Customer-facing assistants
Trained on your pricing, policies, catalog and past tickets. Answers questions on your site, in chat and in social inboxes, at 3am, in your tone of voice, and escalates cleanly when unsure.
AI that does the repetitive work, so your team does not have to.
Not a demo, and not a chatbot that recites a menu. We connect language models to your actual data and systems so they answer customers correctly, move information between tools, and hand off to a human the moment they should.
Where AI earns its keep
Five places it pays for itself first
Document and data processing
Purchase orders, invoices and forms arriving in twenty different formats, read and entered into your systems automatically. This is where most manual hours quietly disappear.
Lead qualification and routing
Incoming inquiries summarized, scored and routed to the right person with the context attached, so your team starts the conversation already knowing what it is about.
Internal knowledge assistants
Your policies, specs and documentation made answerable in plain language. New hires stop interrupting senior staff for things that are written down somewhere.
Workflow automation
The connective work between tools: statuses updated, notifications sent, reports assembled, records synced. Often the fastest return of any AI project, because it needs no new interface at all.
Paid pilot, then scale
We start with one process, run it under supervision for two to three weeks and measure what it actually handled. You decide to scale based on your numbers, not on a vendor promise.
How we keep it honest
The failure modes are known, so we design against them
AI systems fail in predictable ways: they invent answers, they leak information they should not, and they cheerfully handle a conversation that a human should have taken over. We build against each one specifically. Assistants answer only from your material, refuse and escalate when confidence is low, keep sensitive data separated from public responses, and log every exchange so you can audit what was said. In the first weeks we review those logs together and tighten the behavior.
Contact us
Pricing
What AI integration costs
The pages that rank for this question are written for enterprise buyers, describing programmes far larger than most companies searching actually want. We work in the band a mid-sized business can justify, and these are the actual numbers. The figures are indicative and exist so you know the order of magnitude before we talk. The final price is set per project, against the scope we define together.
AI readiness audit
1,900 EUROne week, fixed price, credited in full against the build. We look at what your customers actually ask and where your team's hours go, and tell you if it is worth doing. Sometimes the answer is not yet.
- Analysis of real customer questions
- Hours-saved estimate with the working shown
- Costed proposal and timeline
- Credited if you proceed
AI assistant
3,900 - 16,000 EUR + 290 - 490 EUR/moTrained on your content, live on your site and one other channel. Answers the questions your team answers twenty times a week.
- Trained on your own documents
- Site plus one channel
- Human handover built in
- Monthly report of unanswered questions
Multichannel agent
from 9,900 EUR + 490 - 890 EUR/moSite, WhatsApp, email and CRM handoff. Reads live data, so it can answer where an order is rather than only what your policy says.
- Multiple channels with one knowledge base
- Live data from your systems
- CRM handoff with full conversation history
- Action log for auditing
Bespoke AI system
from 19,000 EURRetrieval over your own documents, multi-step agents, document processing. Built for a specific process, with a human approving anything that matters.
- Retrieval over your document set
- Multi-step workflows with approval gates
- Evaluation suite so quality is measured
- Team training and handover
Year one, all in
Before you commit, we put the first-year total in writing: build, hosting, support and model usage in one number. Everyone else quotes only the build, and the running cost is the part that surprises people six months later.
Language model token cost is passed through with no markup and itemized every month, so you can see exactly what the assistant consumed. Unlike per-resolution pricing, it does not rise simply because more customers wrote to you. Your data is not used to train anyone else's model.
What moves the price
Three things, in the order they matter:
How many channels it has to cover
Your site alone is one deployment. Site, WhatsApp, email, Instagram and a phone line are five, each with its own approval process and its own failure modes.
Whether it needs to read live data
An assistant answering from documents is straightforward. One that tells a customer where their order is has to connect to your store and your carrier, and that is software development rather than configuration.
How clean your data is
If the answers live scattered across spreadsheets and inboxes, an assistant will confidently give wrong ones. Then the first job is tidying the data, and occasionally that first step turns out to be the only one you needed.
What we need to quote accurately
Four answers and the quote is firm:
- Which questions you answer every day, and how often
- Which channels customers reach you on today
- Where the correct answers currently live
- What you would call success after three months
Selected work
AI systems in production
FAQ
Frequently asked questions
How quickly can an AI assistant go live?
One to two weeks from receiving your material for a typical customer-facing assistant. The first weeks in production are supervised, with answers reviewed and refined against real conversations.
Is our data used to train someone else's model?
No. Your material is used to answer your customers, under agreements that exclude training on your data. Sensitive material stays separated from anything customer-facing, and we can deploy in your cloud environment when policy requires it.
What if the assistant does not know the answer?
It says so and hands the conversation to a person, with the context attached. Refusing gracefully is a design requirement, not an accident, because one confidently wrong answer costs more trust than ten honest escalations.
Will this replace our staff?
In practice it takes the repetitive layer: the same questions, the same data entry, the same reminders. Teams we work with rarely shrink; they stop spending mornings on work nobody wanted to do and handle more volume with the same people.
Which models and platforms do you use?
We are deliberately model-agnostic and choose per use case based on cost, accuracy and privacy requirements, including options that run in your own environment. Being able to switch as the market moves is part of the value.
How much does AI integration cost?
The tiers are published above on this page, in the table, with the build cost and the monthly cost stated separately in every one: a readiness audit, an assistant, a multichannel agent and a bespoke system. The pages that rank for this question are written for enterprise buyers, not for the companies actually searching.
What will it cost in the first year, all in?
We put it in writing before you commit: build, hosting, support and model usage as one first-year number. We do that because everyone else quotes only the build cost, and the running cost is what surprises people six months later.
Who pays for the model tokens?
You do, at cost, itemized, with no markup from us, and you see the figure every month. Unlike the per-resolution pricing some platforms use, it does not climb simply because more customers wrote to you that month.
Is our data used to train someone else's model?
No. We use enterprise API tiers where the provider contractually does not train on submitted data, and we say in the proposal which provider handles what. If you need the whole thing to run on infrastructure you control, that is possible and we will tell you what it adds to the cost.
