Case Study · 2026

AI avatar setup that checks itself.

Connecting an AI avatar meant juggling an image spec, domain whitelisting, and a script embed, with no way to tell if any of it was right until the end. I designed a guided four-step wizard with real-time validation and a live preview, so users see the avatar is correct before it ever goes live.

Role
Solo Product Designer
Scope
Flows · Validation · UI
Type
Product design
Platform
B2B SaaS · web app
AI Avatar setup with a four-step wizard, an image upload, and a real-time preview with validation checks
Overview

Connecting an AI avatar shouldn't feel like wiring up a server.

WideBot lets a business drop an AI avatar onto its own site to talk to visitors. But turning it on meant stitching together three unrelated technical tasks: uploading an image to an exact spec, whitelisting the domains it could run on, and embedding a script, with no feedback on whether any of it was correct until the very end. The people doing this are often non-technical, so they stalled, or shipped an avatar that didn't work. As the solo product designer, working with a PM and engineers, I designed the setup flow, its validation, and the UI.

The Problem

What wasn't working

Users couldn't see whether it was working, so they couldn't trust that it was.

The Insight

The hard part wasn't the steps. It was seeing that it worked.

Every problem pointed the same way. The individual steps were manageable; what was missing was confidence. Users needed to see that each piece was right as they went, not discover a broken avatar after publishing. So the design bet was to make correctness visible: validate as you go, and show the result before anyone commits.

Show it works before you ship it.

Design Decisions

Four decisions that turned setup into a guided path

Scroll to explore the full flow →

AI Avatar setup flow with validation on the image step A four-step wizard: AI Avatar widget, Avatar Image, Whitelist domains, and Add script. The Avatar Image step expands into a validation loop: drop an image, validate resolution and mouth-closed, then a decision that either passes to a live preview and save, or fails with a clear reason and a re-upload that re-validates. Setup flow four guided steps, with validation on the image step GUIDED SETUP 1 · AI Avatar widget 2 · Avatar Image 3 · Whitelist domains 4 · Add script AVATAR IMAGE STEP · VALIDATION zoom in Drop or choose an image Validate image 1920×1080 · mouth closed Valid? Live preview → Save & proceed Show reason · re-upload pass fail re-validate

One flow, four steps, validation where it counts. The wizard keeps the four stages in view, and the image step, the one most likely to go wrong, validates on upload and routes a bad image to a clear reason and a re-upload instead of a broken avatar.

1

A four-step wizard with a persistent stepper

Setup spans unrelated tasks, so I split it into Widget, Avatar Image, Whitelist Domains, and Add Script, with a stepper that always shows where you are and what is left. I considered one long settings page, but only a stepper made progress and recovery legible across steps that have nothing to do with each other.

2

Validate the image at upload, not at the end

The image has hard requirements: 1920×1080, and a closed mouth so the avatar animates cleanly. Instead of accepting anything and failing on save, the design checks resolution and detects an open mouth the moment a file lands, so a bad image never reaches the avatar.

3

A real-time preview before commit

The uploaded image renders exactly as the avatar will appear, with the validation results beside it. This is the decision the whole flow rests on: it replaces trusting the result after publishing with seeing it is right before saving.

4

Inline guidance over a separate guide

Tooltips and on-screen hints sit at each step, with documentation one click away rather than required reading. Help lives where people actually get stuck, not in a doc they have to leave the flow to find.

The Solution

What the user sees

AI Avatar setup wizard with a four-step stepper and an image upload area

A guided flow, always visible. The stepper keeps the four stages, Widget, Avatar Image, Whitelist Domains, and Add Script, in view, so the technical setup never feels open-ended.

On the Avatar Image step, the upload is checked instantly for resolution and a closed mouth, then rendered in the live preview shown at the top of the page, so the user confirms the avatar will look right before saving rather than finding a problem after it goes live.

Measuring Success

How I'd judge the design

These are the success targets I defined to evaluate the design. They are goals to measure against in production, not post-launch results.
MetricTargetWhy it matters
Setup completion (start to script live)≥ 80%Adoption
Image valid on first upload≥ 85%Efficiency
Setups reaching a broken avatar≤ 2%Quality
Time to connectunder 5 minEfficiency
Setup satisfaction≥ 4/5Experience
Impact & Reflection

What I'd expect to change

Setup stopped being a leap of faith and became something you can see working.

If I did it again: I'd validate the flow with real operators hitting the whitelist and script steps, pressure-test the mouth detection for false positives (a closed mouth wrongly flagged is its own frustration), and instrument the funnel to confirm the completion and error targets actually hold in production.

Key tradeoff

Strict validation vs. friction. Blocking an open-mouth or wrong-size image prevents a broken avatar, but risks false positives that stop a valid image. I accepted the stricter check because a silently broken avatar is worse, and paired it with a clear reason and an obvious way to re-upload.

Thanks for reading

Want to see more, or talk?