DATA5000 — Artificial Intelligence Programming in Business Analytics

Application of AI-driven Frameworks
in Business Analytics

Workshop 12

From individual tools to integrated AI solutions

Today's Agenda

Our final workshop — bringing it all together

Part A: Theory & Synthesis

Your complete AI toolkit, the analytics stack, and an end-to-end case study with FreshCart

~45 minutes

Part B: Workshop

Design an AI-driven solution for PulseFit using the complete framework you have learnt

~45 minutes

Part C: Wrap-Up

Key takeaways from DATA5000, career readiness, and your next steps

~15 minutes

Learning Objective

By the end of this workshop, you will be able to design an end-to-end AI-driven analytics solution that integrates predictive, causal, generative, and agentic AI techniques to address a real business problem.

Your AI Toolkit — The Complete Picture

Over 11 weeks, you have not learnt 11 separate topics. You have built an integrated AI capability.

Foundation

Weeks 1–4

AI fundamentals, predictive ML, causal ML, meta-learners

AI Architectures

Week 5

Advanced neural architectures, transformers, deep learning

Generative AI

Weeks 6–8

GenAI, prompt engineering, AI-assisted coding

Agentic AI

Weeks 10–11

Agent frameworks, evaluation, ethics

The Key Insight

These are not competing approaches — they form a connected AI-driven analytics stack that modern organisations deploy together. Today, we integrate them into a single, coherent framework.

The AI-Driven Analytics Stack

Remember the business analytics pyramid from Week 1? Here is what it looks like with your complete toolkit.

Prescriptive "What should we do?" — Agentic AI systems that recommend and execute actions (Weeks 10–11)
Predictive "What will happen?" — LightGBM, NeuralProphet for forecasting and classification (Week 2)
Diagnostic "Why did it happen?" — Causal ML, SHAP, EconML to find true drivers (Weeks 3–4)
Descriptive "What happened?" — AI-assisted coding for data exploration and visualisation (Weeks 6–8)

Full circle: In Week 1, this pyramid was abstract. Now you have concrete tools, techniques, and hands-on experience at every level.

How AI Systems Are Structured in Organisations

A non-technical view of how the analytics stack is implemented in practice.

Human-in-the-Loop Business analysts interpret outputs, make decisions, provide oversight. You sit here.
Agent Layer Agentic AI orchestrates workflows, uses tools, applies guardrails, escalates to humans when uncertain.
Model Layer Predictive models (LightGBM), causal models (EconML), generative models (LLM APIs) run analysis.
Data Layer Customer data, transaction records, operational metrics — quality and governance matter most here.

Your Role as Business Analyst

You are the orchestrator — not the coder who builds these layers, but the professional who designs the solution, interprets the outputs, and communicates insights to decision-makers.

Case Study: Meet FreshCart

An end-to-end AI-driven analytics solution

FreshCart — Australian Online Grocery Delivery

FreshCart is an online grocery delivery platform operating across Sydney and Melbourne, serving approximately 200,000 active customers. The company delivers fresh produce, pantry essentials, and household goods directly to customers' doors within 2-hour windows.

Challenge 1: Customer Churn

22%monthly churn rate

Customers are leaving at an unsustainable rate, but the leadership team does not know why.

Challenge 2: Delivery Efficiency

31%late deliveries

Nearly one-third of orders arrive outside the promised delivery window.

Challenge 3: Demand Planning

18%weekly stockout rate

Popular items frequently run out, forcing substitutions that frustrate customers.

The CEO asks: "How can we use AI to reduce churn by 15% within the next quarter?"

FreshCart — Descriptive Analytics

"What happened?" — AI-Assisted Data Exploration (Weeks 6–8)

What the Analyst Does

  • Use AI-assisted coding (Google Colab + Gemini API) to explore 200K customer records
  • Generate automated visualisations of order patterns, delivery performance, and customer segments
  • Prompt the LLM to summarise key trends from the data exploration

Tools from Your Toolkit

  • Python + Pandas for data manipulation
  • GenAI-assisted coding for rapid EDA
  • Prompt engineering for structured analysis

Key Findings (Descriptive)

  • Peak orders occur Tuesday–Thursday evenings
  • Average basket size is $87 (down from $102 six months ago)
  • Customers in inner-city suburbs order 2.3x more frequently
  • Most churned customers had 3+ late deliveries in their last 5 orders
  • Stockouts are concentrated in fresh produce (dairy, vegetables)

These are observations, not explanations. We need the next level.

FreshCart — Diagnostic Analytics

"Why did it happen?" — Causal Machine Learning (Weeks 3–4)

The Critical Question

We observed that churned customers had more late deliveries. But does late delivery cause churn, or do both result from a confounding factor (e.g., suburban customers have longer routes AND higher expectations)?

Causal ML Approach

  • Treatment: Experiencing 2+ late deliveries in a month
  • Outcome: Customer churn within 30 days
  • Confounders: Location, order frequency, basket size, tenure
  • Method: EconML causal forest with SHAP explanations

Causal Findings

  • Late delivery does cause churn: ATE = +14 percentage points (customers with 2+ late deliveries are 14pp more likely to churn)
  • Heterogeneous effects: Impact is strongest for new customers (tenure < 3 months): CATE = +23pp
  • Stockouts amplify the effect: Late delivery + substitution increases churn probability by 28pp
  • SHAP reveals: Delivery timeliness is the #1 driver, followed by substitution rate, then price sensitivity

Business implication: Fix delivery reliability, especially for new customers.

FreshCart — Predictive Analytics

"What will happen?" — ML Forecasting and Classification (Week 2)

Model 1: Churn Prediction (LightGBM)

  • Task: Classify which customers will churn in the next 30 days
  • Features: Delivery performance, order frequency, basket trends, tenure, complaints
  • Performance: AUC = 0.87, identifying 78% of at-risk customers
  • Output: Daily risk scores for all 200K customers

Why LightGBM?

Handles mixed data types, fast training on large datasets, interpretable with SHAP — ideal for business analytics where explainability matters.

Model 2: Demand Forecasting (NeuralProphet)

  • Task: Forecast weekly demand per product category per warehouse
  • Features: Historical sales, seasonality, promotions, weather, holidays
  • Performance: MAPE = 8.3% (down from 15% with manual planning)
  • Output: 4-week rolling forecast by product and location

Why NeuralProphet?

Combines neural network flexibility with Prophet's interpretable trend/seasonality decomposition — accessible for business users to understand forecasts.

FreshCart — Prescriptive Analytics

"What should we do?" — Agentic AI System (Weeks 10–11)

Now we design an agentic AI system that acts on the predictions and causal insights automatically.

Monitor

Agent checks daily churn risk scores and demand forecasts

Analyse

Cross-references risk factors with causal drivers (delivery issues, stockouts)

Decide

Selects personalised retention action based on customer segment and risk level

Act

Sends targeted offers, adjusts delivery priority, or escalates to human team

Evaluate

Tracks outcomes, measures effectiveness, refines approach

Automated Actions

  • High-risk new customer with late delivery → Priority re-delivery + $15 credit
  • Medium-risk customer with stockout pattern → Personalised product alternatives email
  • Low-risk customer with declining basket size → Targeted promotion on frequently purchased items

Guardrails (Week 11)

  • Budget cap: Maximum $20 credit per customer per month
  • Human escalation: Any action affecting 500+ customers requires manager approval
  • Bias monitoring: Weekly audit of offers by demographic segment
  • Evaluation: A/B testing of retention strategies with control groups

FreshCart — The Complete AI-Driven Solution

All four analytics levels working together as one integrated system.

Descriptive

AI-assisted exploration reveals 22% churn, delivery problems, declining baskets

GenAI Coding

Diagnostic

Causal ML confirms late delivery causes churn (+14pp), especially for new customers

EconML + SHAP

Predictive

LightGBM identifies at-risk customers daily; NeuralProphet forecasts demand weekly

LightGBM + NeuralProphet

Prescriptive

Agentic system executes personalised retention actions with human oversight

Agentic AI + Guardrails

Projected Business Impact

-37%

Reduction in churn
(22% → 14%)

-52%

Reduction in stockouts
(18% → 8.6%)

$2.4M

Estimated annual
revenue retained

Knowledge Check

Testing your understanding of the FreshCart case study

Question 1: In the FreshCart case, why was causal ML necessary instead of just observing the correlation between late deliveries and churn?

A) Causal ML runs faster than descriptive analytics
B) A confounding factor (e.g., location) might explain both late deliveries and higher churn, making the correlation misleading
C) Descriptive analytics cannot process large datasets
D) Causal ML is always more accurate than predictive ML

Question 2: Why does the agentic system include a human escalation rule for actions affecting 500+ customers?

A) AI cannot process more than 500 records at once
B) It is a legal requirement in Australia
C) High-impact decisions require human oversight to manage risk, prevent errors at scale, and ensure accountability
D) Human managers are faster at making decisions than AI agents

The Business Analyst's Role in AI Systems

Where you fit in organisations deploying AI-driven analytics.

What You Do

  • Define the problem: Translate business challenges into analytical questions
  • Design the solution: Select which AI techniques to apply at each analytics level
  • Interpret outputs: Critically evaluate model results and causal findings
  • Communicate insights: Present findings to executives in business terms
  • Govern the system: Set guardrails, monitor for bias, ensure ethical use

What You Delegate

  • Writing production code: Software engineers build the pipelines
  • Infrastructure: Data engineers manage databases and cloud systems
  • Model optimisation: Data scientists fine-tune hyperparameters
  • Deployment: MLOps teams manage model serving and monitoring

The Vibe Coding Advantage

Your ability to prototype with AI-assisted coding means you can build proof-of-concepts before involving technical teams — making you a more effective communicator and faster decision-maker.

From Tool User to Solution Architect

The progression you have made in this course mirrors a real career trajectory.

Weeks 1–4

Tool User

Run individual models, interpret outputs

Weeks 5–8

AI Collaborator

Use GenAI to augment analysis and coding

Weeks 10–11

System Designer

Design agentic workflows with evaluation

Week 12

Solution Architect

Integrate everything into a complete business solution

Critical Thinking Over Blind Trust

At every stage, your value comes from asking the right questions, not from accepting AI outputs uncritically. The best business analysts are sceptical collaborators — they leverage AI's speed while applying human judgement on context, ethics, and business strategy.

Where AI in Business Analytics Is Heading

Emerging trends shaping the next 3–5 years

1. Multi-Agent Enterprise Systems

Organisations are deploying teams of specialised AI agents — one for customer analysis, one for supply chain, one for financial reporting — that coordinate with each other. The business analyst becomes the agent orchestrator.

2. Domain-Specialised AI Models

General-purpose LLMs are being fine-tuned for specific industries: FinanceGPT for banking, MedPaLM for healthcare, legal AI for compliance. Business analysts who understand their domain deeply will guide this specialisation.

3. AI Governance and Regulation

Australia's voluntary AI Ethics Framework and the EU AI Act are establishing compliance requirements. Every AI-driven solution will need documentation of fairness, transparency, and accountability — skills you practised in Week 11.

4. AI-Augmented Decision Intelligence

The convergence of causal ML + GenAI + agents is creating "decision intelligence" platforms — systems that not only predict outcomes but explain why and recommend specific actions with confidence intervals.

Your DATA5000 Skills Portfolio

What you can now articulate on your CV and in interviews.

W1–2 Machine learning model development (LightGBM, NeuralProphet)
W3–4 Causal inference and treatment effect estimation (EconML, SHAP)
W5 Understanding of advanced AI architectures (transformers, deep learning)
W6–7 Generative AI application and prompt engineering (LLM APIs)
W8 AI-assisted coding and rapid prototyping (Python, Google Colab)
W10 Agentic AI framework design and implementation
W11 AI evaluation, governance, and ethical AI practice
W12 End-to-end AI solution architecture for business analytics

How to Communicate This

On your CV: "Designed and prototyped AI-driven analytics solutions integrating predictive modelling, causal inference, generative AI, and agentic frameworks for business decision-making." — This is what differentiates you from someone who simply "used ChatGPT."

Part B

Interactive Workshop

Design an AI-Driven Solution for PulseFit

~45 minutes — group activity

Link to notebook

Workshop Scenario: PulseFit

Your team has been hired as AI analytics consultants

PulseFit — Subscription-Based Fitness App

PulseFit is an Australian fitness app with 85,000 paid subscribers ($29.99/month). The app provides personalised workout plans, nutrition tracking, progress analytics, and live virtual classes. After strong initial growth, the company is facing serious challenges.

Challenge 1: Subscriber Drop-Off

35%cancel within 90 days

Over one-third of new subscribers cancel before the end of their third month.

Challenge 2: Low Engagement

2.1sessions per week (avg)

Active subscribers average only 2.1 workout sessions per week, well below the 4+ target.

Challenge 3: Content Mismatch

67%use <20% of content

Most subscribers only engage with a small fraction of available workouts and features.

The CEO asks: "Design an AI-driven strategy to reduce 90-day churn to under 20% and increase average sessions to 3.5 per week."

Workshop Design Template

In your groups, design a complete AI-driven solution using this framework. You have 30 minutes.

1. Descriptive Analytics

What data would you explore? What questions would you ask using AI-assisted coding? What patterns do you expect to find?

2. Diagnostic Analytics

What causal questions need answering? What is the treatment? Outcome? Potential confounders? Which causal ML method would you use?

3. Predictive Analytics

What would you predict? Which model(s) would you use and why? What features would be important? How would you evaluate performance?

4. Prescriptive Analytics

What automated actions would the agentic system take? How would it personalise interventions by subscriber segment?

5. Guardrails & Ethics

What could go wrong? What human oversight is needed? How would you monitor for bias? What evaluation framework would you use?

6. CEO Pitch (3 minutes)

Summarise your solution for a non-technical CEO. What is the expected business impact? What resources are needed? What is the timeline?

Tips for Success

Be specific — name actual tools (LightGBM, not "a model"). Reference causal thinking (confounders, treatment effects). Think about what the agent actually does, not just what it analyses. And always connect back to business value.

Workshop Time

30 minutes — work in your groups

30:00

Remember

  • Cover all six template sections
  • Be specific about tools and techniques from the course
  • Prepare a 3-minute CEO pitch
  • Think critically — what could go wrong?

Group Presentations & Peer Review

Each group presents their 3-minute CEO pitch

Presentation Guidelines

  • 3 minutes maximum per group
  • Present as if pitching to PulseFit's CEO
  • Focus on business impact, not technical jargon
  • Explain your approach at each analytics level
  • Address at least one ethical consideration

Peer Evaluation Criteria

  • Completeness: Did they cover all four analytics levels?
  • Specificity: Did they name actual tools and techniques?
  • Causality: Did they distinguish correlation from causation?
  • Feasibility: Is the agentic system realistic and well-governed?
  • Communication: Would a CEO understand and be convinced?

Class Vote

After all presentations, the class will vote on the most innovative solution and the best CEO pitch. Remember: creativity and business thinking matter as much as technical accuracy.

DATA5000 — Five Principles to Carry Forward

Part C: Course Wrap-Up

1. AI is a Collaborator, Not a Replacement

AI augments your analytical capability. Your value lies in judgement, context, and strategic thinking that AI cannot replicate. The human-AI partnership is where the best outcomes emerge.

2. Correlation is Not Causation

The most dangerous business decisions come from acting on correlations as if they were causes. Always ask: "Is there a confounder?" Causal ML gives you the tools to answer rigorously.

3. Business Value Over Technical Complexity

The best AI solution is not the most complex one — it is the one that solves the business problem. Always start with the question: "What decision does this enable?"

4. Critical Thinking is Non-Negotiable

Never accept AI outputs uncritically. Verify results, check for bias, question assumptions, and always consider what the model does not know. You are accountable for AI-informed decisions.

5. Ethics and Governance Are Your Responsibility

Fairness, transparency, accountability, and human oversight are not optional add-ons — they are foundational requirements of every AI system you design or manage.

Thank You — DATA5000

You started this course as beginners in AI. You leave as AI-literate business analysts.

Continue Your Learning

  • Practice: Use Google Colab and LLM APIs to prototype solutions for problems you care about
  • Stay current: Follow AI developments — the field evolves rapidly
  • Build your portfolio: Document projects that demonstrate your end-to-end AI analytics capability
  • Think critically: Apply the causal and ethical frameworks you have learnt to every AI system you encounter

Your Competitive Advantage

Most people can prompt an LLM. Very few can design a complete AI-driven analytics solution that integrates predictive modelling, causal inference, generative AI, and agentic frameworks — with proper governance. You can.

Best of luck in your careers.

Logo