Signalchain

Problem statement

Everyday communication is fragmented across platforms, leading to inefficiency, missed messages, and a loss of control.

Goals

Consolidate personal channels in one mobile-first app; use AI to prioritise, retrieve, and organise; reduce hunting across apps while maintaining trust.

Role & responsibilities

End-to-end product designer — research synthesis, IA, flow diagrams & journey maps, interactive prototyping, usability testing.

Constraints

• 8-week timeline
• AI behaviours simulated, not integrated

User research summary

Interviews with potential users focused on multi-platform communication and app roles. A literature review covered communication in information-dense environments; a competitor audit examined project management tools, social media platforms, and AI email tools. These insights refined assumptions about merging work and personal communication, informing shallow IA and contextual actions.

Personas & user stories

After reviewing the interview transcripts and coding the data, I built out personas that informed the project going forward. This process also involved understanding edge cases and developing user stories. The personas acted like a North Star — at each stage, it was important to ask if the choices made were beneficial to these potential user groups.

A persona for Sarah Thompson, one of two developed. The personas acted like a North Star in the project; at each stage, it was important to ask if the choices made were beneficial to these potential user groups.

Key pain points

To streamline the experience, the design enabled inbox-level, contextual actions and AI retrieval, addressing user frustration with time lost hunting for older messages across various applications. Prioritisation cues were incorporated to surface critical information first, solving the problem of important messages getting buried by notification noise. Finally, the scope was refocused on personal communications to prevent the unwanted blending of work and personal channels.

A screenshot of preliminary competitor research being done in Miro that was later expanded to include feature offerings and other relevant information.

Competitor & literature insights

An audit of competitors (project management, social media, AI email) and a literature review on communication in information-dense environments — particularly how medical professionals manage patient data — revealed three key insights:

• Prioritise situational awareness — UI should minimise extraneous information and use goal/task-oriented language.
• Important actions in the inbox — information architecture should be shallow, and contextual menus should assist users.
• Reminders improve response rates — follow-up prompts likely increase response rates.

These insights guided the development of user flows that quickly routed users to necessary replies.

Although initial user sentiment favoured a unified inbox for all communications, early paper and digital prototype testing revealed discomfort with merging work and personal messages in the same UI. This prompted a project refocus on personal communication.

An early digital wireframe showing a top menu option allowing for switching between work-related communication and personal messages.

Interaction model updates

“Undo send” initially delighted users, but its on-screen presence caused frustration and cognitive overload. Redesigned as transient, it now appears briefly and is dismissed by tapping the message itself. Next steps would include progressive onboarding for safety controls — modelled on Slack's tips — making them prominent for new users then subtle for experienced ones, balancing functionality with less visual clutter.

First version of the undo send feature with a button that would appear after each message was sent.

During moderated studies of replying to SMS messages, the prototype reduced time on task by 20% and required 50% fewer interactions. These savings came from allowing actions to take place directly through the inbox and reducing the layers of the information architecture.

Outcome vs. goal

The project exceeded the success criteria established during the early phases — quick SMS replies were 20% faster and required 50% fewer interactions in moderated tests, and participants preferred versions that explained the AI's rationale.

Before and after revisions to the tone of the AI in the prototype. The revised version has clearer, more transparent language of the information sources consulted.

Transparency and AI

Testing showed that features which explained how the AI reached its conclusions were preferred over versions that prioritised smaller word counts and hid this information, especially when dealing with personal communications. This built trust and eased users into the idea of an LLM having access to their communication.

AI language should prioritise transparency over brevity when building trust — especially in personal channels, where users want to see the reasoning before they accept the reply.

The flow below covers receiving a message, getting AI response ideas, searching by content, and sending a reply. This sequence shows how context-driven actions and clear AI cues build user trust and speed. Purposeful loading states signal retrieval steps and increase perceived reliability.

1. Receives message
2. →
3. Prompts user with action
4. →
5. Finds answer
6. →
7. Generates reply & confirms

The message-to-reply flow, end to end — proactive prompts, visible AI thinking states, and confirm/edit gates.

Key moments

The flow hinges on three patterns: proactive prompts, visible “thinking” states, and confirm/edit gates. Non-content states and animations communicate that thorough consideration is being given to the user's prompt, while confirmation and editing options ensure the user retains a sense of agency in the process.

#1 Starting the AI engagement

The initial prompt to engage with the AI features proactively offers an action without forcing user engagement. Background blurs and footnote-style typography visually isolate the content for user interaction.

#2 Non-content state to communicate consideration

Breaking the action across various compositions helps convey the different actions being taken, while also acting as a progress indicator and signifier of the direction the AI is heading in.

#3 Confirmation and addressing agency

Once the message is drafted, the user is prompted to send as-is or edit. This retains agency in the user and ensures that nothing is sent without their consent. It is integral to building trust in the process.

I differentiated each card component to enhance user comprehension, ensuring selection and deselection were indicated through colour variations, borders, iconography, or typographic adjustments. Clearly discernible shadows were implemented. Where colour differentiated states, it was combined with WCAG-compliant typographic elements and adequate contrast for colour-blind users, signifying a shift in state for improved visual feedback.

A screen illustrating the use of elevation and components using minimum touch targets and clear, action-focused language.

By the numbers

• 20% faster quick personal replies
• 50% fewer interactions in moderated tests

Delivered

• High-fidelity app prototypes for the most important flow
• Research plan and synthesis

Next steps

• Flesh out additional communication types within the prototype
• Test using an integrated LLM to better understand constraints such as hallucinations and unpredictable outputs
• Further user testing with larger sample sizes

Key takeaways

• AI language should prioritise transparency over brevity when building trust.
• Contextual actions decrease time on task and increase situational awareness in messaging.
• Establishing clear personal / public boundaries creates trust in users.