What does an AI assistant for employees cost?
An internal AI assistant that answers based on the company's own documents and systems typically costs SEK 300,000–1.2 million to build. Permission control, making sure each employee only gets answers from data they're allowed to see, is the priciest and most important part. Always weigh the math against ready-made licenses like Copilot.
An internal AI assistant – a service where employees can ask about everything from vacation policy to project documentation and get answers from the company’s own knowledge – is one of the most sought-after AI initiatives right now. It shouldn’t be confused with the customer service chatbot: here the users are your employees, the knowledge is internal, and the hard question isn’t what the assistant can answer, but who gets to see which answer. The price tag for a custom-built solution is typically SEK 300,000–1.2 million.
The pricing, and what drives it
| Level | Typical cost | Characteristics |
|---|---|---|
| Scoped assistant | SEK 300,000–500,000 | One knowledge source with open access, e.g. the employee handbook and policies |
| Expanded assistant | SEK 500,000–800,000 | Multiple sources, basic group-level permission control |
| Full-scale assistant | SEK 800,000–1,200,000 | Many sources and system data, document-level permissions, logging and tracking |
Permissions are the priciest, and most important, part
What sets an internal assistant apart from a simple document chat is permission control. A company’s knowledge isn’t a uniform mass: payroll data, personnel cases, board materials, unpublished contracts, and ongoing negotiations sit mixed in with handbooks and process descriptions. The assistant has to answer every employee using only the data that person is personally allowed to see.
Technically, that means a connection to your identity solution, permission filtering on every single search, and making sure the source systems’ access rules carry through into the AI layer, even when documents move or rights change. It’s often the single largest work item in the project, and you can’t cut corners on it: an assistant that shows the wrong pay slip to the wrong person even once is dead as an internal tool, and a data protection incident on top of that.
So ask every vendor this question: how does the solution guarantee that answers respect per-user permissions, and how is that tested? The answer says more about how serious the quote is than the price does.
Build it yourself, or buy licenses? Do the breakeven math
The alternative to building is ready-made products like Microsoft 365 Copilot, where the license typically costs on the order of SEK 300–400 per user per month, roughly SEK 4,000 per employee per year. A simplified comparison over three years:
| Headcount | Licenses, ~3 years | Custom-built, ~3 years |
|---|---|---|
| 50 employees | ~SEK 600,000 | ~SEK 1,000,000–1,600,000 |
| 150 employees | ~SEK 1,800,000 | ~SEK 1,000,000–1,600,000 |
| 400 employees | ~SEK 4,800,000 | ~SEK 1,200,000–2,000,000 |
The math assumes a build cost within the ranges above, plus operations and maintenance at SEK 10,000–30,000 a month. The breakeven then often falls around 100–150 employees: below that, licenses are hard to beat on price; above it, the custom build gets cheaper with every new employee, since your own operations don’t scale per license. Run the math with your own numbers before deciding – the number of actual users is often lower than the headcount.
But the comparison isn’t purely financial. License products are strongest within their own environment (documents, email, calendar), while a custom-built assistant can reach your business systems, follow your rules in detail, and be swapped out in step with the model market. Many companies land on a combination: licenses for office work, a custom assistant for business knowledge.
How to get started without overinvesting
Start narrow: pick a body of knowledge with clear ownership and broad access – the employee handbook, IT guides, process documentation – and launch it to a pilot group. Measure usage and the questions that don’t get good answers; they show exactly which sources should be connected next. The permission integration can then be sized to real needs instead of theoretical ones.
Also plan for the assistant needing an internal owner: someone who follows the logs, sees which questions keep coming up without good answers, and pushes to keep the sources current. An assistant with no content owner loses trust within six months, not because the technology stopped working, but because the answers slowly became outdated.
At Weapp we build internal AI solutions with permission control as a first-class citizen in the architecture. Want help with the breakeven math for your headcount? Get in touch and we’ll run the numbers for your situation.
Frequently asked questions
What's the difference from our customer service chatbot?
The customer chatbot faces external users and answers within a narrow subject area under tight control. The internal assistant faces employees and needs to cover broader knowledge, policies, processes, system data, where permissions determine who gets to see what. They're different solutions with different risk profiles, even though the technology is similar.
Isn't Microsoft Copilot enough?
For many companies, Copilot or something similar is a good first step, especially if the knowledge already lives in the Microsoft environment. A custom-built assistant is justified when the knowledge lives in your own systems, when you want to control behavior and permissions in detail, or when the license cost at a large headcount exceeds the build cost.
How does the assistant handle sensitive documents?
By inheriting the permissions from the source systems: the assistant only searches documents that the employee asking is themselves allowed to open. That requires a connection to your identity solution and filtering on every single search. Test this rigorously before launch – a single leaked pay slip destroys trust.
What value does an internal AI assistant deliver in kronor?
The most common effect is shorter search time: employees find answers in minutes instead of digging through systems or asking colleagues. Measure time-to-answer for common question types before and after, plus the relief on internal support functions like HR, IT, and finance, and you'll have a concrete impact calculation.
How long does it take to roll out an internal AI assistant?
A scoped version against a well-structured knowledge source often takes 2–3 months. A full build-out with multiple sources and permission integration takes 4–8 months. The wisest approach is to launch narrow, one department, one body of knowledge, and expand in step with usage.