Build vs buy: should AI build your product tours?

AI can write a product tour in an afternoon. What it can’t do is own one: the editor your team needs, the targeting, the analytics, and the fixes every release brings. Here is what building really costs, a calculator to check it with your own numbers, and when building is the better choice.

The short answer

Build if you need one or two fixed walkthroughs that rarely change and only engineers will edit. Buy if product, marketing or customer success will write, target, test and fix guidance themselves. AI makes the first version cheap; it doesn’t make the hundredth change cheap. Cueston’s Launch plan costs $468 a year, about 5 hours of one engineer’s time at US rates.

What AI gets you in an afternoon

Ask Claude, Cursor or ChatGPT for a product tour and you get a working one: a popover pinned to a button, Next and Back, a dimmed backdrop, and a flag in localStorage so it doesn’t show twice. Or it wires up an open-source library such as Driver.js or React Joyride. It takes an afternoon, and it looks finished.

Check the licence before you pick a library. Shepherd.js and Intro.js need a paid commercial licence in a commercial product; Driver.js and React Joyride are MIT. All four draw the tour. Everything around it is still yours.

That first tour isn’t where the cost is. The cost is in what a tour needs once it has real users, and in every change after that. Each row below is something you’d write, test and keep working yourself.

What you’d still have to build

What it’s for	If you build it	In Cueston
Changing a tour without a release	Copy, steps and targeting live in code. Every change is a ticket, a review and a deploy.	Your team edits in the dashboard and picks elements on your real site with the Chrome extension. Changes go live when they publish.
Remembering who saw what	localStorage forgets on a new browser or device. Doing it properly takes a table, an API and your user IDs.	“Show once” holds across browsers and devices for users your app identifies, with frequency options for every item.
Showing it to the right people	Role, plan and page checks written into each tour.	Audience rules and saved segments: user properties, pages, events in the last 90 days, whether someone finished another guide.
Not piling up pop-ups	Two tours and a banner fire on the same page, and you write the rules for who goes first.	A priority on every item, plus workspace limits: most per visit, most per day, minutes between.
Knowing whether it worked	Events for views, steps, completions and dismissals, then a dashboard to read them.	Funnels for every tour and checklist, and goals with a control group of up to 50%, reported at 95% confidence.
Noticing when it breaks	A release renames a button, the step stops showing, and nobody knows until a user asks.	A broken targets report lists every step whose element went missing, and admins get a daily email.
Shipping changes safely	Preview, staging and rollback are whatever your deploy pipeline gives you.	Drafts, preview on your own site, staging, approvals, and rollback to any earlier version.
Other languages	Every step’s text goes into your translation files, and someone keeps them in sync.	11 languages, machine translation as a first draft, and CSV or XLIFF files for your translators.
More than tours	Checklists, surveys, NPS, announcements and a help launcher are each a new project.	All of them in the same editor, with the same targeting and analytics.
Keeping styles apart	Your app’s CSS restyles the tour, or the tour’s CSS leaks into your app.	Guidance renders in its own Shadow DOM, so neither side’s styles reach the other.
Deciding who can change what	Anyone with access to the repository.	Roles, custom roles, an audit log, and single sign-on with SCIM.

What building costs, with your numbers

Change any number. The defaults lean towards building: they assume AI writes most of the code and leave out work most teams end up doing anyway.

Your estimate

First year

Building it$16,368

Cueston Ascent$948

Over three years

Building it$34,224

Cueston Ascent$2,844

Building costs 17 times as much in the first year. Cueston Ascent costs as much as 10 hours of engineering a year.

Cueston prices are list prices billed yearly. See pricing for monthly billing and what each plan includes.

Where the 80 hours come from

A first version with basic targeting, an admin screen and analytics, with AI writing most of the code:

Tour, tooltip and banner components that work on desktop and phone widths	8 h
Saving who saw, finished or dismissed what, per user, across devices	8 h
Targeting by role, plan and page	8 h
An admin screen so people outside engineering can edit copy and steps	24 h
Events, and a dashboard for views, completion and drop-off	12 h
Priority and limits, so pop-ups don’t pile up	6 h
Testing across browsers, route changes and your app’s CSS	14 h
Total	80 h

Not counted: translations, experiments with a control group, approvals, version history, checklists, surveys, a report of broken steps, the product manager’s and designer’s time, and what those engineers would have built instead.

Where the 8 hours a month come from

Repointing steps after releases move or rename elements	3 h
Copy, targeting and timing changes other teams ask for	3 h
Library, framework and browser updates	2 h
Total	8 h

Maintenance: the cost that keeps coming

A tour points at elements on your page, and your product keeps changing. So a homegrown tour breaks quietly, a step at a time, and the work lands on whichever engineer is free that week.

A redesign renames a button. The step that pointed at it stops showing, and nothing tells you.
Customer success wants one sentence changed. It waits for the next sprint.
Marketing wants a banner for paying customers only, starting Tuesday. That’s targeting code, and a release on Tuesday.
The engineer who prompted it leaves. Nobody else knows how the AI-written code works, and nobody wants to touch it.
Someone asks whether the tour helps activation. There’s no control group, so nobody can say.

AI helps write each of these fixes. It doesn’t notice that they’re needed, and it doesn’t ship them without an engineer.

When building is the right call

You need one or two fixed walkthroughs, they rarely change, and only engineers will edit them.
The guidance is really product UI: a setup wizard that creates records, or steps tied to your backend. That’s product work, not a tour.
Your security rules don’t allow third-party scripts in your app. Cueston works through a script on your page.
You need guidance in a native iOS or Android app. Cueston is for web apps only.
A team wants to own onboarding as a product, and has the time to keep it working for years.

Keep the AI, skip the upkeep

If what you like about building is asking an AI and getting a tour back, keep that part. Connect Claude, ChatGPT or Cursor to Cueston’s MCP server and ask in plain language. The assistant looks at your pages, builds a draft with selectors it has checked, and sends you a preview link. Nothing goes live until someone on your team publishes it, and every action it takes is in the audit log.

Paste into Claude or ChatGPT

Use the Cueston tools. Build a five-step tour of our invoice flow for new admins. Check every selector on the real page, then send me a preview link. Don’t publish.

See Cueston with ChatGPT, Claude and Cursor for the setup.

Frequently asked questions

Can’t I just ask Claude or ChatGPT to build our product tours?▾

Yes, and for one fixed walkthrough that’s a fine choice. What you get is the tour itself. Changing it without a release, targeting, remembering who saw what across devices, analytics and repointing steps after redesigns are still yours to build and maintain.

How much does it cost to build product tours in-house?▾

With AI writing most of the code, expect about two working weeks for a first version with basic targeting, an admin screen and analytics, then about a day a month to keep it working. At $93 an hour, US median developer pay plus benefits, that comes to about $16,400 in the first year. Cueston’s paid plans cost $468 to $1,668 a year.

Are open-source tour libraries free?▾

Not all of them, for a commercial product. Shepherd.js and Intro.js are free for open-source and non-commercial use and sell commercial licences. Driver.js and React Joyride are MIT-licensed. All four draw the tour; targeting, analytics, editing and saving state are still up to you.

How long does Cueston take to set up?▾

A developer adds one script to the head of your app, plus an identify call if you want to target by role or plan. After that your team builds and publishes guidance without code. Single-page apps are detected automatically.

Will Cueston slow down our app?▾

The script loads asynchronously, so it doesn’t block your page from loading. Guidance renders in its own Shadow DOM, so it doesn’t restyle your app.

We already built our own tours. Can we move them to Cueston?▾

There’s no import. Teams rebuild their tours in the Chrome extension by clicking the elements each step points at, then remove the old code once the Cueston version is live.

Is there a free plan?▾

Yes. The free plan has one live item of each type: a tour, a tooltip, a banner, a checklist, a survey and an announcement. You can try it on your own app before paying.

Try it before you build it

Start on the free plan, put a tour on your own app, and compare it with what you’d build.

Start free