Which AI Solution Fits Your Task
Answer about a single task. The quiz doesn't sell technology — it shows which of the four approaches reaches a result more cheaply.
What gets in the way most?
Why the choice of class matters more than the choice of model
"Which model should be used" is the first question asked and the last one that matters. Models change every few months, and swapping one for another in a finished system is a day's work. The solution class changes everything: timelines, budget, team composition, and what comes out the other end at all.
By and large there are four classes.
An assistant answers questions from your documents. It launches fastest and costs least, but it does nothing — it only talks.
Document processing turns paper into data. The most predictable class in money terms: the volume is known, the time per document is known, and the savings come out of simple arithmetic.
An agent runs the process end to end, from event to result. The most valuable and the most demanding: without access to the systems and written-down rules it doesn't work.
A private perimeter isn't a class of task, it's a decision about where things run. It's made before the others, because it determines the architecture and the infrastructure budget.
What makes the choice wrong
Three mistakes repeat more often than the rest.
The first is starting with an agent. A full process looks like the best deal on paper and turns out to be the longest in practice. An agent is assembled from pieces that already run smoothly; going straight for the whole thing is a reliable way to spend the budget on rework.
The second is taking a private perimeter without needing one. The requirement comes from internal policy more often than from law. The difference in infrastructure cost is roughly double, and it's worth paying deliberately.
The third is buying development instead of strategy while the task is still undefined. A contractor will build exactly what was asked for, and it will work. Whether it was needed is another question.
What this quiz doesn't do
It doesn't count money — there's a budget calculator for that. It doesn't check whether the process is ready for automation — there's a separate quiz for that, and it makes more sense to take it before choosing a class.
And it doesn't replace a conversation about your task. Six questions give a direction; the specifics appear once your documents, your systems, and your constraints are on the table.
FAQ
Can several solutions be combined?
That's how it turns out almost every time. Document processing becomes part of an agent, and the assistant draws on the same base. The quiz shows where to start: the class that pays back first usually becomes the foundation for the rest.
How is an AI assistant different from a button-driven chatbot?
A bot walks you down a tree drawn in advance and gets lost at the first question that's off-script. An assistant understands the wording as it comes and answers from your documents. The practical difference shows up in the share of requests that never had to be handed to a person.
What is RAG in plain words?
A way to make a model answer from your documents instead of from general knowledge. The system first finds the relevant fragment of a policy, then formulates an answer from it. That's why the answer can be checked: its source is visible.
Is an in-perimeter model really necessary
Less often than it seems. The requirement usually comes from an internal security policy rather than from law. Check the wording: if it's about personal data, a Russian cloud is often enough, and that's half the cost.
What if none of the options fit?
Then the task isn't fully described yet — and that's a normal stage, not a dead end. Working the task out takes one conversation, after which the solution class is usually obvious.
Check it on your case?
We will look at your process and give numbers for it, not for averages.
Related tools
Updated: July 26, 2026