How to choose an AI onboarding agent

“AI onboarding” covers a lot of different products now. Some answer questions, some build tours, some do the setup for the user. They can all be a good choice, for different problems, so the first job is to be clear about which problem you have.

This is a checklist for that choice: the kinds of AI onboarding help, the questions worth asking any of them, and how Stiro answers each one, so you can hold it to the same standard.

Three kinds of AI onboarding help

1. Answers from your docsTells the user how, in plain words
2. AI-generated toursShows the user where to click
3. Agents that do the stepsDoes the setup while the user watches

Answers from your docs. A chat assistant reads your help centre and answers questions with a link to the source. Good for “what does this setting do?” and “do you support SSO?”. The user then does the steps, and the answers are as current as the docs behind them.

AI-generated tours. AI helps you write and target walkthroughs, tooltips and checklists faster. Good for showing a new user around the main paths, especially when you want a consistent first look. The user still clicks, and the tours follow the paths someone set up.

Agents that do the steps. The agent works inside your app, reads the live screen and does what the user asked: creates the project, fills in the form, changes the setting. Good for the “set this up for me” moments across the whole product. It moves through real pages, so it is slower than a single answer, and because it can press real buttons, it has to know when to stop.

Plenty of products use more than one. A docs assistant for questions and an agent for doing sit together well.

Start from where new users get stuck

Before you compare tools, look at your own onboarding:

• If new users mostly ask questions your docs answer, answers from docs fit.
• If they mostly don’t know where things are, a tour helps.
• If they know what they want but stall doing it, such as connecting data, inviting a team or configuring a setting, an agent that does the steps fits.

Your support queue from users’ first week is the quickest way to tell. The AI customer success guide covers how to read it.

The buyer’s checklist

Whichever kind you look at, these are the questions that matter once real users meet it.

What can it touch?

It should only reach what the signed-in user could reach in your app, through your own permissions. Ask how it is limited to your domains, and whether a copied key could be used elsewhere.

What does it do when it can’t?

Ask what happens when a user requests something your app doesn’t have, or a record that doesn’t exist. “That isn’t possible here” is a good answer. Guessing, or changing the wrong thing, is the costly one.

What does it hand back?

Some actions should always be the user’s call: sending an email or invite, spending money, deleting anything. Ask which actions it leaves for the user to press.

What happens when it’s stuck?

A stuck new user should reach a person, with the context of what they asked. Ask where that message goes and whether it can go to the help desk you already use.

What does set-up and upkeep take?

Ask what you have to build: flows to record, content to write, tools to code per action. Then ask what happens when you ship a redesign. Upkeep is where onboarding tools cost the most over a year.

What does it see, and where does it go?

Ask what data leaves the browser, whether passwords and card numbers are kept out, whether the AI providers train on it, and what your dashboard shows of each request.

How was it tested?

Ask whether it was tested on apps it had never seen, and how results were checked. A reply that says “done” is easy to produce; a change in the database is not.

How is it priced?

Per seat, per message and per task look very different once real users arrive. Ask what you pay when it fails, and what happens at the plan limit.

How Stiro answers the checklist

Stiro is an agent that does the steps. Here is how it answers each question, from what the site already says.

Question	Stiro’s answer
What can it touch?	Only what your signed-in user could do. Each site’s key works only on the domains you list, and you can accept only user ids signed on your server.
When it can’t	It stops, tells the user why and changes nothing.
What it hands back	Anything that sends, spends or deletes is set up and left for the user to press. They can stop it at any moment.
When it’s stuck	A signed-in user can message your team from the same panel, with the request attached, or one line sends it to your own help desk.
Set-up and upkeep	One script tag, nothing to train and no flows to record. It reads the live screen, so a redesign doesn’t break it.
Data	Password, card and ID fields it recognises never leave the browser. The AI providers it uses never train on it, and your dashboard shows requests with filled-in values replaced by placeholders.
Testing	Cold, in three open-source apps it had never seen, every result checked in the app’s database.
Pricing	Per task done. A run that fails on Stiro’s side is free, messages to your team are free, and every plan stops at its tasks instead of billing past them.

The test results are on the benchmark page:

everyday requests passed, cold, in three apps it had never seen

44 of 44

on a repeat run of the same code

43 of 44

median time from request to reply

19.5 s

And the prices are on the pricing page: the free Developer plan has 50 runs a month on localhost and a staging address, Starter is $149 a month for 2,000 tasks ($0.075 a task), and Growth is $599 a month for 10,000 tasks ($0.06 a task).

Try the checklist on your own app

The surest test is your own product with your own setup flow. Put Stiro’s AI onboarding on a staging address, ask it for the steps your new users stall on, and check the results where your app stores them. The free plan is enough to run the whole checklist before any user sees it.