Answers at any hour
Your customers get an immediate answer, in the evening, at weekends and in their own language, instead of waiting or going to a competitor.
Custom chatbots, internal assistants and AI agents
AI assistants for your business, your website and your teams
Your customers ask the same questions day and night, your teams hunt for information across dozens of documents, and simple tasks wait for someone to retype them. We design, build and integrate AI assistants that work for your business: an assistant on your website that answers visitors from your own content, an internal assistant that knows your procedures, and agents connected to your CRM, ERP or calendar that take action. Every answer draws on your sources, your data stays protected and a human steps in whenever needed. We built Gaspard, the AI assistant of Valais Connect, and bring the same standards to your projects, entirely remotely.
Your outcome
A solution that's useful today and solid tomorrow.
Freed-up teams
Repetitive questions and retyping are taken care of: your teams focus on the requests that need them.
Answers grounded in your sources
The assistant answers from your up-to-date documents, cites its sources and hands over when it does not know, instead of making things up.
You stay in control
Your data, knowledge and code belong to you; the AI provider can change without rebuilding everything, and costs stay capped.
Demo
An assistant at work, in three situations
It understands the question, searches your sources, answers by citing them, then acts in your tools or hands over to your team.
Your assistant
Your website's chat window
Sample conversation, for illustration only.
The service in detail
AI assistants: what the service includes
Who it's for: For businesses, shops, firms, schools and associations that receive a lot of questions, whose teams lose time searching for information, or that want to automate repetitive tasks with AI without exposing their data.
01
Scoping and use cases
We identify where an assistant genuinely saves you time or wins you customers, and where it is pointless. Before spending anything, you know what the assistant will do, what it will not do and what it will cost to run.
- Workshops with your customer-facing teams
- Inventory of repetitive questions and tasks
- Use cases ranked by benefit and risk
- Estimate of the monthly running cost
What you get: A clear roadmap: the first assistant to build and the indicators that will tell you whether it succeeds.
02
Assistant for your website
An assistant built into your site answers visitors at any hour, in their language, from your pages, catalogue and FAQ. It guides them, qualifies requests and passes to your team whatever deserves a human.
- Chat window in your brand colours, on desktop and mobile
- Works with WordPress, WooCommerce, Shopify, Angular, React or a custom site
- Appointment booking, quote requests and lead qualification
- Handover to a human by email, chat or ticket
What you get: Visitors who get an immediate answer and enquiries that arrive already qualified.
03
Internal assistant for your teams
Your colleagues ask questions in plain language and get an answer drawn from your procedures, contracts, product sheets or technical documentation, with a link to the original document.
- Search across SharePoint, Google Drive, Notion, Confluence or your PDFs
- Access rights respected: everyone sees only what they are allowed to see
- Available in Microsoft Teams, Slack or your intranet
- Summaries, draft replies and meeting notes
What you get: Fewer repeated questions to the same people, and new colleagues who become autonomous faster.
04
Knowledge of your business
An assistant is only as good as what it knows. We connect the AI to your content through retrieval-augmented generation (RAG): it answers from your up-to-date documents, cites its sources and says so when it does not know, instead of making things up.
- Import, cleaning and chunking of your documents
- Vector knowledge base hosted for you
- Automatic update when your content changes
- Sources cited in every answer
05
AI agents that take action
Beyond answering, an agent can act in your tools: create a ticket, check an order, book a slot, prepare a quote or follow up with a customer. Sensitive actions always wait for human approval.
- Connection to your CRM, ERP (Odoo, SAP), calendar or mailbox
- Automations with n8n, Make, Power Automate or custom code
- Human approval before any send, payment or deletion
- Log of every action taken
06
All your channels
The same assistant, with the same knowledge, can answer wherever your customers and teams already are, without multiplying the tools you have to maintain.
- WhatsApp Business, Messenger and Instagram
- Email sorting and draft replies
- Integration into your mobile apps
- Voice dictation and spoken answers
07
Data protection and privacy
Your data is not used to train providers' models. Depending on how sensitive your information is, the assistant uses an AI provider with clear contractual commitments or a model installed on your own servers.
- Model hosted in Europe or on your servers
- Names, emails and numbers removed before anything is sent externally
- GDPR compliance and the transparency required by the European AI Act
- Safeguards against off-topic answers and misuse
08
Assistant and infrastructure security
A public-facing assistant is a new way into your business. We protect it like a sensitive application: it can only reach what it strictly needs, every exchange is encrypted, and we attack it ourselves before your visitors do.
- Protection against prompt injection and hijacking attempts
- Least privilege: the assistant only gets the rights it needs, tool by tool
- Encryption of exchanges and stored data, API keys kept out of the code
- Authentication of internal users (SSO, Microsoft Entra ID, Google)
- Abuse limits: message rates, bots and costs capped
- Attack testing (red teaming) before launch and with every change
- Logging, alerts on abnormal behaviour and an emergency kill switch
What you get: An assistant that doesn't reveal what it shouldn't, only does what you've authorised, and can be shut down in one click.
09
AI compliance and governance
We help you use AI by the rules and prove it: user information, contracts with providers, retention periods and written internal rules.
- Clear disclosure: users know they're talking to an AI, as the AI Act requires
- Data processing agreements (DPAs) with AI providers and a GDPR record of processing
- Data protection impact assessment (DPIA) when sensitive data is involved
- Retention periods, right of access and right to erasure of conversations
- AI usage policy for your staff
- Traceability: which sources and which model produced each answer
10
Monitoring, measurement and improvement
An assistant improves with its real conversations. We measure what it gets right, spot the questions where it fails and improve it every month, while keeping an eye on its cost.
- Test question set replayed after every change
- Dashboard of conversations and handed-over requests
- Regular enrichment of the knowledge base
- Usage cost tracking and caps
Out of scope or optional
- Use of AI models (OpenAI, Anthropic, Mistral, Google…) and hosting are billed in your name or re-invoiced at cost; we give you an estimate before we start.
- No assistant is infallible: we sharply reduce errors and always provide a human handover, but we do not promise 100% correct answers.
- We do not train AI models from scratch: we assemble and adapt the best existing models to your business.
Our work
Gaspard
The AI assistant we designed for Valais Connect: it answers from any page, knows the app inside out and prepares lessons from each student's courses. It runs first on a model installed on our own servers, removes personal data before any external call and deletes conversations after 30 days.
Discover the projectDeliverables
AI assistants: what you receive
- Use case study and cost estimate
- Prototype tested on your real questions
- Assistant integrated into your website, tools and channels
- Knowledge base, safeguards and test set
- Dashboard, documentation and training for your teams
- Security and compliance file: risks, measures, attack tests and GDPR documents
Expertise
AI assistants: technologies and practices
OpenAI (GPT)Anthropic (Claude)Mistral AIGoogle GeminiLlama / OllamaLangChainpgvector / QdrantMicrosoft Copilot Studion8n / MakeWhatsApp Business API
Our method
We start where you are.
A brand-new idea, a product already live or an urgent situation can't be handled the same way. We adapt how we start the engagement before applying a rigorous, transparent process.
Website01
Your website keeps getting the same questions
Opening hours, prices, lead times, availability: we add an assistant to your site that answers from your content and passes real enquiries to your team.
First steps
- List of the most frequent questions
- Collection of your pages, FAQ and catalogue
- Prototype on your test site
Teams02
Your teams look for information everywhere
Procedures, contracts, technical sheets: we build an internal assistant that answers from your documents while respecting everyone's access rights.
First steps
- Inventory of document sources
- Access rules by team
- Test with a pilot team
Automation03
You want a task to run on its own
Handling incoming requests, preparing quotes, updating a CRM: we design an agent that acts in your tools under human control.
First steps
- Step-by-step description of the current process
- List of the tools and access needed
- Definition of the human approval points
How the engagement unfolds
Every step produces tangible proof.
You always know what's been done, what you're receiving and what needs to be approved before moving on.
- 013 to 7 days
Scoping
We pick the use case that pays off most, define what the assistant must be able to do and estimate its cost.
What we do
- Workshops with the teams involved
- Inventory of questions, documents and tools
- Choice of model and hosting
Why this step: An assistant that tries to do everything ends up doing everything badly. This scoping focuses the effort on the most profitable use case, sets what the assistant must never do and chooses the model and hosting suited to how sensitive your data is, before a single line of code is written.
How we proceed, line by line
- 1 A first video call sets your goals: winning customers, relieving support, speeding up an internal task.
- 2 With your teams, we list the repetitive questions and tasks, based on your emails, tickets or chat histories.
- 3 We take stock of the documents and tools the assistant will need to know or use.
- 4 We rank the use cases by expected benefit and risk of error, and select the first one.
- 5 We choose the AI model and hosting according to your privacy requirements and budget.
- 6 We write the assistant's personality, tone and limits, and the success indicators.
Who does what
An AI consultant runs the workshops and writes the specification, and an architect checks technical feasibility and estimates costs. On your side, a business owner describes the needs and a few customer-facing colleagues share the questions they receive.
What you provide
- Real examples of questions or requests you receive.
- The list of your documents, tools and confidentiality constraints.
- One hour of availability from the people who handle these requests today.
Deliverable
Assistant specification and cost estimate
Approval checkpoint
You approve the use case, scope and budget.
- 021 to 2 weeks
Prototype
We quickly build a first version fed with your content and test it on your real questions.
What we do
- Import and preparation of your content
- First instructions and safeguards
- Test question set and answer measurement
Why this step: With AI, only real questions show whether an idea works. A short prototype tested on your actual cases shows the quality that can be reached and the gaps in your content before you invest in the full integration.
How we proceed, line by line
- 1 We import, clean and chunk your documents, pages and FAQ into a knowledge base.
- 2 We write the assistant's instructions: role, tone, permitted topics and what to do when it does not know.
- 3 Together with you, we build a test question set, with the expected answer for each.
- 4 We measure answer accuracy, source citation and response time.
- 5 Your teams try the prototype online and rate the answers.
- 6 We adjust the instructions and fill in missing content until the agreed quality level is reached.
Who does what
An AI engineer builds the prototype and the knowledge base, while the AI consultant prepares the test questions and analyses the results. On your side, a few testers judge the answers and a business contact settles doubtful cases.
What you provide
- Your documents, pages, FAQs or catalogues in their up-to-date version.
- A review of the expected answers in the test set.
- A few testers available for a few days.
Deliverable
Online prototype and quality report
Approval checkpoint
Your teams judge the prototype's answers.
- 032 to 6 weeks
Build and integration
We integrate the assistant into your website, tools and channels, and connect it to your systems when it needs to act.
What we do
- Integration into the website, Teams, Slack or WhatsApp
- Connection to the CRM, ERP or calendar
- Security, human handover and dashboard
Why this step: A good assistant on its own remains a gadget. Integrated where your customers and teams already are, connected to your systems and surrounded by safeguards, it becomes a reliable work tool that alerts the right person when it needs to hand over.
How we proceed, line by line
- 1 We embed the chat window in your site, in your brand colours, or the assistant in Teams, Slack, WhatsApp or your app.
- 2 We connect the assistant to your systems through their APIs: CRM, ERP, calendar, online shop or ticketing tool.
- 3 We add human approval before any sensitive action and a log of every action.
- 4 We set up the handover to a human, with a summary of the conversation so the customer does not have to repeat themselves.
- 5 We add data protection: anonymisation, retention period, access rights and user information.
- 6 We set up the conversation dashboard, alerts and spending cap.
- 7 A weekly video demo shows you the progress.
Who does what
An AI engineer and a developer integrate the assistant and its connections, an architect looks after security and a project manager organises the demos. On your side, an administrator opens the access and a business contact validates the behaviour.
What you provide
- Access to your website, your tools and their APIs.
- The handover rules: who takes over, on which channel and at what times.
- Your brand assets and the notices to show users.
Deliverable
Complete assistant in a test environment
Approval checkpoint
Every integration and every action is checked with you.
- 041 to 2 weeks
Testing and go-live
We test the assistant to its limits, launch it gradually and train your teams.
What we do
- Quality, security and load testing
- Gradual launch to part of your users
- Training and documentation
Why this step: Users always ask questions nobody anticipated, and some try to misuse the assistant. Testing it to its limits and then launching gradually lets us fix issues on a small scale, before your customers see the flaws.
How we proceed, line by line
- 1 We replay the test question set and check the agreed quality level.
- 2 We try to misuse the assistant: off-topic questions, requests for other customers' data, prompt injections.
- 3 We check how it holds up under load and its response time.
- 4 We open the assistant to part of your visitors or to one team, then widen access based on the results.
- 5 We train your teams to follow conversations, take over and update the knowledge base.
- 6 We hand over the technical documentation and a simple user guide.
Who does what
The AI engineer runs the tests and the launch, the AI consultant trains your teams and the project manager coordinates go-live. On your side, the people who will take over conversations attend the training.
What you provide
- Approval of the test results.
- Training slots for the teams concerned.
- Your green light for each launch stage.
Deliverable
Live, documented assistant and trained teams
Approval checkpoint
You approve the launch for all users.
- 05Ongoing
Monitoring and improvement
We analyse conversations, improve the assistant every month and keep its costs under control.
What we do
- Monthly analysis of conversations
- Knowledge base enrichment
- Model updates and cost tracking
Why this step: Your offers, prices and procedures change, and AI models evolve every few months. Without monitoring, an assistant gets old and ends up giving outdated answers; with monthly follow-up, it gets better the more it is used.
How we proceed, line by line
- 1 Each month we analyse conversations, unanswered questions and requests passed to a human.
- 2 We fill gaps in the knowledge base and adjust the instructions based on these results.
- 3 We replay the test set after every change to prevent regressions.
- 4 We evaluate new AI models and switch when they are better or cheaper.
- 5 We send you a clear report: volume, success rate, costs and next improvements.
- 6 Once the first assistant has proved itself, we suggest the next use cases.
Who does what
A Dealmood AI engineer monitors and improves the assistant, and the AI consultant presents the monthly report. On your side, a business contact flags changes to offers or procedures and validates priorities.
What you provide
- Changes to your offers, prices or procedures.
- Feedback from your teams and customers.
- Half an hour a month for the review meeting.
Deliverable
Monthly report and improved assistant
Approval checkpoint
Success indicators improve every month.
One conversation is enough to choose the right starting point.
We look at your context before proposing a clear scope, timeline and responsibilities. Our team is spread around the world. Depending on your project, if a team member is present in your country of residence, they can meet you to follow up on your project.
AI assistants: frequently asked questions
How is this different from a standard chatbot?
A standard chatbot follows fixed scripts and gets stuck as soon as a question goes off-script. An AI assistant understands freely worded questions in many languages, answers from your documents and can act in your tools. We keep strict rules where they are needed: permitted topics, actions approved by a human.
Which AI model do you use?
The one that suits your case: OpenAI's GPT, Anthropic's Claude, Mistral, Google's Gemini, or an open model such as Llama installed on your servers. We take no commission and build the assistant so that you can switch models without starting over.
Is my data used to train the AI?
No. We use providers' business plans, which do not train their models on your data, or a model hosted on your side. Personal information can be removed before anything is sent externally, and conversations are deleted after a period you choose.
What happens if the assistant does not know the answer or gets it wrong?
It is designed to say so rather than make things up, and to pass the request to your team with a summary of the conversation. Its answers cite their sources, and the monthly review fixes the errors we spot.
How much does the assistant cost to run?
The cost depends on the number of conversations and the model chosen; for a small business, it is often in the range of a few tens to a few hundred euros per month. We estimate it during scoping and set up a spending cap.
Can you add it to my existing website?
Yes. The assistant can be added to a WordPress, WooCommerce, Shopify, Wix or custom-built site without rebuilding it. It can also join Teams, Slack, WhatsApp or your mobile app.
How long does it take to get a first assistant?
A first assistant for your website, fed with your content, can be live in three to four weeks. An internal assistant or an agent connected to several systems usually takes six to ten weeks.
Can a hacker manipulate the assistant or make it reveal information?
It's the first risk we deal with. The assistant only has access to the information and actions it needs, its instructions are protected against injection, one customer's data is never accessible to another, and sensitive actions require human approval. We run attack tests before launch and with every change, and an alert warns us of any abnormal behaviour.
Where is the data hosted and who can read the conversations?
You choose: hosting in Europe, with your cloud provider or on your own servers. Conversations are encrypted, kept for as long as you decide, and only the people you designate can view them. Every access is logged.
Will my assistant comply with the GDPR and the AI Act?
We design it to: clear user information, processing agreements with AI providers, retention periods, rights of access and erasure, and an impact assessment when sensitive data is involved. We hand you a compliance file; final legal sign-off remains with your counsel or data protection officer.
Your project deserves a clear framework
Let's build a solution that fits you.
Tell us what you need. We'll reply with clear initial guidance.
Start the conversation