Explore the challenges

Problem statement 1

Build a Chatbot using MantaGo

Learning objective

How might we turn the customer data a business already holds into personalised, automated conversations that measurably increase retention and repeat purchase?

Learning outcomes

Read the data

Interpret customer records, purchase history and behavioural signals to identify who is worth engaging and when.

Design the journey

Map an end-to-end AI-powered customer journey with defined triggers, decision points and exits.

Automate the conversation

Specify personalised conversational flows that adapt to segment, history and intent rather than sending one message to everyone.

Grow lifetime value

Connect the designed journey to retention, repeat purchase and customer lifetime value, and state how each is measured.

Skills gained

• AI workflow and customer journey design
• Customer segmentation and personalisation strategies
• Conversational AI implementation for customer retention
• Structured thinking for solving customer engagement challenges
• Framework design for scalable AI-powered customer experiences

Workshop and consultation

• MantaGo Workshop: 3 October, 2:00–6:00 PM (SGT).
• Eslitec consultation: 6 October, 7:30–8:30 PM (SGT).

Problem statement 2

ONE LESS THING

Build an Autonomous AI Agent That Gives People Their Time Back

Move beyond AI that only answers. Build an agent that can understand a goal, plan the work, take safe action and prove what it completed.

The problem

Every day, people lose valuable time switching between apps, searching for information, remembering follow-ups, organising documents, coordinating schedules and completing repetitive personal or work tasks. Most AI assistants can provide an answer, but the user still has to perform the actual work.

The challenge

How might we build a trustworthy autonomous AI agent that understands a person’s goal, plans the work, takes multiple actions across relevant tools, adapts when circumstances change and asks for human approval at the right moments—so that one meaningful personal or workplace task can be safely delegated from start to finish?

What teams may build

Teams may choose any genuine personal or workplace task. The task should be meaningful, repetitive or frustrating enough that delegation creates clear value.

• Coordinate a group project and follow up on incomplete actions.
• Plan appointments or activities across several people and constraints.
• Research a topic and prepare a decision-ready brief.
• Organise incoming documents, extract actions and track deadlines.
• Help a small business complete a recurring administrative workflow.
• Support a user who struggles with complex digital tasks.
• Monitor a changing situation and take approved follow-up actions.

Minimum solution requirements

Goal-driven

Accept a meaningful goal rather than only responding to one isolated prompt.

Plans the work

Break the goal into a visible multi-step plan.

Uses tools

Use at least two tools, services or information sources.

Takes action

Perform meaningful actions and show evidence of completion.

Keeps context

Maintain relevant context throughout the task.

Handles change

Recover from an error, missing input or changed condition.

Requests approval

Pause before sending, publishing, purchasing, deleting or exposing sensitive information.

Creates an audit trail

Provide a clear activity record and final outcome.

Proves value

Measure time saved, steps removed, errors reduced or accessibility improved.

What does not qualify

• A chatbot that only answers questions or generates text.
• A fixed automation that cannot plan, adapt or explain its actions.
• A demo that claims external actions without evidence.
• An agent that performs consequential actions without an appropriate approval step.
• A solution that exposes personal, confidential or copyrighted information without permission.

Recommended design principles

Useful

Solve one real task well rather than automating everything.

Autonomous

Let the agent plan and execute multiple steps within defined boundaries.

Trustworthy

Make actions, assumptions, evidence and uncertainty visible.

Human-controlled

Keep people responsible for consequential decisions.

Resilient

Handle missing information, tool failure and changing conditions.

Measurable

Demonstrate a clear before-and-after improvement.

Suggested demonstration flow

1. Show the original frustration. Explain who experiences the problem and why the current process wastes time or creates errors.
2. Give the agent a goal. Use a realistic request that requires multiple steps rather than a scripted single action.
3. Show planning and action. Let judges see the plan, tool use, status updates and evidence produced.
4. Introduce a surprise. Change one condition or remove one required input to demonstrate recovery.
5. Trigger an approval gate. Show the agent pausing before a consequential action.
6. Prove the outcome. Compare the old process with the agent-assisted process using a measurable result.

Challenge-specific success questions

• Does the agent complete a meaningful end-to-end task rather than merely offer advice?
• Are its plans and actions understandable to the user?
• Can the user see what was done and verify the result?
• Does it recover sensibly when something goes wrong?
• Are approval boundaries appropriate to the risk?
• Is the benefit measurable and relevant to a real user?

Closing direction

Do not try to automate everything. Find one frustrating task, redesign it around safe delegation, and prove why the user can trust your agent to act.

Workshop and consultation

• IBM Bob Workshop: 3 October, 9:00 AM–1:00 PM (SGT).
• Nexius consultation: 5 October, 7:30–8:30 PM (SGT).