2.4 New Technologies in E-Commerce

The IPU syllabus explicitly lists six "new technologies for e-commerce":

1. Chatbots
2. Recommendation systems (Personalisation)
3. Smart Search
4. Product Comparison
5. Augmented Reality (AR)
6. Big Data
7. Cloud Computing

We cover each below.

---

1. Chatbots

Definition: software programs that simulate conversation with users via text or voice — typically powering customer support, sales, and discovery.

Types of chatbots

Chatbots evolved through four overlapping generations, each more sophisticated than the last. Rule-based bots are essentially decision trees with pre-scripted responses — the IVR-style "Press 1 for English, 2 for Hindi" flows. They are cheap, predictable, and break the moment a user phrases something unexpectedly. Retrieval-based bots match the user's query against a curated knowledge base and return the best-fitting canned answer — most pre-2022 customer-support bots (Intercom, Drift) worked this way. Generative bots are powered by Large Language Models (GPT, Llama, Claude, Gemini) and can produce fluent, novel responses to almost any phrasing — they are conversational but can hallucinate facts if not grounded in the brand's actual data. Almost every production-grade customer-support bot today is hybrid: rules for high-stakes flows (order status, returns), retrieval for FAQ-style queries, and a generative layer for the open-ended remainder.

E-commerce use cases

Indian chatbot platforms

• Yellow.ai — enterprise AI assistants
• Verloop — customer experience automation
• Haptik (Reliance Jio) — conversational AI
• Tata Cliq's Cliq — in-house chatbot
• WhatsApp Business + Wati / Gallabox — WhatsApp-based commerce bots

Why chatbots work

• 24×7 availability without human cost
• Multi-language — Hindi + regional
• Instant response — no wait time
• Scalable — handle 10K queries simultaneously
• Consistent — same answer every time
• Free up humans for complex queries

---

2. Recommendation Systems (Personalisation)

The "Customers also bought" / "Recommended for you" engines drive 30-35% of Amazon revenue and 75% of Netflix watch time. Building them is a science.

Three classical approaches

There are three textbook approaches to product recommendation. Content-based filtering recommends items that are intrinsically similar to ones the user has liked — same brand, same category, similar attributes, similar text descriptions. It works the moment a new item is added to the catalog (no "cold start" problem on the item side), but it tends to produce narrow, predictable recommendations: a user who bought running shoes keeps getting more running shoes. Collaborative filtering flips the problem around — it ignores the item attributes and instead looks at user behaviour: "people who bought what you bought also bought X." It surfaces serendipitous discoveries (a runner getting recommended a fitness tracker) but suffers a cold-start problem for both new users (no history yet) and new items (no one has bought yet). Hybrid systems combine both signals — content similarity blended with collaborative filtering — and are what every modern recommender uses in production.

Modern approaches

Recommendation use cases in e-commerce

---

3. Smart Search

Modern search goes far beyond keyword matching.

Smart search features

Tools

• Algolia — sub-50ms search-as-a-service
• Elasticsearch — open-source, customisable
• Typesense — modern open-source
• Pinecone, Weaviate — vector databases for semantic search
• AWS OpenSearch — managed Elasticsearch

Indian search example

• Flipkart — uses Elasticsearch + custom ML ranker
• Myntra — visual search via image upload
• Amazon India — Alexa voice search integration

---

4. Product Comparison

A side-by-side view of multiple products for high-consideration purchases (electronics, mobiles, large appliances).

Comparison features

• Up to 4 items side-by-side
• Common attributes shown in rows (RAM, battery, price)
• Highlights (which product has better value per attribute)
• Pros / cons summarised
• Reviews comparison (rating breakdown)
• Price history charts
• Stock and delivery comparison

Examples

• Flipkart "Compare" on mobiles, laptops
• Amazon "Compare with similar items"
• Smartprix, MySmartPrice — dedicated comparison sites
• Pricebaba, Cashify — also price tracking

---

5. Augmented Reality (AR) in Commerce

AR overlays digital products on real-world views via smartphone camera.

E-commerce AR use cases

Technologies enabling AR

• ARKit (Apple), ARCore (Google) — mobile AR SDKs
• WebAR — AR in browser (no app install)
• 3D models — .glb, .usdz formats
• Markerless tracking — uses phone sensors, not QR codes

AR impact

• ~70% lift in conversion rates when AR is used (Shopify data)
• ~25% reduction in returns (right-fit before buying)
• Higher engagement, time on site

---

6. Big Data in E-Commerce

E-commerce generates massive amounts of data. Big-data tech makes it usable.

Sources of e-commerce data

Big-data technologies used

What big data enables

• Personalisation at scale
• Demand forecasting — what to stock, where
• Dynamic pricing — adjust prices based on demand
• Fraud detection — pattern anomalies
• Customer segmentation — high-LTV vs at-risk
• Inventory optimisation
• Supply chain optimisation

---

7. Cloud Computing in E-Commerce

Cloud is the default infrastructure for modern e-commerce.

Cloud benefits for e-commerce

Cloud service models (recap)

The three service models layer on top of each other. IaaS (Infrastructure-as-a-Service) gives raw virtual machines, storage, and networking — used when a team wants to build its own stack from the OS up; AWS EC2, Azure VMs, and Google Compute Engine are the canonical examples. PaaS (Platform-as-a-Service) adds a managed runtime so developers deploy code without worrying about OS or scaling — Heroku, Vercel, AWS Elastic Beanstalk fit here. SaaS (Software-as-a-Service) skips the engineering entirely — the e-commerce platform itself is provided as a service. Shopify, BigCommerce, Magento Cloud, Salesforce Commerce Cloud are SaaS commerce platforms.

Cloud providers

• AWS (Amazon Web Services) — dominant in e-commerce; Amazon itself runs on AWS
• Microsoft Azure — strong with enterprise (Walmart famously moved off AWS to avoid competing on infra)
• Google Cloud — strong with data + AI
• JioCloud, TataCloud, Yotta — Indian options

Common cloud services for e-commerce

---

Combining these technologies — modern e-commerce architecture

Modern e-commerce combines all six tech areas:

A modern Indian quick-commerce platform like Blinkit combines:

• Cloud-native infrastructure (AWS)
• Big-data demand forecasting to stock dark stores
• ML routing to assign orders to nearest store
• Chatbots for customer support
• Real-time recommendation in app
• Personalised search ranking
• Indian-language voice search (vernacular)

---

Key Terms — Lesson 2.4

The seven technology areas above each carry their own vocabulary. The terms below are the highest-yield ones — examiners frequently ask you to "explain X with one e-commerce use case", and these are X.

Chatbot — A software program that simulates conversation with users through text or voice, used in e-commerce for customer support, product discovery, cart recovery, order tracking, and lead generation. Chatbots reduce per-query support cost to near zero, operate 24×7 in multiple languages, scale to thousands of concurrent users, and are increasingly powered by Large Language Models (LLMs) rather than rule-based decision trees.

Large Language Model (LLM) — A neural network trained on vast amounts of text (hundreds of billions of words) that can generate fluent natural-language responses, summarise documents, classify intent, and reason within limits. GPT-4 (OpenAI), Claude (Anthropic), Gemini (Google), Llama (Meta) are the main families. LLMs power modern chatbots, semantic search, automated review summarisation, and content generation in e-commerce.

Conversational Commerce — Commerce that happens through a chat interface rather than a traditional website — most often WhatsApp or in-app chat. Indian examples: Meesho's WhatsApp-led reseller flow, Bharat Pe / PhonePe ChatPay, Tata Neu's WhatsApp commerce.

Recommendation System / Recommender Engine — A system that predicts which products a user is most likely to want and surfaces them. Drives ~35% of Amazon's revenue and ~75% of Netflix watch time. Three classical approaches: content-based (similar items), collaborative (similar users), and hybrid (both). Modern systems use neural embeddings, vector similarity, and reinforcement learning to optimise for long-term engagement, not just next-click.

Collaborative Filtering — A family of recommendation algorithms that uses user-item interaction patterns — clicks, views, purchases, ratings — to infer that users similar to you will like similar items. Item-based collaborative filtering ("customers who bought this also bought") and user-based (similar users) are the two main flavours.

Content-Based Filtering — A recommendation approach that ignores other users and instead matches item attributes to a profile of what the user has liked — same genre, same brand, similar text description. Content-based works well at item launch (no waiting for others to interact) but tends to produce narrow, predictable recommendations.

Cold Start Problem — The well-known weakness of recommendation systems: a brand-new user (no history) or a brand-new item (no interactions) cannot be confidently recommended for or against. Workarounds include onboarding surveys (ask user to pick favourite categories), demographic priors, and content-based fallbacks until enough behaviour accumulates.

Embedding / Vector Representation — A way of representing an item or user as a list of numbers (a vector, often 64–1024 dimensions) such that similar items have nearby vectors. Embeddings are produced by neural networks and form the backbone of modern semantic search, recommendations, and visual search.

Vector Database — A specialised database (Pinecone, Weaviate, Qdrant, AWS OpenSearch with k-NN, pgvector) that stores embeddings and answers "find the 20 items closest to this vector" in milliseconds, even with billions of vectors. Vector DBs are what makes semantic search and modern recommendations practical at scale.

Semantic Search — Search that understands meaning, not just keywords. A user typing "comfortable formal shoes under ₹3,000" should get business-casual loafers and oxfords, even if the listings don't use the exact word "comfortable." Built on embeddings + vector search.

Visual Search — Search by uploading or pointing the camera at an image to find similar products in the catalog. Myntra, Pinterest, Flipkart Lens, and Google Lens all offer visual search; the technique is computer-vision embeddings + vector similarity.

Augmented Reality (AR) — Technology that overlays digital content (3D models, info) onto the real world, viewed through a smartphone camera or AR glasses. E-commerce uses include furniture placement (IKEA Place, Pepperfry), eyewear try-on (Lenskart), cosmetics try-on (Nykaa), and apparel try-on. Conversion lifts of ~70% and return reductions of ~25% are reported.

Virtual Reality (VR) — A fully immersive computer-generated environment, viewed through a VR headset (Meta Quest, Apple Vision Pro). Commerce use cases are nascent but include virtual showrooms (real estate, automobiles) and virtual try-ons.

ARKit / ARCore / WebAR — The three main AR delivery platforms. ARKit (Apple, iOS), ARCore (Google, Android), and WebAR (in-browser, no app install required). WebAR is the most accessible — users tap a link and AR works without downloading anything.

Big Data — Datasets so large (terabytes to petabytes), fast (millions of events per second), or varied that traditional relational databases cannot handle them. E-commerce big data sources include clickstream events, search queries, cart events, order data, customer support tickets, and reviews. The "three Vs" — Volume, Velocity, Variety — are the standard textbook framing.

Data Pipeline / ETL — The plumbing that moves data from operational systems (databases, app servers, third-party APIs) into a data warehouse for analysis. ETL (Extract-Transform-Load) is the older pattern; ELT (Extract-Load-Transform) is the modern cloud-native pattern. Tools: Apache Kafka for streaming, dbt for transformation, Airflow for orchestration.

Data Warehouse — A purpose-built analytical database optimised for large, complex queries over historical data. Snowflake, Google BigQuery, Amazon Redshift are the dominant cloud data warehouses. The warehouse is where business analysts run their queries, BI tools build dashboards, and ML teams build their training data.

Cloud Computing — On-demand access to compute, storage, networking, and managed services over the Internet, billed by usage. The three service models — IaaS, PaaS, SaaS — were detailed in Lesson 1.4 and again above. Cloud is the default infrastructure for modern e-commerce because of elasticity (handle Big Billion Day without buying servers), pay-as-you-go economics, and an enormous ecosystem of ready-made services.

Elasticity / Auto-Scaling — The cloud property that lets the number of servers (or other resources) grow and shrink automatically based on real-time demand. E-commerce sites rely on elasticity to absorb 10× spikes during sales without overpaying for idle capacity the rest of the year.

Multi-AZ / Multi-Region — Cloud resilience patterns. Availability Zones (AZ) are physically separate datacentres within a region; regions are geographically distant clusters of AZs. A "multi-AZ" deployment survives the loss of one datacentre; a "multi-region" deployment survives the loss of an entire geography. E-commerce sites that take payments typically deploy multi-AZ at minimum.

Dynamic Pricing — Algorithmic pricing where the price changes in real time based on demand, competitor prices, inventory level, time of day, and user-segment signals. Airlines (Indigo, Air India), hotels (MakeMyTrip, OYO), and ride-share (Ola surge, Uber) are the classic users; quick-commerce and general retail are adopting it via cloud-hosted pricing engines.

Demand Forecasting — Big-data and ML-driven prediction of how much of a product will sell, where, and when. Demand forecasting drives inventory placement (which SKUs in which dark store), discount planning, and supply-chain procurement. A 1% improvement in demand-forecast accuracy can be worth ₹100s of crores for a large retailer.

---

Study deep

1. Personalisation drives commerce. From Amazon's "you may also like" to Netflix's home feed, recommendation systems are the single most measurable AI contribution to commerce.

1. Chatbots have matured fast. Pre-2022 chatbots were rule-based and limited. Post-ChatGPT, LLM-powered support is industry standard. Yellow.ai, Haptik, Verloop and global players all integrate LLMs.

1. AR/VR is still emerging in India. Western brands (IKEA, Sephora, Nike) lead. Indian brands (Lenskart, Nykaa, Tata Motors) are catching up. Furniture and eyewear see the biggest impact.

1. Big data without action is useless. Many companies collect terabytes and learn nothing. The discipline is in turning data into decisions — pricing, inventory, marketing, product roadmap.

1. Cloud is becoming "boring infrastructure." Like electricity in 1920, cloud is now assumed. The differentiation moves up the stack — to services (Stripe for payments, Algolia for search, Twilio for SMS).

PYQ pattern (very common): "Discuss new technologies in e-commerce — chatbots, recommendation systems, AR, big data, cloud." — Define each, give one e-commerce use case, list one example tool. Connect them to better customer experience.