Logic Builder · Chatbot builder SaaS · Onboarding
How we taught the product
to teach users
No calls, no demos, no team in the room. I redesigned onboarding for a no-code chatbot builder, so new teams could learn a complex tool on their own.
Role
Senior Product Designer.
Design decisions were mine,
in sync with the design team.
Team
Ops team of 3 and engineers. Weekly syncs, hallway tests.
Scope
Research, onboarding system, UX, voice and motion
Almost every new customer needed several live onboarding calls, sometimes 4 to 6 hours in total. New users opened the builder, saw an empty canvas and didn’t know where to start.
Make onboarding part of the product, so non-technical users could learn a complex tool on their own.
I ran the research, set the principles and designed the onboarding system end to end: clearer settings panels, voice-guided tours, an AI starting point and team templates. I also produced the voice and motion myself.
Half as many repeated onboarding calls, 30% fewer onboarding support requests, and 30% more first flows in the first week.
01
Context
Customers grew faster than onboarding could
Logic Builder is a tool for building support chatbots without code. You build a chatbot as a flow: a chain of steps on a canvas. Each step is called a node.
It started as an internal tool built by engineers. Then it became a SaaS product for other companies, used every day by non-technical ops and support teams.
As more customers came in, onboarding stopped scaling. New users saw an empty canvas and unfamiliar logic and didn’t know what to do first. Almost every customer needed several live calls with our team, sometimes 4 to 6 hours in total.
My goal was to turn onboarding into part of the product, so people could become independent without our team in the room.
02
What we learned
People got lost before their first flow
Before designing anything, I wanted to know where people got stuck and why our team spent so many hours helping them. I collected signals from four places.

Four sources, one pattern: people got stuck before they built their first flow.
FINDING 01
Problems start early
Most onboarding requests came in the first sessions, before people had built anything.
FINDING 02
Structure was the barrier
People struggled with how flows and logic fit together, more than with single buttons.
FINDING 03
Engineering language
Many labels and node settings used terms that non-technical users didn’t know.
FINDING 04
Text tours got skipped
The content was clear. People just didn’t read long text in tooltips.
03
The hypothesis
One learning system, not more help screens
The obvious fixes were all about adding: better tooltips, smarter tours, ready-made content. Each one had a reason to say no. The real problem was not missing help. It was too much complexity, too early.
THE QUESTION
How do we get people to independence without onboarding calls?
IDEA A
Better tooltips
Wouldn’t fix skipping. It was an attention problem, not a clarity problem.
IDEA B
AI-personalized paths
Would hide the structure instead of teaching it. People needed the mental model first.
IDEA C
Ready-made templates
Every customer had its own logic and brand voice. One set wouldn’t fit them all.
WHAT WE DID INSTEAD
Chosen
One learning system inside the product
Clear settings, voice tours, an AI start and team templates, working together. The product teaches step by step, inside the builder, instead of in calls and help articles.
WHY
Each idea on its own fixed one symptom. Together they could replace the calls: people learn by doing, in context, at their own pace.
04
Four principles
How I made onboarding decisions
Designing onboarding for a complex product means many small decisions. To keep them consistent, I worked from four principles. Most decisions below trace back to one of them.
01
A system, not a moment
First-time tours fade fast. Help should be there whenever people need it, not only on day one.
SHOWS UP IN
Voice tours
Team templates
02
Respect how people learn
Some learn by doing, some explore, some skip ahead. Every path should work.
SHOWS UP IN
Three ways to start
03
Remove before you add
Every tooltip, animation and label has to solve a real problem. If it doesn’t, it goes.
SHOWS UP IN
Settings panels
04
The product teaches itself
If people have to leave the product to understand it, something is broken.
SHOWS UP IN
Voice tours
Start with AI
05
What we built
Four parts of one learning system
Each part fixes a different version of the same problem: the fear of starting in a complex product. Together they cover the first session, the first week and every new flow after that.
01
Clear settings
Settings panels in plain language
02
Voice tours
Guidance people actually follow
03
Start with AI
A draft instead of an empty canvas
04
Team templates
Reuse what already works
FIRST SESSION
FIRST WEEK
EVERY NEW FLOW
1
Settings panels people can read
In usability tests I ran, people got stuck the moment they opened a node, not on the canvas. The settings panels showed system logic in engineering terms. Adding tooltips on top wouldn’t help, because the structure itself was confusing.
So I rebuilt the panels with clearer labels, grouping and hierarchy. Boolean operators became business rules, and field paths became data sources. The interface started to speak the user’s language, not the system’s.

The Option Select node settings, before and after.
2
Voice-guided tours
I split onboarding into small steps that show up in context, when they become relevant. People choose how to start: a guided tour, building with AI, or exploring on their own.
The first tours were text only, and people skipped them. The problem wasn’t the content, it was attention. So I added AI voice narration with a Mute button. Same content, but now people followed the tour instead of clicking through it.
1
A short tip appears when it becomes relevant
2
The voice reads it out, so people notice it
3
Mute or skip the tour at any time

The same tip, before and after. Voice turns a tip people skip into one they follow.
OWNERSHIP
Voice wasn’t in the original scope, and the client had no budget for voice production. I tested AI voice tools, built a working prototype myself and helped the team ship it without an outside studio.
Producing the motion myself
Animated onboarding usually needs motion designers, voice actors and editors. We had none of that, so I built a light workflow myself.
01 · FIGMA
Motion
Every animation prototyped next to the real UI.
02 · AI TOOLS
Voice and sound
Narration and sound effects generated with AI.
03 · SHIP
Product and help center
The same files go into the product and the knowledge base.
When the UI changed, onboarding was updated the same day, without waiting for a production cycle.
The onboarding tour with AI voice. Turn the sound on. Motion, voice and sound were made in-house.
3
Start with AI, not an empty canvas
Even after the tour, the empty canvas was still a barrier. People understood the system but didn’t know what to build first. The product already had an AI Copilot, but it was treated as a secondary feature. I moved it into the first screen and made Build with AI one of three main ways to start, next to Guide me and Explore on my own.

1
Explore on my own
An empty canvas, for people who learn by doing.
2
Build with AI
Copilot drafts a first flow to edit.
3
Guide me
The voice tour, one short tip at a time.
The first screen of a new flow: three ways to start, with AI as one of the main options.
Copilot itself, and how it builds a whole flow, is a separate case: Adding AI where it really helps →
4
Templates teams build themselves
The request sounded simple: add templates. But every customer worked differently, with their own processes, chatbot logic and brand voice. No single set of templates would fit everyone.
So instead of making templates for customers, I pushed for letting teams create and save their own. One person builds a flow that works, and the next person starts from it instead of from scratch.

1
Grouped by topic
Billing, devices, plans: find a starting point fast.
2
Author and date
Every template shows who made it and when.
3
Preview and AI summary
See the flow and a short summary before using it.
4
Your templates
The team saves its own flows for everyone.
Teams save their own flows as templates and start new ones from them.
It didn’t fix the first session right away. But over time teams stopped facing an empty canvas for every new flow, and one person’s work became the whole team’s starting point.
06
Outcome
What changed
ONBOARDING
repeat calls
BEFORE
Several live calls per customer, up to 4 to 6 hours.
AFTER
Voice tours and in-product help do the teaching.
−30%
SUPPORT
support requests
BEFORE
Questions about node settings and where to start.
AFTER
Settings in plain language and a guided start.
ACTIVATION
first flows, week 1
BEFORE
An empty canvas and no clear first step.
AFTER
Start from an AI draft or a team template.
−35%
OPERATIONS
team load
BEFORE
Hours of calls for every new customer.
AFTER
Onboarding that runs without us in the room.
Based on ops team tracking and product analytics, 3 months after launch.
07
Reflections
Less complexity, not more explanations
The hardest part wasn’t designing onboarding screens. It was seeing that nothing was missing from the explanations. The product simply showed too much complexity too early.
So instead of adding more onboarding on top, I reduced friction in the product itself: simpler settings, guided attention, and an editable starting point instead of an empty canvas.
What I’d do better: measure onboarding from day one. We had no clear tracking of where people dropped off, so many decisions relied on observation and support feedback instead of behavior data.
This wasn’t really an onboarding project. It was a complexity reduction project, at the moment people first meet the product.

