- Chat, voice & email
- AI disclosure built in
- Human handover
AI & automation · Web chat, voice and email
AI chatbot development: agents that answer, qualify and book
We build AI agents for your website, phone line and inbox that answer from approved information, qualify enquiries, book appointments and hand over to a person when it matters. They say they are an AI from the first message.
Personal reply, No obligation, English, Español
What it is
Chatbot or AI agent? The difference is what it gets done
A chatbot answers; an AI agent also acts. Good AI chatbot development produces agents that understand natural language, check your approved information and take concrete steps: offer an appointment, log the enquiry in your CRM or pass the case to your team with full context.
Button-and-menu bots still suit narrow tasks but break the moment someone goes off-script. AI agents for business built on language models hold a natural conversation, ask for missing details and know when not to answer. That ability to stop is what makes them dependable.
As a conversational AI agency, we design, build and monitor agents for website chat, the phone (an AI voice agent or AI receptionist) and email; WhatsApp has its own WhatsApp chatbot page. Four principles apply throughout: transparency, human handover, data protection by design and clear measurement.
In depth
Where does an AI agent work? Website, phone and inbox
An AI agent works wherever customers already write or call: website chat, your phone line and email. The knowledge base stays the same; only the interface, tone and permitted actions change. Each channel can run in several languages, including Catalan for customers across the Balearics.
A fourth 'channel' is growing: AI browsers and assistants that visit your site for people. Google's AI search optimisation guide (updated July 2026) notes they may read screenshots, the DOM and the accessibility tree, so semantic, accessible pages help. We handle that through UX/UI and accessibility or web development.
- Chatbot for your website: product and service questions, quotes, bookings and order status.
- AI voice agent: after-hours calls, seasonal peaks, appointments and call routing.
- Email: sorting by intent and urgency, with drafts ready for review.
- Typical use cases: restaurant bookings, wellness appointments, software demos or online-shop returns.
In depth
Knowledge base and guardrails: it only says what you approve
A dependable agent answers from your approved information (services, terms, hours and policies), not from whatever the model happens to know. We connect that content through retrieval-augmented generation (RAG) and agree in writing which topics it covers and which it hands over.
Guardrails come first: no invented prices or availability, no medical, legal or financial advice, and a clear 'that's outside my scope' when needed. We test it with real questions, edge cases and manipulation attempts (prompt injection) before launch and after every change.
- Versioned sources, each with a named owner in your team.
- Allowed topics, topics to hand over and excluded topics.
- Brand tone and courtesy phrases defined for each language.
In depth
CRM and calendar integration: from message to opportunity
An agent earns its keep when conversations reach your systems without copy and paste. We connect the agent via API or automation workflows to your CRM, calendar and booking tools, so each conversation ends in a useful record: contact, intent, qualification, consent and next step.
We agree with your sales team what counts as a qualified lead and who receives it. The agent feeds your marketing automation rather than duplicating it, and qualified leads can return to your campaigns as conversions, with consent.
- CRM: new or updated contacts, deals and notes with a conversation summary.
- Calendar: real availability, booking, confirmation and reminder.
- In-house systems: order, booking or ticket status via their API.
In depth
Human handover: when and how a person takes over
Every agent we build has a clear route to a person. Customers can ask for one at any time, and the agent escalates itself on complaints, sensitive topics, questions it can't answer safely or urgent sales opportunities.
Whoever takes over gets a summary and the history, so nobody repeats themselves; if no one is free, the agent offers a call-back. In Spain, Ley 10/2025 bars large companies and basic-service providers from making automated systems their only customer-service channel; we apply that standard to every client.
- Chat: live transfer to a team member or a support ticket.
- Voice: call transfer to an extension or an on-call mobile.
- Email: assignment to the right owner, with priority and context.
In depth
Data protection by design
Conversations hold personal data, so privacy is designed into the architecture. We collect only what's needed, show a short privacy notice from the first message and, on calls, say upfront whether they are recorded or transcribed.
We use providers that sign a data processing agreement, favour set-ups where they commit in writing not to train on your data, and document every data flow, international transfers and safeguards included, plus technical input for any impact assessment.
- Minimisation: the agent never asks for data it won't use.
- Defined retention periods and scheduled deletion of transcripts.
- Role-based access, activity logs and a register of providers.
In depth
Ongoing monitoring: no launch-and-forget agents
An AI agent needs upkeep like any service channel: automatic alerts (integration errors, handover spikes, out-of-scope answers), human review of conversation samples and fresh tests whenever the model or your offer changes. If anything goes wrong, it pauses instantly and traffic goes to people.
Conversations are also business intelligence: what customers ask, which objections recur and what your website lacks. We feed that into conversion rate optimisation and your content.
What's included
What our AI agent service includes
From use cases to monitoring, with every deliverable documented.
Use-case map
What to automate, what to assist and what must stay human, based on your real conversations.
Conversation design
Scripts, tone per language, an AI-disclosure welcome, qualifying questions and handover routes.
Curated knowledge base
Approved, structured and versioned content, with an owner for updates and coverage tests.
Website chatbot
A lightweight widget in your visual identity, usable by keyboard and screen reader.
AI voice agent
A virtual receptionist that answers, books and transfers calls, on its own number or forwarded.
Email triage
Messages sorted by intent and urgency, drafts for review and routing to the right owner.
Integrations
CRM, calendar, booking or order systems and internal tools, with every agent action logged.
Compliance & monitoring
Transparency and privacy texts, processing records, a metrics dashboard and a continuous review plan.
How we work
How we build an agent
Phased, with testing at every step: we start narrow and expand what proves itself, as in our method.
Phase 01 / 06
Discovery
We review real conversations, calls and emails and define goals, limits and handover criteria.
Phase 02 / 06
Design
Scripts, knowledge base, architecture, data flows and legal texts, agreed with you before anything is built.
Phase 03 / 06
Build & integrate
We configure the agent, connect it to your systems and set up the metrics dashboard and alerts.
Phase 04 / 06
Testing & pilot
Real questions, edge cases and manipulation attempts, then a pilot on limited traffic.
Phase 05 / 06
Gradual roll-out
More channels, hours and languages as the data justifies it, with your team ready to take handovers.
Phase 06 / 06
Monitor & improve
Regular conversation reviews, content updates and fresh tests whenever the model or your offer changes.
The SERMAN approach
Why build your agent with SERMAN
Compliance built in from day one
AI disclosure, human handover and privacy are designed in, not patched on, and documented so your team can explain them.
Marketing and technology in one team
More than an AI automation agency: the agent connects to your campaigns, website and analytics, and its insights feed your strategy.
Documented and under your control
Provider accounts can be in your company's name, and we document configuration and workflows: never a black box.
Judgement before automation
We'll tell you what shouldn't be automated, even when the answer is a good form or a person.
What it is not
What this service is not
- Not a general-purpose assistant with your logo: the agent only covers the business topics you approve.
- Not a replacement for your team: it takes on repetitive work and passes on what needs human judgement.
- Not a source of medical, legal or financial advice, nor of decisions that significantly affect people.
- Not a template installed and abandoned: without monitoring, no agent stays reliable.
- Not a tool for mass sales calls or messages sent without consent.
- Not a promise of a set number of sales or leads: we measure openly and adjust.
What we measure
How we know the agent is working
Service quality and business impact, with every event sent to your analytics.
Resolved conversations
Volume by channel and language, and the share resolved without human help within scope.
Qualified leads
Contacts that meet your sales team's criteria and reach the CRM complete.
Handover rate
The share of conversations passed to a person, and the reason.
Appointments & bookings
Bookings, appointments or demos confirmed by the agent, and actual attendance.
Response times
Time to first reply and to resolution or handover, in and out of office hours.
Answer quality
Accuracy against the source, correct handovers and customer ratings, from reviewed samples.
Tools & platforms
Tools and platforms we work with
- OpenAI API
- Anthropic Claude API
- Google Gemini API
- ElevenLabs
- Twilio
- Vapi
- n8n
- HubSpot
- Pipedrive
- Salesforce
- Google Calendar
- Calendly
- Microsoft 365
- Langfuse
Who it's for
A particularly good fit if…
- You get lots of repeat enquiries, including out of hours.
- You miss calls in high season or at peak times.
- Your sales team spends too long filtering poor-fit contacts.
- You serve customers in several languages.
- You are a large company adapting its customer service to Spain's Ley 10/2025 by 28 December 2026.
- You have a CRM and calendar, but conversations don't reach them with clean data.
Compliance
The rules we apply to every agent
A practical summary as of October 2026, not a substitute for legal advice.
EU AI Act, Article 50
Since 2 August 2026, people interacting with a chatbot or voice agent must be told at the start that it is an AI, unless that is obvious. Our agents always say so in their first message or sentence. Fines can reach €15 million or 3% of annual worldwide turnover.
Spain's customer service law (Ley 10/2025)
It covers large companies (250+ employees, turnover above €50 million or a balance sheet above €43 million, including at group level) and basic-service providers, which must adapt by 28 December 2026. Service cannot be automated-only, customers can ask for a person, and co-official languages (Catalan in the Balearics) apply where the service targets those regions. Smaller firms aren't directly covered; we add handover anyway.
GDPR and Spain's LOPDGDD
People talking to the agent must get the GDPR Article 13 information: controller, purpose, legal basis, recipients, retention, rights and the right to complain to the AEPD (Spain's data protection authority). Spanish law allows layers: a short one in the conversation, linked to the full policy.
WhatsApp Business policy
As of October 2026, Meta's policy allows businesses to use AI on WhatsApp for their own customer service, bookings and sales; its restrictions target general-purpose AI assistant providers. Our agents are scoped to your business; see WhatsApp chatbot.
Industries
Where we apply it
Hospitality, Travel & Travel Tech
Direct bookings, reputation and visibility on Google, Maps and AI assistants for hotels, restaurants, experiences and travel tech.
Private Clinics, Aesthetics & Wellness
Marketing for dental, aesthetic, medical and wellness clinics: local search, AI answers, genuine reviews and multilingual booking.
Professional, Financial & B2B Services
Marketing for law, tax and advisory firms, financial services and B2B companies: citable expertise, AI visibility, LinkedIn and a measurable pipeline.
Education & Training
Education marketing for private universities, business schools, international schools and academies: student recruitment in Spain and abroad.
FAQ
Frequently asked questions
What is an AI agent for business?
A conversational system built on language models that serves customers by chat, voice or email and also acts: it qualifies enquiries, books appointments, logs data in your CRM or hands over to a person. Ours use only your approved information, within limits agreed in writing.
What's the difference between a chatbot and an AI agent?
A traditional chatbot follows a decision tree or returns fixed answers. An AI agent understands natural language, keeps context, consults sources and completes tasks in other systems. The line is gradual; what matters is what it must resolve, and with which controls.
Is it legal to use an AI chatbot for customer service in the EU?
Yes, provided it discloses that it is an AI from the start (Article 50 of the EU AI Act, applicable since 2 August 2026), handles data under the GDPR and, at large companies in Spain, offers human attention under Ley 10/2025. AESIA is Spain's national AI supervisory authority.
Do I have to tell customers they're talking to an AI?
Yes. Since 2 August 2026, people must be told at the start, unless it is obvious. Our agents always say so in their first message or sentence, for example: 'Hi, I'm the AI virtual assistant for…'.
Can an AI voice agent answer calls in several languages?
Yes. It detects the caller's language or asks, then replies in Spanish, English, German, French, Catalan or other languages. We check voice and vocabulary in each language before going live, as naturalness varies by provider.
How does an AI agent connect to my CRM?
Through the CRM's API or automation workflows. We define what is saved (contact, intent, qualification, consent, summary) and what follows: assignment, a task or a follow-up sequence. We work with HubSpot, Pipedrive, Salesforce and in-house systems with an API.
Does Spain's new customer service law ban chatbots?
No. Ley 10/2025 stops chatbots being the only channel at large companies and basic-service providers, which must adapt by 28 December 2026, and lets customers ask for a person. A well-designed agent fits: it resolves simple cases and hands over the rest.
What happens when the agent doesn't know the answer?
It says so rather than guessing, then collects details for your team or passes the conversation to a person. Unanswered questions are reviewed and, where useful, added to the knowledge base.
How long does AI chatbot development take?
It depends on channels, languages and integrations. A focused pilot, such as website chat with bookings, can be ready in a few weeks; voice, several languages and in-house systems take longer. The diagnosis gives you a phased timeline.
Do you work with businesses outside Mallorca?
Yes. We are based in Palma de Mallorca and build agents for businesses across Spain and abroad, in their customers' languages. To decide first which marketing processes benefit from AI, see AI consulting.
Related services
Pairs well with
WhatsApp AI Chatbot
AI chatbots on the official WhatsApp Business API that answer, qualify, book and hand over to your team when it matters.
Lifecycle Marketing, CRM & Automation
One system for CRM, email, WhatsApp and SMS that guides every contact from sign-up to repeat purchase, built on consented first-party data.
AI Marketing Strategy & Training
AI marketing strategy for teams: prioritised use cases, the right tools, human-reviewed workflows and hands-on training, with governance built in.
Conversion Rate Optimisation (CRO)
Research, A/B testing and UX fixes that help more visitors become customers, with no dark patterns and accessibility as a conversion lever.
Web Development
Custom web engineering with Next.js and headless: measurable performance, integrations with your systems and care plans that protect the build.
Insights
Go deeper
1 Oct 2026 · 11 min read · AI, data & automation
AI agents for lead generation: WhatsApp, web and voice (2026 guide)
How to use AI agents for customer service and lead generation on WhatsApp, web and voice: qualification flow, WhatsApp costs, AI disclosure, human handover and metrics.
Contact
What could your agent handle?
We review your current conversations and show you what an agent could handle, what should stay with your team and how we would measure it.
Personal reply, No obligation, English, Español
Start with SERMAN
Tell us about your project
First conversation
Diagnosis and proposal