How much does an AI that knows your documents cost?

"And how much does it cost?"

The second question, right after ROI. And the honest answer is: it depends on what you count. The sticker price — a licence, a subscription — is never the real cost of an AI connected to your documents. Here are the lines that make up the true bill, the three ways to do it, and the one cost that can outweigh all the others.

The sticker price isn't the real cost

Whatever the tool, the total cost has four parts — only one is on the quote:

• The licence or subscription. The visible line. Often the smallest.
• Data preparation. Connecting your sources, cleaning, organising. The effort no one prices, and where most projects slip.
• Adoption. Training, support, the time for the tool to become a habit. A tool paid for but barely used is expensive for nothing.
• Upkeep. A company's knowledge changes constantly; without updates, quality — and therefore value — decays.

Remember this: two solutions at the same subscription price can have a total cost that differs threefold, depending on what they demand across those other three lines.

Three ways to do it (and what they really cost)

1. ChatGPT or Claude, on their own

The lowest entry ticket: a few tens of euros per user per month. But these tools don't know your documents: you paste the context every time, answers don't cite their sources, and nothing guarantees the confidentiality of what you paste. The price is low; the cost of generic or wrong answers is not. (See why your AI hallucinates.)

2. Build your own RAG

Technically doable, and appealing on paper. But it's a development project, not a subscription: source integration, a preparation pipeline, an interface, security, then lifelong maintenance. The real cost is counted in engineer-days — the most expensive line in the company — and it never stops. Relevant for a technical team with a very specific need; expensive and slow for everyone else. (See three ways to connect an AI to your documents.)

3. A turnkey platform

A subscription, but one that absorbs data preparation and maintenance: connect sources in a few clicks, sourced answers, automatic updates. The cost shifts from engineering to a predictable subscription plus an adoption effort. It's the option that makes the total cost readable upfront — exactly what a committee wants.

Orders of magnitude

Let's be concrete without lying: cost depends mostly on the number of users and the state of your data. A small team on an already-clean base starts at a few hundred to a few thousand euros a year. A larger organisation, or messy data, means an additional preparation budget. At Ragnight, pricing is public and transparent — no imposed hidden setup cost.

And to turn those costs into a decision, set them against the gains: that's the whole point of calculating AI ROI in the enterprise.

The cost that outweighs all the others

There's a line that appears on no quote and can cost more than everything else combined: a wrong answer, at scale. A quote at the wrong price, a customer answer that contradicts policy, a clause misquoted in a contract — multiplied by hundreds of uses a day. An AI that doesn't make things up and cites its source isn't a luxury: it's the line that protects all the others.

And that reliability doesn't depend only on the tool: it depends on the quality of your data. Hence the step that controls the bill before you even buy — auditing your knowledge base.

How to keep the bill under control

• Start small. One team, one use case, measure, then expand.
• Audit first. Knowing the state of your data avoids paying to prepare what you won't use.
• Track adoption. It's what turns a cost into a return.
• Demand pricing transparency. A predictable total cost beats a teaser price followed by surprises.

In short

The right instinct isn't to chase the lowest price, but the most readable total cost — licence + preparation + adoption + upkeep — and to set it against what a reliable AI saves you. The rest is just a line on a quote.

See our pricing: transparent, with no hidden setup cost, to estimate your real budget.