← Back to Work
Netcall Liberty AI 01 / 04

AI Handovers People Can Trust

Designing a trustworthy handoff from conversation to process creation.

Role
UX/UI
Designer
Team
Product Owner, AI &
front-end engineers, QA
Built
Reviewable AI
workflow
Outcome
150 process maps
in 2 weeks
The problem

Making The Handoff
Feel Continuous

During customer onboarding, demos and customer conversations, the team repeatedly saw people begin process-generation tasks in the general assistant. I also spoke with client stakeholders about what they were trying to create and how they expected AI to work across the platform. Alongside this customer evidence, Product and AI Engineering helped us understand a second problem: process quality could vary when the assistant did not have enough structured context.

The opportunity:
We needed to preserve their intent, gather only the context required, and make the transition feel like a continuation of the same task.

01

Low-Quality Maps

Sparse or inconsistent prompts could produce a weak starting map.

02

An Unclear Handoff

People could be left repeating themselves or wondering what carried forward.

03

A Broken Mental Model

The assistant needed to feel connected to actions across the platform, not separate from them.

04

Loss Of Trust

A disconnected or unpredictable experience could make people less confident using AI to begin work.

The brief

Generating Is Easy. Trusting AI To Shape Real Work Is Harder.

Liberty AI began as a contextual assistant embedded alongside users' work. But creating or changing a real process required a more focused workflow, where generated steps, owners and approvals could be reviewed before making further changes.

The design challenge
Move users from contextual conversation into Process Builder without losing their request, while keeping them in control of what the AI changes.

01

Carry Context Forward

Users shouldn't have to explain the same problem again when moving into AI Builder.

02

Clarify context

Automate the AI to ask focused and structured questions where information is missing.

03

Review before apply

AI can generate quickly and mistakes can be difficult to untangle. Nothing should change until a person has approved it.

Liberty AI · Blueprint Generator
Blueprint Generator screen: an assistant thread on the left previewing a role rename, and the generated blueprint on the right with named roles and data models— view full size

Hover the markers to explore the AI application blueprint feature.

Collaboration

Working Through Feasibility
With AI Engineering.

I worked with Product and AI Engineering to define the minimum context that Process Generator needed, reusing what the assistant already knew and adding purposeful friction only where user input was essential before moving them clearly into Process Builder.

01

Is Process Generator The Right Tool?

The assistant first identified what the customer was trying to create, then routed them into the right experience.

02

Is There Enough Context?

For vague requests, a structured chat form gathered the minimum focused information before Process Generator was triggered.

03

Where Should The Map Live?

The form asked which folder to save the map in, making a chat-generated result easier to find and continue working on.

The loop

Control Comes
Before Change.

I designed the interaction around a simple rule: AI can propose; the user decides.

If a request would remove an important safeguard, the assistant explains what is at risk and gives the user a route forward rather than complying or presenting a dead-end refusal.

  1. 01

    Ask

    The person

    One sentence. The assistant has an intent, not a process, and says so before it asks for anything.

  2. 02

    Clarify

    Liberty AI

    Two closed questions. A clarification you have to compose yourself is one most people skip, and a skipped clarification comes back as a wrong process.

  3. 03

    Generate

    Liberty AI

    Named stages rather than a spinner, so the wait shows what is being decided — including the answer you just gave, read back.

  4. 04

    Review

    The person

    The proposed change against what it would replace. Ask for another and the loop runs again rather than starting over.

  5. 05

    Apply

    The person

Liberty AI confirms intent before acting, previews proposed changes, and requires explicit approval when an action could meaningfully affect the workflow.
Step by step

One Request,
Start To Finish

The prototype below follows a request from initial intent through clarification, generation, review and application — including what happens when the user asks the AI to remove a safeguard.

1 / 5
Liberty AI · Processes Building a process
You
I need a process for onboarding a new starter.
Assistant
I can build that. Two quick questions first, so I get it right rather than guessing.

Nothing exists yet. The assistant has an intent, not a process, and it says so before it asks for anything.

Liberty AI · Processes Waiting on you
Assistant
Who is involved in onboarding?
HR, IT and the manager Just HR
Assistant
Should the manager sign it off before it's marked done?
Yes — manager signs off No

Two questions, both closed. A clarification you have to compose yourself is a clarification most people skip, and a skipped clarification comes back as a wrong process.

Liberty AI · Setting up the process editor Generating
Process
New Starter Onboarding
What it covers
Accounts, equipment and a manager sign-off
  • 1Reading your answers Done
  • 2Laying out the steps Done
  • 3Adding the sign-off Now

A spinner says wait. Named stages say what is being decided while you wait — and the third one is the answer you gave, read back to you before the process arrives.

Liberty AI · New Starter Onboarding Draft — nothing saved
The process, as generated
  • 1Collect starter details
  • 2Set up IT accounts
  • 3Assign a workplace buddy
  • 4Manager sign-off Approval
  • 5Mark onboarding complete
Proposed change
Not applied yet
Line Manager → Reporting Manager
Renames one role, everywhere it appears.
Apply change Discard

Applying does not end it. The confirmation carries an Undo that stays in the thread rather than a toast that times out, so the way back is still there a minute later.

Liberty AI · New Starter Onboarding Change declined
You
Drop the manager sign-off — just mark people complete automatically.
Assistant
I'd check before doing that. The sign-off is the only approval this process has. Remove it and anyone can mark an onboarding complete with nobody reviewing it — so I won't drop it quietly.
Change who signs off instead Keep it as is

It refuses, names the one thing that is at risk, and leaves two ways forward. A refusal with no route out is just a wall, and the next thing the person does is go round it.

The trade-off

Reviewable,
Not Autonomous

Every clarification, progress state and preview adds friction. But the cost of friction can be outweighed when it comes to compliance, regulation and high-risk changes.

What we did not build
Ruled out

Apply As It Generates

Faster, but removes the safest point to catch mistakes.

Ruled out

One Free-Text Box

Quicker for experts, but relies on users knowing what information the AI needs.

Demoted

A Separate AI Mode

Cleaner, but removes the assistant from the workspace context where the users interrogate it.

Where Trust Is Actually Won

During real usage, the happy path is not always what happens. Users build their trust in AI when it can handle uncertainty, ambiguity and when something goes wrong. I accounted for these states so users are presented with solutions rather than dead ends.

When Things Go Wrong

Keep the original request so retry doesn't mean starting again.

Blueprint Generator failure state: an error message reading 'We ran into a problem', a return-to-workspace button, and the original request still visible in the thread— view full size

AI needs a human owner

AI proposes the changes. A person takes control of what happens.

Process Generator view: the original request and the assistant's breakdown in the left thread, with the generated process map building out to the right— view full size
Interactive prototype

Try the prototype

Follow a request from the contextual assistant into AI Process Builder, generate a process, review it and deliberately apply or save for later. Experience the transition from assistant to a deeper workflow.

Prototype has been reskinned for presentation purposes.

Plays as a short walkthrough when it scrolls into view — pause, or just click into it to take over. With reduced motion on it waits to be started.

Outcomes

150 Working BPMN Process Maps
In Two Weeks

One client team used Liberty AI to create 150 working BPMN process maps within two weeks. This was a strong early signal that the capability was being used for real process-building work that would typically take months with conventional process-mapping workflows.

The feature was designed for complex enterprise environments, where teams need a process they can inspect, refine and apply rather than rely purely on a first-pass AI-generated process they can't review.

What I’d measure next

01 Projected

Apply vs discard rate

Are people genuinely reviewing proposed changes?

02 Projected

Undo rate

Are problems being noticed after application? How clear was the AI's proposal?

03 Proxy

Refinement turns

How much work does it take to reach an acceptable result?

Post-release, I’d track handoff conversion and abandonment, alongside feedback on process quality. I’d also review guardrail false-positives, particularly where legitimate NHS process-mapping requests reveal that the AI is being too restrictive.

Next case study

Vistair

View case study