Skip to content

Selected work04AI support · Product Designer @ Chams Access

Helpnest.Closing the gap between bot and human.

An AI-native support platform that blends intelligent triage, seamless human escalation and cross-product analytics into one workspace, for B2B teams who can't choose between speed and context.

Role
Product Designer @ Chams Access
Timeline
2025 · 8 weeks
Scope
User research · 30+ dashboard screens
Surfaces
Admin and Super-admin web dashboards
Helpnest admin dashboard: total conversations, bot resolved, satisfaction rate, escalations, response time, active chats and a seven-day conversation volume chart
01The problem
When a bot hands a conversation to a human, the context disappears. Every escalation starts from zero.
02The bet
Treat the handoff as a first-class design problem, not an afterthought, and design everything else around it.
03My role
Product designer: extensive user research, then the Admin and Super-admin dashboards end to end.
04The outcome
30+ dashboard screens across eight surfaces, from the live agent console to the escalation rules.

01Context

Every escalation started from zero.

B2B companies running several products face a compound support problem.

Bots resolve the common questions. But the moment a conversation needs nuance, the handoff to a human erases everything before it. Agents walk into escalations blind, managers can't see which products cause the most friction, and customers feel the gap.

Helpnest was designed from the ground up to close that gap, by treating the bot-to-human transition as the core design problem.

The best escalation is the one the customer never notices has happened.
Design principle · Helpnest

02Problem

Five failures, five transitions.

  1. 01Context dies at escalation.When the bot hands off, the thread resets. The agent can't see what the customer already tried, what the bot explained, or how long they've waited.
  2. 02The queue has no priority.A failed payment blocking someone's pension contribution sits in the same undifferentiated queue as a routine password reset.
  3. 03Performance is invisible.Resolution rates, response times and satisfaction scores live in separate reports, if they exist at all, so managers have no unified view across products.
  4. 04Knowledge lives outside the flow.Agents switch tabs to find documentation while customers wait, which adds delay and makes answers inconsistent.
  5. 05Escalation is reactive.Conversations escalate after customers have already said they're frustrated. By then, satisfaction is gone.
Each failure is a transition: bot to human, knowledge to chat, individual to team, reactive to proactive. Helpnest is designed around those transitions.

03Research

Three roles, one shared surface.

The research covered the people on both sides of a support conversation, and the person managing everyone in it.

  • Works the queue

    Support agent

    “Everything I need, without switching tabs.”

    Handles live conversations across several client products. Time to first response is the main pressure, so the queue, history, knowledge base and chat must sit together.

  • Runs the team

    Support manager

    “Data I can act on, not just dashboards.”

    Monitors team performance, configures escalation logic and reviews analytics across products, to make staffing and policy decisions quickly.

  • Asks for help

    End customer

    “A fast answer, from whoever has it.”

    Meets Helpnest through the chat widget. The product succeeds when they don't notice the switch from bot to human.

04Thinking

Designing for the second between bot and human.

  1. 01Carry context forwardOne constraint shaped everything: the handoff must carry context forward, not drop it. The agent sees the current issue, full history and activity timeline before typing a word.
  2. 02Triage from the listThe queue was redesigned around priority, so agents can triage from the list without opening each conversation, which cuts first-response lag.
  3. 03Escalate before frustrationEscalation rules watch for negative sentiment, repeated failed attempts and wait time, so a human steps in before the customer gives up.
  4. 04Insights before chartsAnalytics follow a manager's real decision loop: AI-written insights about anomalies come first, raw charts second, and everything exports to PDF, Excel or CSV.

05Design

Eight surfaces. One coherent workspace.

Live agent console

Queue, context and chat, in one uninterrupted view.

The moment of handoff, designed. The conversation the agent inherits already shows the bot's attempt, when it escalated and when the agent joined.

Live agent console: a prioritised queue on the left, the active conversation with bot and agent messages in the centre, and quick actions below

The live agent console, with four annotated details.

  1. 1Triage before opening

    Every queued conversation shows product, wait time and a priority badge, so the urgent one is obvious from the list.

  2. 2The handoff, on the record

    The thread marks when the bot escalated and when the agent joined, and keeps the bot's messages in view.

  3. 3History, one tap away

    Wait time sits in the header, and the customer's full history opens without leaving the chat.

  4. 4Answers inside the input

    Canned responses, the knowledge base and file sharing live where the agent types, not in another tab.

Conversation history

A customer's full record, before you say hello.

Conversation history: the current issue, six past tickets as cards, and an activity timeline
History
Ticket detail: a resolved password reset with resolution time, messages exchanged, assigned agent, priority, the chat thread and tags
Ticket detail

Every past ticket as a card with status, summary and date, then the full thread of any one of them: resolution time, messages, assigned agent and tags.

Escalation rules

Configure the bot. Define the rules it plays by.

Proactive escalation, as settings a manager can actually reason about: how many failed attempts, how long a wait, and which signals hand the conversation to a person.

Escalation rules: auto-escalate after 3 failed attempts or 10 minutes, with triggers for negative sentiment, requests for a human, complex technical queries and after-hours
Smart escalation
Sentiment and explicit requests escalate immediately; complex queries and after-hours messages queue for the next available agent.

For managers

The rest of the workspace.

Inbox, analytics, knowledge base, staff and settings share one layout, one stat-row pattern and one set of status badges, so moving between them never means relearning.

Conversations inbox: daily totals, active, resolved, escalated and helpfulness, with All, Active, Resolved and Escalated tabs and a list of conversations

  1. 1An operational read first

    Five numbers summarise the day before any thread is opened.

  2. 2Four tabs, one inbox

    All, Active, Resolved and Escalated filter the same list, each with a count.

Design system

Calm authority, legible under pressure.

Typography

Segoe UI

Display · 400 weight · readable in under a second

Support agents work at speed, so every label, status and action has to read at a glance. One calm blue carries actions and data; amber and red are kept for what needs attention.

Palette

  • Primary#2F6D9B · Actions, data
  • Background#FFFFFF · Page
  • Amber#F79009 · Attention, medium priority
  • Red#F04438 · High priority, failure

Status, never colour alone

  • Resolved
  • Medium
  • High

Actions

06Outcome

What shipped.

  • 30+

    Dashboard screens

    From the main dashboard and live agent console through history, inbox, analytics, knowledge base, staff and settings.

  • 2

    Views, one system

    Admin and Super-admin dashboards, sharing one layout, stat-row pattern and badge language.

  • 8

    Surfaces, one workspace

    Every surface an agent or manager needs, arranged so the bot-to-human handoff never loses context.