A structured guide to the core theories, frameworks, and exam questions from Porter & Heppelmann (2015), Iansiti & Lakhani (2021), and Medaglia's lecture.
A scalable decision engine that treats decision-making as an industrial process. Analytics convert data into predictions, which guide or automate operational actions — moving humans off the critical path.
Products with physical, smart, and connectivity components. They can monitor, control, optimise, and act autonomously — and they fundamentally reshape the value chain of any firm that makes them.
Systematically extracting data from activities that already occur in any business. The Nest thermostat datafied home climate. Netflix datafied TV watching. Ride-sharing datafied urban transport.
When an AI-driven firm serves the same customers as a traditional firm with a similar (or better) value proposition but a far more scalable operating model — e.g. Amazon vs retailers, Ant Financial vs banks.
More usage → more data → better algorithms → better service → more usage. This self-reinforcing loop creates formidable scale and learning advantages that are hard for incumbents to replicate.
The new infrastructure smart products require: product hardware/software, connectivity, the product cloud (with analytics engine, app platform, database), plus security and integration with enterprise systems.
"The AI factory treats decision-making as a science. Analytics systematically convert internal and external data into predictions, insights, and choices."
— Iansiti & Lakhani, Competing in the Age of AIThe four essential components that make the AI factory work:
Gathers, inputs, cleans, integrates, processes, and safeguards data in a systematic, sustainable, and scalable way. Netflix, for example, ingests billions of ratings, stream plays, queues, and search terms. The challenge: data from incumbent firms is often siloed, fragmented, and inconsistent.
Algorithms generate predictions about future states or actions of the business. Three families: supervised learning (predict from labelled data — e.g. spam filters, churn prediction), unsupervised learning (find hidden groupings — e.g. customer segmentation), reinforcement learning (explore/exploit trade-off — e.g. Netflix artwork personalisation, AlphaGo).
Hypotheses about improvements are tested as randomised controlled trials (A/B tests) to distinguish genuine causal effects from spurious correlation. Google runs over 100,000 experiments per year. Netflix tests every significant product change before release.
Embeds the pipeline in a consistent, componentised software and computing environment and connects it to internal and external users via APIs. A well-designed data platform lets an agile team build a new application in days rather than years.
The four new product capabilities these layers unlock:
Products report on their own condition and environment — generating insights previously unavailable (real-time engine health, soil moisture, usage patterns).
Users can operate products remotely, customise function and performance, and operate in hazardous or hard-to-reach environments.
Algorithms substantially improve performance, utilisation, and uptime — and how a product works with related products in a broader system (smart building, smart farm).
Products learn, adapt to users and environment, service themselves, and operate independently — the combination of all three prior capabilities.
| Function | Before (traditional product) | After (smart connected product) |
|---|---|---|
| Product development | Mechanical engineering; fixed-generation design; physical testing only | Interdisciplinary systems engineering; evergreen design; continuous real-world quality monitoring; low-cost software variability |
| Manufacturing | Discrete process ending at shipment; standardised physical assembly | Continuous process — the cloud stack is part of the product; smart factories with networked machines; software loaded post-shipment |
| Logistics | RFID/barcode scanning; periodic location tracking | Continuous real-time tracking of location, condition, and environment; drone delivery; optimised fleet routing |
| Marketing & Sales | One-time transaction focus; segment by demographics; customer surveys | Continuous value delivery; fine-grained usage-based segmentation; product-as-a-service models; focus on systems not products |
| After-sale service | Reactive, on-site visits; technician diagnoses on arrival; expensive second trips | Remote & preventive service; one-stop visits; AR-assisted repair; predictive analytics (Diebold ATMs); new service revenue streams |
| HR | Mechanical engineers; traditional sales and service skills | Software engineers, data scientists, UX designers; new clock speeds; new compensation norms; co-location with tech hubs |
New organisational units that emerge
Led by a Chief Data Officer. Consolidates data collection, aggregation, and analytics across all functions. No single function can manage the volume, complexity, and security of data by itself. (Gartner: 25% of large firms by 2017.)
Bridges product development ("dev") with IT and manufacturing operations ("ops"). Manages continuous product operation, software updates, patches, and new post-sale services. Shortens release cycles, runs cloud infrastructure.
Monitors product use and performance to ensure customers extract maximum value — especially critical in product-as-a-service models to drive renewals. Uses the product's data stream rather than waiting for complaints.
Iansiti & Lakhani's research shows AI maturity leaders had a 55% gross margin vs 37% for laggards, and 11% net income vs 7%. The five principles for transformation:
Strategic clarity and full commitment — no plan B. Not about spinning off an AI division; it requires rebuilding the core. Nadella's "The cloud is our future and we have fundamentally no choice" is the model.
Everyone must understand what the future operating architecture looks like. Data must be integrated, not fragmented. Without a consistent data catalog and clear standards, AI is impossible to safeguard and scale.
AI-centric applications must embed a deep understanding of their settings — like any product. Agile teams building on a shared data platform can deploy new applications in days. Gone are the years-long custom IT builds.
Hire data scientists, software engineers, and — crucially — data and analytics product managers who can bridge business context and technical execution. The market for this talent is hot and competitive.
AI's societal impacts (privacy, bias, fairness, cybersecurity) must be governed across legal, engineering, and product functions together. Microsoft's six AI principles (fairness, reliability, privacy, inclusiveness, transparency, accountability) are the benchmark.
"For traditional firms, becoming a software-based, AI-driven company is about becoming a different kind of organisation — one accustomed to ongoing transformation. This is not about spinning off a new organisation, setting up the occasional skunkworks, or creating an AI department."
— Iansiti & Lakhani, Chapter 5Session readings