From Signup to Running Project in Under 2 Minutes: Inside Onplana's AI Project Kickstart

We used to think the hard part of adopting a project management tool was migrating data from the old one. Then we watched twenty new-signup sessions.

The hard part isn't migration. It's staring at an empty project template and realizing you have to create every task, every dependency, every timeline estimate from scratch before the tool does anything useful. That's a two-hour commitment before you've felt a single benefit. Most people don't finish. They open another tab, get pulled into a meeting, and the trial account goes cold.

We decided the first two minutes after signup should feel like magic, not homework. This is how we got there.

The problem, concretely

Every PM tool greets you the same way: an empty canvas and a dozen configuration dropdowns. That works if you already have a project plan in your head — the tasks, the sequence, the rough durations. If you don't, the empty canvas is paralyzing. You came here for the tool to help you structure the work, and the tool's answer is: you structure it first, then we'll help.

Even users who can sketch the structure end up burning their Tuesday evening on data entry. Type a task, pick a priority, set a date, add an assignee, repeat thirty times. By the time the Gantt chart has enough data to be interesting, the novelty is gone.

And the aha moment for a PM tool — seeing your timeline, your dependencies, your critical path materialize — only happens once there's populated data. You need the payoff to justify the input, but the input is what's blocking the payoff.

Every PM tool company we've talked to says the same thing, in some form: first-session engagement predicts retention more than feature depth. If users don't create five or more tasks before they leave the first time, they usually don't come back.

The solution: AI Project Kickstart

The new /onboarding flow has three steps and a single goal: get you to a populated project before you have time to second-guess signing up.

Step 1 — Welcome. One screen, a role-aware subheading, and preview cards showing what Gantt, Kanban, and the AI menu will look like once there's something to put in them. It's there to orient, not to collect.

Step 2 — Your organization. The org name is pre-filled from your email domain (@acme.com becomes "Acme"), with opt-in team invites. Consumer-email signups skip this step entirely. Average time on this screen: under ten seconds.

Step 3 — Your first project. A single textarea, five preset chips ("Launch a mobile app", "Marketing campaign", "CRM migration", "Office relocation", "Quarterly OKRs"), and one button: Create with AI.

Behind the scenes, clicking that button does two things. First, POST /api/ai/intake sends the freeform description to the configured AI provider with a strict JSON schema — project name, summary, estimated duration, epics, tasks with subtasks, milestones, and any risks the model can identify from the description. Second, POST /api/projects/from-intake takes that structured output and atomically creates the project, epics, tasks, subtasks, milestone markers, and risk entries in a single Prisma transaction.

End to end, from clicking the button to landing on a fully populated project, takes eight to twelve seconds.

Here's a concrete example. Paste this into the textarea:

"Launch a new mobile-first iOS and Android app for small-business customers over 12 weeks. Scope: user authentication with Google/Apple sign-in, offline data sync, push notifications, in-app payments via Stripe, and a simplified 3-step onboarding flow. Private beta with 20 customers at week 8, public GA at week 12."

The output, on a typical run: three epics (Discovery & Design, Build, Launch Readiness), about twelve tasks with subtasks grouped under the right epics, milestones at week 8 (Private Beta) and week 12 (Public GA), and one risk flagged against third-party payment SDK integration timing. It's not perfect — you will still prune a couple of tasks and sharpen a few estimates — but it's a scaffold, not a blank page.

What happens when AI isn't available

The flow has to work when the AI provider doesn't. Three failure modes, all handled:

• Quota exhausted (402 from the provider) — the UI falls back to a curated template picker with five hand-crafted templates: Software Launch, Marketing Campaign, Event Planning, Microsoft Project Online Migration, and OKR Quarterly Planning. Each is a real, opinionated plan, not an auto-generated one.
• Provider unavailable (503, network error) — same template picker, plus a "Load sample project" button that seeds a fully-built demo project so the user has something to poke around in while they wait.
• Timeout (30 seconds, which we've almost never seen in practice) — same fallback.

The fallback was designed to feel like a first-class option, not a consolation prize. A non-trivial share of users pick a template even when AI is available — sometimes they know exactly what they want and "Event Planning" is faster than describing an event in prose.

Under the hood

A few engineering details that matter for reliability:

• We use Azure OpenAI's response_format: json_object so the model returns parseable JSON by construction. For the Anthropic provider we wrap the response in a tolerant parseAiJson that strips code fences and handles the handful of edge cases where the model adds a preamble.
• Project creation is a single Prisma transaction. If any step fails — bad field on a task, an invalid enum, a transient DB error — the whole thing rolls back. No orphaned epics or half-built task trees to confuse the user.
• We defensively coerce the AI output before writing it. Estimated days are clamped to 1–180. Dates fall back to today and today+30 if the model hallucinates something weird. Priority is forced to the enum set. Task names are trimmed and length-capped.
• Timesheets, when they start landing, snapshot the current rate card so future financial reports stay consistent even if rates change later.

None of this is glamorous. All of it is why the flow doesn't blow up in the 1% case.

Beyond the first two minutes

The setup doesn't end when the project is created. A persistent OnboardingChecklist sits in the corner of the app for the first week, with role-specific next steps — invite your team, configure project permissions, connect your calendar, set up your first integration. It never pops up modal-style and never blocks the UI.

A short product tour, built on react-joyride, highlights Gantt, Kanban, and the AI menu the first time they're opened. It's dismissible, it never repeats, and every step has a "skip for now" link that actually skips it.

The goal throughout is the same as the first two minutes: show the value before asking for the work.

Try it

Sign up at app.onplana.com — free, no credit card, no sales call. Google, Microsoft, or email all land in the same flow.

This is especially relevant if you're on a team migrating off Microsoft Project Online ahead of the September 2026 retirement. We built AI Project Kickstart, in part, to make switching fast enough that "we don't have time to evaluate alternatives" stops being a defensible excuse.

Try AI Project Kickstart →

For the full running list of what we've shipped this quarter, see What's New.