How Much Does an Internal AI Assistant Cost?

Your teams spend hours hunting for the right procedure, the right contract, the right technical sheet. They ask the same questions of their manager. Sometimes they work from outdated document versions. An internal AI assistant solves exactly this problem: employees ask their question in plain language, and the assistant answers by drawing on your actual documents.

But what does it really cost? The answer varies enormously depending on the option you choose. An off-the-shelf SaaS subscription can start at a few hundred euros per month. A custom assistant connected to your DMS, ERP, and internal knowledge bases can represent a 50,000-euro investment. Between those two extremes sits a middle option that is often underestimated.

We have deployed internal AI assistants for SMBs and mid-market companies across France, including for Actia Group, where teams cut their document retrieval time by 70%, and for Copro Assistance, which halved the time spent on reports. This article gives you real price ranges, the hidden costs no one mentions, and the cost-of-inaction calculation to help you make an informed decision.

What an internal AI assistant actually is

An internal AI assistant is a conversational system that answers your employees' questions by drawing on your actual documents: internal procedures, contracts, technical sheets, reports, project history, regulatory reference material.

The underlying technology is called RAG (Retrieval-Augmented Generation). When an employee asks a question, the system first retrieves the most relevant passages from your documentation, then a language model synthesizes them into a clear, sourced answer. The user can see where the answer came from and verify the source.

This is not an FAQ chatbot that answers from a list of pre-programmed questions. It is not ChatGPT making things up from general knowledge either. It is your documentation made conversational and instantly accessible to the entire team.

Concrete SMB use cases

• A maintenance technician asks: "What is the seal replacement procedure for pump P-12?" and gets the answer extracted from the manual in 5 seconds, including the part number to order.
• A sales rep asks: "What are our warranty terms for key account customers?" and receives the exact clause from the applicable master agreement.
• A new HR employee asks: "How does our leave approval process work?" and gets the current procedure, not the 2019 version they found on the file server.
• A construction project manager asks: "Have we had issues with this foundation type on clay soil before?" and receives extracts from past site reports.

This concrete promise is what justifies the investment. For more on use cases by sector, see our article on the most profitable enterprise RAG use cases.

The 3 options and their real price ranges

There is no single way to deploy an internal AI assistant. There are three, with very different levels of cost, flexibility, and control. Here is an overview before going into the detail of each.

Option 1: off-the-shelf SaaS (200 to 1,500 euros per month)

Platforms such as Microsoft Copilot (formerly Bing Chat Enterprise), Notion AI, Guru, and Glean offer AI assistants that connect to your existing tools (SharePoint, Google Drive, Confluence, Notion) with no custom development.

Indicative pricing in 2026:

• Microsoft 365 Copilot: approximately 30 euros per user per month (on top of the Microsoft 365 license). For 20 users: 600 euros per month.
• Notion AI: 10 euros per user per month as an add-on to an existing Notion subscription. For 20 users: 200 euros per month.
• Glean: starting around 10 to 15 dollars per user per month, but real entry point is roughly 1,000 dollars per month (contractual minimum). Oriented toward larger enterprises.
• Guru: approximately 18 to 20 euros per user per month for the AI-enabled tier. For 20 users: 400 euros per month.

What is included: a ready-to-use chat interface, connectors to major document platforms, basic access management, automatic updates.

What is not included: connection to your specific business systems (ERP, CMMS, proprietary DMS), adaptation to your sector's vocabulary, EU-only hosting for sensitive data, integration into your internal tools (your intranet, your business application).

Key takeaway: SaaS is fast to deploy but you are bound by what the platform supports. If your most important documents are in a proprietary system or in non-standard formats, SaaS will hit its ceiling quickly.

Option 2: simple custom build (8,000 to 20,000 euros)

This is the entry point for an internal AI assistant built on your actual data, with a controlled scope. Typically: a corpus of fewer than 2,000 well-structured documents (native-text PDF, Word, Markdown), a single data source, and a chat interface accessible internally.

What this budget covers:

• Document audit and preparation (cleanup, formatting, naming conventions)
• Ingestion and vector indexing pipeline
• Chat interface accessible internally (web or Teams/Slack integration)
• Basic access rights management
• Cloud deployment (OVH, Scaleway, or AWS depending on requirements)
• 2 to 4 weeks of user acceptance testing and tuning

Recurring costs after deployment run between 150 and 600 euros per month, depending on usage volume and the language model selected. This is the option we recommend to validate the use case before investing further. It can be built in 4 to 8 weeks. Our article on RAG project costs and TCO covers the architecture and real cost breakdown for this type of deployment.

For a technical overview of the approach, our RAG primer gives a solid grounding without unnecessary jargon.

Option 3: advanced custom build (30,000 to 80,000 euros)

This is the solution for companies that need an internal AI assistant integrated into their actual working environment, with multiple document sources and high security requirements.

This level of project typically includes:

• Connection to multiple heterogeneous sources: DMS, ERP, SharePoint, emails, SQL databases, scanned documents
• Advanced parsing of complex documents (tables, diagrams, image-based PDFs via OCR)
• Granular access controls: each user only sees what they are authorized to see
• AI agents capable of triggering actions (creating a ticket, filling a form, launching a workflow) in addition to answering questions
• Sovereign hosting for sensitive data (Mistral on French infrastructure)
• Response quality monitoring, usage dashboards
• Change management support and team training

The upper end (60,000 to 80,000 euros) applies to projects with multiple system integrations, strong security requirements, or very large document volumes (more than 10,000 heterogeneous files). For the full technical cost breakdown of this type of deployment, see our article on RAG project costs from POC to production.

Hidden costs nobody mentions

Internal AI assistant quotes regularly omit several line items. Here is the full picture, so there are no unpleasant surprises.

Document preparation: the consistently underestimated cost

An internal AI assistant cannot work miracles with poor-quality documents. Before development even begins, you need to:

• Audit the corpus: which documents are relevant, which are outdated, which are duplicates?
• Clean up formats: a PDF from a low-quality scan is not treated the same as a native-text PDF. OCR and structuring can represent 10 to 20 days of work on a difficult corpus.
• Validate content: update procedures that are no longer current, remove obsolete versions. This work falls on your business teams, not the vendor.
• Define naming conventions: who names documents, how are they categorized, what is the metadata schema?

On the projects we run at Tensoria, this preparatory work represents between 20 and 35 percent of the total budget depending on the state of the corpus. It is often the first surprise in a project.

Document governance: a recurring human cost

An internal AI assistant is only as reliable as the documents it draws on. If your procedures evolve and are not updated in the system, the assistant will deliver incorrect information. You need to designate:

• A document owner who validates new versions before ingestion
• A clear update process (who updates what, when, and how)
• Periodic quality reviews of responses (how many questions received a wrong answer this month?)

This cost is invisible in quotes but very real in practice. Budget 2 to 4 hours per week of a designated employee to keep an internal AI assistant in good working order over the long term.

User support and onboarding

Adoption of an internal AI assistant does not happen automatically. Employees need to learn how to ask good questions, how to interpret answers with their sources, and how to flag cases where the assistant is wrong.

Plan for at minimum an initial training session (2 to 4 hours for the whole team) and a feedback channel for the first few months. Without this, adoption stays low and ROI does not materialize.

Language model updates

AI models evolve rapidly. GPT-4o will be superseded by something else within 12 to 18 months. A well-built custom solution anticipates these updates, but they always require regression testing and prompt adjustments. Budget 1 to 2 development days per major model update, roughly 2 to 4 times per year.

Exit costs on SaaS solutions

If you deploy a SaaS solution and want to switch in 2 years, how much does migration cost? Is your data exportable in a standard format? Is your documentation locked into the platform? These questions must be asked before signing, not after.

The cost of inaction: what you lose without an AI assistant

Before discussing budget, the most relevant question is often: how much does doing nothing cost?

Calculating the time lost to document search

Industry studies estimate that an office worker spends an average of 1.5 to 2 hours per day searching for information. That is an aggregate figure across all types of search. For document-heavy roles in industrial SMBs and consulting firms, our field experience puts the loss specifically from searching internal documentation at 25 to 45 minutes per day.

Here is the calculation for a team of 15 people losing an average of 30 minutes per day:

A simple custom assistant at 12,000 euros that cuts that lost time in half pays for itself in under 5 months. And that does not account for errors from misretrieved information, extended delays, or the mental energy wasted.

The cost of errors from misapplied procedures

In regulated sectors (construction, industrial manufacturing, healthcare, finance), a procedure followed incorrectly because it was unfindable, or because the wrong version was consulted, can cost far more than an entire AI project: non-compliance penalties, rework, contractual fines, quality incidents.

We have worked with construction companies where a single quality incident linked to a missed procedure cost 15,000 to 30,000 euros in direct costs. In that context, the budget for an internal AI assistant becomes a question of insurance as much as productivity.

The cost of slower onboarding for new hires

A new employee typically takes 3 to 6 months to become fully autonomous with internal documentation. An internal AI assistant significantly shortens that period: it answers basic questions instantly, around the clock, without pulling in an expert or a manager. For an SMB hiring 3 to 5 people per year, the onboarding gains alone can justify the project.

Which option to choose based on your situation

Here is a direct decision framework, based on projects we run in practice.

Choose off-the-shelf SaaS if...

• Your documents are already centralized on SharePoint, Google Drive, Notion, or Confluence.
• You have fewer than 50 users and a reasonable document volume (fewer than 5,000 documents).
• Your data is not highly sensitive and you are comfortable with hosting at Microsoft, Google, or a US vendor.
• You want to test adoption before investing in a custom solution.
• Your IT team is limited and you need a maintenance-free solution.

Main risk: you will be constrained by what the platform supports. If your most important documents are not in compatible formats, you will be paying for a solution that does not address your real need.

Choose a simple custom build if...

• You have a document corpus specific to your business (technical standards, business procedures, project history) that does not fit the generic SaaS use cases.
• You want a solution hosted in the EU, GDPR-compliant, with no data transfers outside Europe.
• You want to embed the assistant in your intranet or an existing internal tool rather than a new standalone interface.
• Your corpus is relatively clean (structured text documents) and of reasonable volume.

This is usually the best entry point for an SMB: custom enough to be genuinely useful, fast and affordable enough to deploy without a 6-month project. Our article on enterprise RAG use cases and ROI illustrates this type of deployment.

Choose an advanced custom build if...

• Your document sources are multiple and heterogeneous (ERP, DMS, emails, scanned documents).
• You need granular access controls (certain users must not see certain documents).
• You want the assistant to be able to trigger actions, not just answer questions.
• Your document volume is large (more than 5,000 to 10,000 documents) and constantly evolving.
• You have strong regulatory constraints (defense confidentiality, health data, sensitive financial data).
• You need contextual access from within your business tools, for instance via a browser-embedded CRM copilot.

At this level of project, we consistently recommend starting with a 2 to 3 day scoping engagement before committing to a quote. Poor estimates in this budget range are expensive. Our AI audit service is designed for exactly this type of scoping.

Questions to ask a vendor before signing

Whether you work with an AI agency or a freelancer, these questions reveal whether the estimate is realistic.

On data preparation

• What is included in document processing? OCR, cleanup, structuring?
• What happens if my documents are poor quality? Is there an additional charge?
• Who is responsible for validating and updating documents once the system is live?

On hosting and security

• Where is my data hosted? In France, in the EU, or in the United States?
• Do my documents pass through the language model provider's servers?
• What happens if the LLM provider's API is unavailable?

On maintenance

• What is included in the annual maintenance contract?
• How are document updates managed?
• How is response quality monitored over time?

For more on choosing a vendor, our article on how to choose an AI vendor covers these aspects in detail.

Summary: budgeting an internal AI assistant realistically

An internal AI assistant is not something you install in a day, but it is not a 200,000-euro project reserved for large enterprises either. For an SMB of 20 to 100 people with a clean document corpus, a simple custom deployment at 8,000 to 15,000 euros is achievable in 4 to 8 weeks, with a visible return on investment in 6 to 12 months.

The three most common budgeting mistakes:

1. Forgetting document preparation: this is often the largest and most underestimated line item.
2. Not budgeting for maintenance: an internal AI assistant that is not maintained degrades quickly.
3. Not calculating the cost of inaction: the right benchmark is not "how much does this cost?" but "how much does this cost compared to what we are losing right now?"

If you want to estimate your specific situation precisely, our article on RAG project costs and TCO gives the development line item breakdown, and our article on AI audit method and cost positions the internal assistant among the different types of AI projects possible.

FAQ: internal AI assistant cost

Further reading

• RAG Project Costs and TCO: detailed development line items from POC to production.
• Enterprise RAG Use Cases and ROI: the most profitable RAG deployments by sector.
• RAG: A Technical Guide: how RAG works end to end, embeddings, chunking, vector stores, and where it breaks.
• RAG vs Simple Chatbot: when to use retrieval and when a rule-based chatbot is enough.
• Production RAG Failure Modes: the recurring failure patterns when RAG ships to production and how to fix them.
• Self-hosted RAG Architecture: when the unit economics of cloud RAG push you toward running your own stack.
• Multimodal RAG: handling PDFs with tables, figures, and scanned forms in a RAG pipeline.
• AI Audit Method and Cost: how to frame an AI project before committing to a budget.
• How to Choose an AI Vendor: the questions that reveal whether an estimate is realistic.
• Our RAG systems service: end-to-end RAG deployment including ingestion, eval infrastructure, and observability.
• AI audit: structured review of your AI use case to recommend the right architecture before you build.