PRODUCT & UX DESIGN ∙ AI PRODUCT EXPERIENCE

Designing an adaptive, AI-powered multilingual form experience

A low-code form and automation platform needed a better way for users to create forms for multilingual audiences. The challenge wasn't simply translating the form, it was helping users navigate the complexity of language configuration, translation, localization, and validation without requiring them to understand how the underlying system worked.


THE CHALLENGE

Creating a multilingual form can quickly become a technical task, where users need to think about:

  • Which languages to support

  • Translation quality and consistency

  • Regional formatting

  • Missing content

  • Validation and error messages

  • Language-specific variations

  • What should happen when translations are incomplete

For less experienced users, exposing all of these decisions upfront creates unnecessary cognitive load. How can we simplify localization for users who shouldn’t have to understand localization?

THE SOLUTION: A GUIDED AND ADAPTIVE WORKFLOW

The opportunity was to make the experience adapt to the user rather than forcing the user to adapt to the product. I designed a guided, adaptive workflow that uses AI to help users move from intent, to creation, to translation, to review, and to publish, while progressively revealing complexity as needed.

Step 1: Goal-Driven Form Creation →

Step 2: AI-Generated Form & Adaptive Guidance →

Step 3: Adaptive Multilingual Setup →

Step 4: Multilingual Content Review & AI Assistance →

Step 5: Adaptive Validation & Publish Readiness→

STEP 1: GOAL-DRIVEN FORM CREATION


One step at a time… starting with defining the user’s goal

Instead of beginning with an empty form builder and asking users to configure fields, languages, and settings, the experience starts with: “What are you building?”

Users describe what they're trying to accomplish, and AI uses that context to generate an appropriate starting point.

Users understand their business goal better than they understand the platform's configuration model.

Key design decisions:

  • A persistent 4-step rail is the single source of truth for position; only the current step is expanded

  • Completed steps collapse to a one-line summary with an Edit affordance, so prior input stays visible but quiet

  • One question per screen with a ‘Question 1 of 2’ counter and an explicit ‘Next’ preview removes ambiguity about what remains

  • Progressive disclosure without hiding the path; future steps are shown locked with a plain-language description

  • The assistant panel narrows to why this question is being asked and what happens after, instead of the full plan

These design decisions allows for lower cognitive load for new or non-technical users.

A screenshot of a web interface for creating a new customer intake form, showing steps including describing the purpose, tailoring questions, reviewing the plan, and creating the form. The interface includes progress indicators, editable sections, and options for different user groups such as customers, employees, partners, and the public.

The user approves exactly what will be built before anything is created

Key design decisions:

  • The plan is grouped into structure, languages and behaviour, so the user reviews decisions rather than a flat field list

  • Every proposed field is individually editable and removable at plan time, approval is not all-or-nothing

  • Each behaviour rule states its cause ('because this form is customer-facing'), making the AI's reasoning auditable

  • 'Nothing created yet' is repeated in the header, the button row and the assistant panel to remove any doubt about state

  • The primary action is 'Create form', explicitly labelled as producing a draft, publishing is never implied here

  • Steps 1 and 2 stay collapsed as one-line summaries so context is available without competing with the decision

Screenshot of an online form creation or editing interface, showing steps including describing, tailoring, reviewing, and creating a customer intake form with multiple fields such as name, email, company, request type, languages, and privacy consent.

The draft is generated transparently, and the user is handed clear next actions

Key design decisions:

  • Creation is shown as an itemised task list, not an opaque spinner, so the user knows what exists and what is still preparing

  • Task states are text-labelled (Done / In progress / Queued) alongside the icon, never colour-only

  • 'Draft — not published' chip on the result summary keeps the publish boundary explicit at the moment of creation

  • Why draft output: presenting the result as reviewable rather than finished keeps the user the decision-maker

  • The assistant separates 'what I created' from 'what still needs a human decision', turning AI output into a review queue

  • Two forward paths (review translations, open editor) plus a way back to the plan avoid a dead-end completion screen

Screenshot of an online platform showing the progress of creating a customer intake form in four steps: Describe, Tailor, Review plan, and Create. The form is titled Building your draft form with a list of completed tasks and remaining tasks, and side notes about the next steps and draft status.

STEP 2: AI-GENERATED FORM & ADAPTIVE GUIDANCE

Use AI as a co-pilot, not an autopilot

AI helps generate the form structure, translations, recommendations, and quality checks, but the user remains in control.

The interface makes AI's role visible through:

  • Suggestions

  • Confidence indicators

  • Explanations

  • Review states

  • Accept / reject controls

  • Manual editing

Automation should reduce work without removing user agency or creating false confidence.

Key design decisions:

  • Every AI-authored element is labelled at the point of use, so provenance is obvious while editing

  • The structure is presented as editable draft output, never as a locked result

  • The assistant explains its reasoning as recommendations, keeping the user the decision-maker

  • A single 'Next recommended step' with progress dots directs attention without blocking other work

Screenshot of an online form creation platform showing a customer intake form being generated with fields for name, email, country, and request details, along with options for language setup and privacy policy agreement.

STEP 3: ADAPTIVE MULTILINGUAL SETUP

Make localization complexity progressive, so it’s a clear human decision

Rather than exposing localization settings immediately, the experience introduces complexity when it becomes relevant.

The primary workflow focuses on: Languages, Translations, Review, and Publish

More technical concepts, such as locale configuration, translation keys, fallback languages, and regional variants—remain available under Advanced settings.

Progressive disclosure keeps the experience approachable for new users while preserving power for experienced users.

Key design decisions:

  • Language choice is framed as an outcome ('who will see which version'), not as a settings toggle

  • Per-language state is text-labelled rather than colour-only, so status survives greyscale printing

  • Advanced localization options stay collapsed until they are relevant

  • The source language is pinned and visually distinct to anchor the translation model

Screenshot of a website page titled 'Make your form multilingual' showing options to select languages for translation like English, French, and Spanish, with a section on how the translation process works and a button to generate translations.

STEP 4: MULTILINGUAL CONTENT REVIEW & AI ASSISTANCE

Adapt the experience to user expertise, and let AI assist the review; the user keeps final control

The workflow recognizes that a first-time user and an experienced builder shouldn't necessarily see the same experience.

For newer users, the interface provides:

  • Contextual explanations

  • "Why this matters" guidance

  • Recommended next steps

  • Examples

  • More visible AI assistance

For experienced users, the same workflow can become more direct, with faster access to configuration and advanced controls.

The workflow doesn't simply duplicate English content across three columns.

It helps users identify:

  • Missing translations

  • Potentially awkward phrasing

  • Inconsistent terminology

  • Regional formatting issues

  • Untranslated validation messages

  • Content that may require human review

AI can suggest alternatives while the user makes the final decision.

The best interface isn't necessarily the simplest one, it's the one that provides the right level of complexity for the person using it. A multilingual experience needs to preserve meaning, context, and usability, not just words.

Key design decisions:

  • Source and translation sit side by side so comparison needs no memory or tab switching

  • AI suggestions are proposals with accept/edit actions, and nothing is applied silently

  • Confidence and issue flags are surfaced per row so attention goes to the risky strings first

  • Bulk actions exist but are secondary to per-string review.

Screenshot of a translation review interface showing translations from English to French and Spanish, with statuses such as reviewed and needs review, and options for acceptance and editing.

STEP 5: ADAPTIVE VALIDATION & PUBLISH READINESS

Validate the experience before publishing

The final step isn't simply: Translations complete = Publish. Instead, the system provides a multilingual readiness check.

It surfaces issues such as:

  • Missing translations

  • Translation quality concerns

  • Untranslated system messages

  • Regional formatting problems

  • Inconsistent terminology

The user gets a clear distinction between critical issues and recommendations.

Catching problems before publication reduces errors, support dependency, and user frustration.

Key design decisions:

  • A readiness score plus itemised checks answers 'can I publish?' in one glance

  • Blocking issues are separated from advisory ones, each with a direct path to the fix

  • Preview sits alongside Publish with per-language switches, so the user can verify before committing

  • Internal sharing (link, reviewers, approval status) turns review into a first-class step rather than an email round-trip

  • Publishing stays a deliberate, explicit action — no auto-publish anywhere in the flow

Screenshot of a web interface for a form builder application called Formary. The page shows a section indicating the form is almost ready, with checkmarks confirming translation of all fields, required fields configured, and validation messages translated. It lists translation quality for English, French, and Spanish, with a warning for one item needing review. The right side shows form readiness status with recommendations, preview options, publish and continue buttons, and review status with reviewers' names and approval options.

Let’s talk about what you’re trying to solve.

I take a small number of new engagements each year. If your challenge sounds like the work above, I’d love to hear about it.

Let's talk
Find your path