake 3: Zero human fallback

What happens: A customer is stuck in a loop with the chatbot, asking to speak to a human, and the AI doesn't offer the option. The customer becomes furious.

How to avoid: Configure trigger words for immediate transfer: "I want to talk to a person", "agent", "human", "supervisor", "I don't want a bot". And add a visible option to "Talk to an agent" in every interaction.

Mistake 4: Outdated knowledge base

What happens: The chatbot gives a 6-month-old price, old hours, or a policy that has changed. The customer complains and loses trust.

How to avoid: Schedule weekly knowledge base reviews. Use the "self-updating base" prompt (use case #3 above) to detect discrepancies.

Mistake 5: Only measuring satisfaction when service goes well

What happens: You only request CSAT when the chatbot resolves the issue. In cases that went wrong, you don't collect feedback. Your numbers look great, but the reality isn't.

How to avoid: Request evaluations for ALL interactions -- especially those that were escalated or had negative sentiment. That's where the real learning is.

Mistake 6: Automating too much, humanizing too little

What happens: The entire journey is robotic. The customer feels like they're talking to a cold machine. Even when the problem is solved, the experience is poor.

How to avoid: Add strategic "humanization touchpoints": use the customer's name, reference past purchases, show that you understood their sentiment. And always keep the human option one click away.

Mistake 7: Not training your team to work with AI

What happens: You implement the tools, but the agents don't know how to use AI suggestions, don't trust the classifications, and end up ignoring everything.

How to avoid: See the training section below.

Team training: how to get your team to embrace AI

The most sophisticated tool in the world is useless if your team won't use it. Here's the 4-step training plan.

Step 1: Demystification (Day 1)

Goal: Eliminate fear and resistance.

What to do:

• Show that AI will NOT replace anyone. It will take care of the boring work.
• Demonstrate live: show the chatbot answering simple questions that agents hate answering.
• Ask: "What repetitive task do you hate most? AI will handle that for you."
• Show the numbers: "You'll handle 50% fewer tickets, but they'll be more interesting."

Step 2: Hands-on training (Days 2-3)

What to do:

• Each agent sets up and tests the chatbot in a training environment.
• Practice how to read and use AI-generated summaries.
• Role-play: one agent plays the customer, another uses the AI tools.
• Document questions and adjust prompts based on team feedback.

Step 3: Assisted operation (Weeks 1-2)

What to do:

• AI is active, but a supervisor monitors all interactions.
• Daily 15-minute meeting: "What did AI get wrong today? What did it get right?"
• Adjust prompts and knowledge base daily.
• Agents have freedom to correct AI in real time.

Step 4: Autonomous operation (Week 3+)

What to do:

• AI operates independently for Tier 1.
• Weekly review meeting (30 minutes): metrics, adjustments, new questions.
• Agents focus on Tier 2 and 3.
• Celebrate results: show the improvement numbers to the team.

Prompt for creating training materials

Create a training guide for customer service agents who will work with AI.

CONTEXT:
- Company: [NAME]
- AI tools implemented: [LIST]
- Team's technical level: [BASIC/INTERMEDIATE]
- Team's biggest concern: [EX: "fear of losing job", "not trusting AI"]

GENERATE:
1. Opening slide (1 page): "Why we're implementing AI"
2. Team FAQ (10 questions with honest answers)
3. Step-by-step guide for using each tool (with simulated screenshots)
4. Practice scenarios (5 situations for the team to simulate)
5. Daily checklist for the first 2 weeks

LGPD compliance in AI-powered customer service

Using AI with customer data requires careful attention to LGPD (Brazil's General Data Protection Law). It's not optional.

What you MUST do

AI notice template for customer service

Hello! Welcome to [COMPANY] customer service.

To speed up your service, we use artificial intelligence on this channel.
Your data is protected according to LGPD and our Privacy Policy.

You can request human assistance at any time by typing "agent".

How can I help you?

What you CANNOT do

• Make automated decisions that significantly affect the customer (e.g., deny credit, cancel account) without offering a human review option.
• Collect sensitive data (health, orientation, religion) via chatbot without explicit consent.
• Use service data for marketing without separate consent.
• Share transcripts with third parties without a legal basis.

To dive deeper into applied AI with compliance, check out the AI courses at TakeAICourse.com—all cover best practices for responsible use.

Implementation Plan: From Theory to Practice

If you've made it this far, you already know WHAT to do. Now you need to know in WHAT ORDER.

Week 1: Foundation

• Gather the 20 most common customer questions
• Document correct, up-to-date responses
• Choose a chatbot tool (recommendation: Typebot)
• Set up your account and explore the interface

Week 2: Building

• Configure chatbot with knowledge base
• Test with 30 real questions
• Adjust prompt and knowledge base based on errors
• Integrate with WhatsApp or website

Week 3: Launch

• Go live for customers (with intensive monitoring)
• Daily 15-minute team sync
• Collect CSAT from first interactions
• Make daily adjustments

Week 4: Optimization

• Implement ticket classification (use case #2)
• Set up sentiment analysis (use case #4)
• First weekly metrics report
• Plan phase 2 (use cases 5-8)

Month 2: Expansion

• Implement conversation summaries for agents
• Set up post-service follow-up
• Activate conversational FAQ on website
• Start automated onboarding

Month 3: Maturity

• Complete, automated metrics dashboard
• Knowledge base self-updates weekly
• Team 100% adapted to AI tools
• Review and adjust CSAT, FCR, and AHT goals

Real Costs: What You're Actually Going to Spend

Full transparency. Here's what each tool actually costs.

*API cost depends on volume. For 500 tickets/month, expect $10-30 with GPT-4o mini.

Total cost for an SMB: $20-100/month for a complete, functional setup.

Compare that to the cost of one support agent ($1,500-2,500/month plus benefits), and the math works out easily.

What's Next: Predictive AI in Customer Service

Everything we've discussed so far is reactive AI: the customer reaches out, the AI responds. The next level is predictive AI: identifying problems BEFORE the customer complains.

Examples of predictive AI in customer service:

• Detect that an order will be delayed and proactively notify the customer (before they even ask)
• Identify usage patterns that indicate the customer will cancel in the next 2 weeks
• Suggest upgrades when the customer is using 90%+ of their plan capacity
• Schedule preventive maintenance before the product breaks down

This is already possible with 2025's tools, but requires structured data and a higher level of integration. It's the natural next step after you've mastered the 8 use cases in this guide.

Start Today, Not Tomorrow

You don't need to implement everything at once. Start with use case #1 (automated responses) and use case #7 (dynamic FAQ). Those two alone can reduce 30-40% of your ticket volume.

Each week, add a new use case. In 60 days, you'll have a customer service system that works better, costs less, and keeps your customers happier.

The best time to start was yesterday. The second best time is right now.

Next Steps

If you want to implement AI in customer service with structured guidance, TakeAICourse.com has everything you need:

• Browse our AI courses -- with hands-on lessons and ready-to-use templates
• Explore the prompt library -- hundreds of prompts organized by area, including customer service
• Check out the guide on AI for your profession -- content segmented by industry
• Create your free account and start transforming your customer service today

AI isn't going to replace human customer service. But customer service professionals who use AI will replace those who don't.