Home / Custom AI chatbot development

Have a chatbot built, or buy one

Custom chatbot development services, connected to your systems

Choose custom chatbot development services when it must read or write live data in your ERP or ticketing system, or run under strict data rules. When your website, PDFs and FAQ hold the answers, SimplyBoost, our own product, is enough.

The question for an IT manager is whether to buy a product or have a chatbot built that works inside your own systems, and who maintains it after go-live. This page sets out both routes, including the one where we send you to our own product.

By Tom Joseph · Last updated: 29 October 2026

Free consultation Pricing

In short

Custom AI chatbot development pays off when the answers are not in your documents but in your systems. If the chatbot has to read an order status from the ERP, open a ticket in TOPdesk or Zendesk, serve employees by role or run inside your own cloud environment, Crux Digits builds it on your systems, alongside your IT partner, and hands over the code. If the answers are in your website, PDFs and FAQ, SimplyBoost, our own product, is faster and less expensive: a subscription instead of a project.

Buy or build

Buy a chatbot or have one built: the decision in one table

Question	SimplyBoost (product, monthly)	Custom chatbot built by Crux
Where do the answers come from?	Website, PDFs, Word files, Q&A lists, Shopify catalogue	The same, plus live data from ERP, CRM, ticketing system or database
What may the chatbot do?	Answer, book appointments, capture leads, hand over	Also open tickets, look up orders, register returns, update records
For whom?	Customers on website chat, WhatsApp and Instagram	Customers, employees or both, with access by role
Handover to a person	To email, Slack, Zendesk, Freshdesk or Gorgias	To any system with an API, such as TOPdesk or your own customer portal
Data rules	SimplyBoost's processing agreement	Model, storage and logging in your own cloud environment, under your policy
Ownership and cost	You subscribe to a service, from €39 a month	Fixed project price; code, prompts and test set are handed over to you

Route 1: the product

When SimplyBoost is enough, and why we say so

SimplyBoost is our own product, which is why we disclose it here. It is an AI agent that answers from your website, PDFs, Word files and Q&A lists, on website chat, WhatsApp and Instagram. It replies in the customer's language, such as Dutch, French, English or German, and hands a conversation to email, Slack, Zendesk, Freshdesk or Gorgias when a person is needed. A data processing agreement is available.

Choose this route when the questions coming in are already answered in your own documentation: delivery times, warranty terms, return policy. Plans start at €39 a month (pricing on simplyboost.io). A custom project for a chatbot that only quotes your FAQ is money badly spent. Run the product for a month and list the questions it cannot answer: that list is the scope of a custom build. For the wider customer service decision, see AI customer support agent: buy or build.

Route 2: custom

Having a chatbot built: four reasons to go custom

Answers from live systems. The customer asks where an order is or whether a part is in stock. That answer lives in your ERP (AFAS, Exact, SAP) or WMS, not in a PDF.
Actions in your systems. The chatbot opens an incident in TOPdesk, registers a return or changes a delivery address, with the right category and operator group. If it works across several systems, it is an agent: see having an AI agent built.
Internal use. Employees ask about procedures or contracts and see only what their role allows. That is a private AI assistant over your own documents.
Strict data rules. Your policy requires the model in your own Azure tenant and prompts in the EU. With an EU Data Zone deployment, prompts and responses are processed in EU member states and are not used to train models. Flagged prompts can still be stored for abuse monitoring unless modified monitoring is approved (Microsoft).

Knowledge sources

Where the chatbot gets its answers

A chatbot is as good as the sources it may read. In a mid-sized organisation there are usually five: the website, PDF manuals, the knowledge base in Zendesk or TOPdesk, the product catalogue from the PIM or ERP, and the ticket history. Each source gets an owner who decides what is current: a manual two versions old produces a confident, wrong answer.

We deliberately do not use the ticket history as a knowledge source. Old tickets hold personal data and answers that were right at the time. Instead we pull out the hundred questions that come back most often, with the answer your team marks as correct. That becomes the test set the chatbot is judged on before go-live. Every answer names the source it came from. When the chatbot finds no source, it says so and hands over.

Integrations

Connecting to your ticketing system: handover with context

A good handover puts the conversation, the category, customer or order details and the reason into the ticket, so nobody asks the customer twice.

The common systems have a documented API for this: the Zendesk Tickets API, the Freshdesk API and the TOPdesk REST API, which reaches incidents at /tas/api/incidents with an application password (TOPdesk developers). Categories and operator groups must match what is configured in your ticketing system; we settle that in week one with the functional administrator. We build the connection. The API account and its permissions come from whoever administers the ticketing system.

Working together

Alongside your IT partner: who does what

Your organisation owns content and process: which questions the chatbot answers, which sources count, when it hands over. Customer service signs off the test set.
Your IT partner or internal IT keeps what already sits with them: identity management, the cloud environment, network, ERP and ticketing administration, and the chatbot's API accounts.
Crux Digits builds the chatbot, the integrations, the test set and the monitoring, records everything in a runbook and hands over code, prompts and test set. We do not take over the running of your IT environment and we are not a hosting provider.
After go-live your own team or your IT partner maintains the chatbot from the runbook, agreed before the build starts.

Rules

Transparency. Since 2 August 2026, AI systems that interact directly with people must tell them they are dealing with AI, unless that is obvious to a reasonably well-informed user (AI Act, Article 50). The duty sits with the provider, and an organisation that has a chatbot developed and puts it into service under its own name is the provider under Article 3. So we build the notice into the first message. The EU AI Act risk checker shows any further duties.

Data processing agreement. When a supplier processes personal data on your behalf, a processing agreement is required, and the supplier may not engage a sub-processor without your prior written authorisation (GDPR, Article 28). For a chatbot the model provider is such a sub-processor, so ask for the list and the region the model runs in.

Cost and timeline

What a custom chatbot costs and how long it takes

Route or step	What you get	Duration	Price (excl. VAT)
SimplyBoost (product)	An AI agent on website chat, WhatsApp and Instagram, set up by your own team	No project	From €39 a month
AI Audit & Strategy	Questions, sources and systems reviewed, buy or build decided, a fixed price for the build	1 to 2 weeks	€2,500, credited against the MVP within 60 days
Production-ready MVP	The chatbot working on your sources and systems, tested against the test set with your customer service team	4 to 6 weeks	€20,000
Production	Robust integrations, roles, monitoring, logging, documentation, training and handover	From 8 to 12 weeks	From €50,000

Measurement

The first weeks, and how to measure whether it works

Week 1: question analysis, sources and owners, categories agreed. Weeks 2 and 3: sources loaded, ticketing integration built, first run against the test set. Weeks 4 and 5: customer service tests with real questions; each wrong answer is traced to the source that caused it. Week 6: go-live for part of the traffic, such as one channel. One named senior engineer stays with it from scope to handover.

Measure four things and agree the thresholds before the build: the share of conversations closed without handover, the percentage of correct answers on the test set, handover quality (did the agent have to ask anything again?) and the share of customers who come back about the same topic within seven days. That last figure, missing from most reports, shows whether a closed conversation was also resolved.

Examples

Examples by sector

Example, technical wholesaler: an installer asks whether an item is in stock. The chatbot reads stock and order status from the ERP; a short delivery becomes a ticket with order number and packing slip.
Example, heat pump brand: an owner reports a fault code. The chatbot looks the code up in the service manual, checks the warranty by serial number and opens a service request. More for brands and importers: AI assistant for dealer networks.
Example, internal service desk: employees ask how to request a laptop or get access to an application. The chatbot answers from the knowledge base and opens an incident in TOPdesk when the answer is not enough.

FAQ

Frequently asked questions

How do you make your own AI chatbot?

With a product such as SimplyBoost you set one up yourself: connect the website, upload PDFs, place the widget. If the chatbot has to read from or write to your ERP or ticketing system, it is a build project: choose the sources, assemble a test set from real questions, build the integrations, then measure.

How much does custom AI chatbot development cost?

A product costs a monthly fee; SimplyBoost starts at €39 a month. A custom chatbot at Crux starts with the €2,500 AI audit, followed by a €20,000 Production-ready MVP and production from €50,000, excl. VAT. Model usage and hosting come on top. More on what a chatbot costs and our pricing.

How long does it take to have a chatbot built?

Allow four to six weeks for a Production-ready MVP and from eight to twelve weeks for production with robust integrations, monitoring and handover. An audit beforehand takes one to two weeks. A product such as SimplyBoost needs no project: your team sets it up.

What makes a good AI chatbot?

A good AI chatbot answers only from sources you approved, names the source with every answer, says when it does not know and hands over with the conversation attached. It has been tested before go-live on real questions from your ticketing system.

How do I build an AI assistant for employees?

An AI assistant for employees is a chatbot over internal sources: procedures, contracts, the service desk knowledge base. The difference is access: each employee only gets answers from documents they have rights to. That is always a custom build; see the private AI assistant.

Do customers have to know they are talking to a chatbot?

Yes. Since 2 August 2026, Article 50 of the AI Act requires that people talking to an AI system are told so, unless it is obvious. The organisation that puts the chatbot into service under its own name is the provider and carries that duty; a short notice in the first message usually covers it.

Buy or build? We decide it with your own questions.

Describe the twenty questions that come in most often and the systems that hold the answers. Within one business day you hear which route fits: SimplyBoost, or a fixed-price scope for a chatbot on your systems.

Book a call