Digital Marketing Analytics

MasterCard as a Payment Network

MasterCard does not issue credit cards or lend money. It is a payment network – the invisible infrastructure that connects card issuers (banks) and merchant acquirers (banks that install point-of-sale machines) to authorize, clear, and settle every card transaction. Think of it as the digital railway that ensures money moves from a cardholder’s bank to the merchant’s bank in seconds.

Market Position and Financials (2013–2016)

In 2013 MasterCard was the global #2 payments network, holding 30% market share, behind Visa’s 45%. The remaining 25% was split among American Express (Amex), Discover, and regional players.

The business is extremely profitable (≈38% net margin in 2016). CEO Ajay Banga demanded that marketing spending be justified by measurable performance – a call for marketing accountability.

Key takeaways

• MasterCard is a pure-play payment network, not a card issuer.
• Market share: Visa 45%, MasterCard 30%, Amex/others 25%.
• Revenue grew ~29% over 4 years; net income ~30%.
• Marketing budget dropped from 10.1% to 7.6% of revenue – still large in absolute terms.
• CEO insisted on linking marketing spend to measurable returns.

The Payment Process: Open‑Loop vs. Closed‑Loop

Every card transaction involves multiple players taking a fee. The two dominant models are open‑loop (Visa, MasterCard) and closed‑loop (Amex, Discover).

Open‑Loop Flow (Visa / MasterCard)

• Cardholder pays the card issuer (e.g., SBI, HDFC) at month-end.
• Merchant receives payment from the merchant acquirer (a different bank that installed the POS machine).
• The network (MasterCard/Visa) handles the routing and data processing.

Fee Breakdown for a $100 Transaction

Exam tip: Open‑loop networks charge a smaller total fee (≈2%) than closed‑loop (≈2.45%). This is why some merchants in India used to refuse Amex – the higher fee ate into margins. However, Amex cardholders often spend more, so many merchants now accept it willingly.

MasterCard’s revenue model: It earns ~0.2% of every transaction processed. With $10.7 B revenue in 2016, the total value of transactions through MasterCard was:

$\frac{10.7 \text{ billion}}{0.002} = 5.35 \text{ trillion dollars}$

At 30% market share, Visa’s 45% share then implies Visa processed about $8 T in the same period.

Key takeaways

• Open‑loop: three separate fee earners (issuer, acquirer, network). Closed‑loop: one fee to the network/issuer.
• Merchant net ≈98% in open loop, ≈97.55% in closed loop.
• MasterCard earns only ~0.2% per transaction – high transaction volume drives revenue.
• The entire authorization cycle takes <3 seconds.

Stakeholders: B2B2C Model

MasterCard operates in a B2B2C structure:

• Business customers (B2B): Banks (issuers) and merchants (acquirers) – MasterCard must persuade them to use its network.
• End consumers (C): Cardholders like you and me – they choose a card based on the issuing bank and rarely notice the payment network.

Why MasterCard advertises to consumers

At the moment of payment (the “low point” of a shopping journey), the cardholder sees the bank’s logo – not MasterCard. MasterCard therefore runs brand campaigns to:

• Build brand salience so consumers prefer cards carrying the MasterCard logo.
• Influence banks and merchants indirectly: if consumers demand MasterCard, banks and merchants will list it.
• Make the payment experience feel priceless rather than transactional.

Marketing Challenges and the “Priceless” Campaign

MasterCard’s Priceless campaign launched in 1997 and was still running in 2012 (now a 25+ year franchise). The CMO, M. V. Rajamannar (former Citibank executive), refocused it on four priceless possibilities – unique, digital-driven experiences that connected cardholders with their passions.

The core tagline:

“There are some things money can’t buy. For everything else, there’s MasterCard.”

The campaign aimed to:

• Differentiate MasterCard in a category where cards look identical.
• Deepen partnerships with banks and merchants by co‑creating memorable experiences.
• Leverage digital technology and social media to intensify linkages between all three stakeholders.

Why “priceless”? The emotional payoff (e.g., spending time with a teenage son) is invaluable – that association makes the brand more than just a transaction facilitator.

Key takeaways

• MasterCard’s marketing must influence three separate groups: consumers, banks, and merchants.
• The Priceless campaign gave the brand an emotional hook in a functionally undifferentiated space.
• CMO Rajamannar shifted toward digital, experiential marketing to build loyalty and differentiate from Visa.
• Despite being a B2B network, consumer advertising is critical for brand preference and partner selection.

MasterCard’s Leveraging of Social Media: The Digital Engine

MasterCard transformed its marketing by building a Digital Engine – a data-driven, real-time system that blends content, transactions, and partnerships. The core idea: move from storytelling (the brand pushing a narrative) to story-making (inviting consumers to co-create and share stories). The engine has seven interdependent components, all aimed at driving card usage, merchant value, and brand preference.

The Digital Engine’s Seven Components

The process is continuous, iterative, and built on real-time optimisation.

Example Campaigns

New Year’s Eve (NYE) Campaign – US (Hugh Jackman)

• Spark video: Hugh Jackman reunites with his mentor.
• Sequence: Spark video → Priceless Surprise video → Winner videos (user-submitted stories).
• Media valuation (cost if done traditionally) was $12.3 million; actual spend lower (~$4–6M).
• Outcome: Reached 50 million consumers, 19 million engagements (38% engagement rate), 3.8 million video views, several million qualified leads.

India – HDFC Independence Day Campaign (targeting Indian tourists to Singapore at Christmas)

• Insight: Families with young kids travel to Singapore, big spenders.
• Merchant partner: Resorts World Sentosa; used Minions characters to appeal to children.
• Performance:
  • Merchant click-through rate 218% higher than market benchmark.
  • Sentosa saw 39% growth in spend and 55% growth in transactions vs. same quarter previous year.

Measuring Impact: The Critical Role of Test vs Control

A single metric in isolation is meaningless. For example, comparing year-over-year growth (e.g., 39% spend increase) could be due to rising incomes. You need a test-and-control (lookalike) group measured over time.

MasterCard’s approach: longitudinal measurement over 3–4 years.

Example: Priceless Cities US Campaign (3-year study)

Exam tip: A campaign’s true effect is shown by the incremental lift compared to a matched control group, measured before, during, and after. The persistence of lift post-campaign indicates lasting behaviour change.

Real-Time Optimisation Through Multiple Metrics

Different objectives call for different KPIs. A single campaign (e.g., a travel offer) can be tested with multiple ad variations:

• The food offer had highest click rate but low engagement.
• The Tuk Tuk offer had highest engagement rate but lower click rate.
• Best practice: balance both metrics based on campaign goals. Use A/B testing to continuously reallocate budget to the best-performing creative.

Brand Perception & Soft Metrics

MasterCard supplemented transactional data with surveys (n≈600 random respondents):

• 62% said Priceless campaigns “greatly” or “somewhat improved” perception of MasterCard brand.
• Merchants and bank partners also reported positive impact.

Key Learnings from MasterCard’s Digital Engine

1. Emotional spark must be context-driven: three factors – appeal to target audience, tap into cultural trends, and align with brand image.
2. From storytelling to story-making: involve consumers to break through clutter.
3. Budget allocation is continuous, not annual or quarterly – fueled by real-time data.
4. Measure the full cycle: reach → engagement → qualified leads → conversion → incremental spend and transactions.
5. Combine art and science: face challenges like attribution and short-term vs long-term effects; perfect ROI is elusive.
6. Win-win for all stakeholders: MasterCard, merchants, banks, and customers must all benefit.

Key takeaways

• The Digital Engine is a seven-step cycle: emotional spark → engagement → partnerships → real-time optimisation → programmatic buying → network effects → measurement.
• Test-and-control (lookalike) groups are essential to isolate campaign impact; longitudinal data (before, during, after) proves causality.
• Real-time A/B testing on digital platforms allows continuous budget reallocation.
• Soft brand metrics (awareness, perception) complement hard transaction data.
• MasterCard’s revenue grew from $10B (2016) to$24B (2024), partly driven by this approach.

Lenovo Digital Charter

Lenovo (computers, laptops, tablets, storage) faces a long purchase cycle — customers keep products for 3–5 years. Digital marketing must therefore acquire, retain, engage, and influence the entire customer journey, not just the moment of sale. The company formalised this through a Digital Center of Excellence (COE) Charter built on four pillars.

Four Pillars of the Digital COE

Innovation & Optimisation Framework

Digital initiatives were classified along two dimensions: innovation type (disruptive vs. sustaining) and resource nature (investments vs. revenue delivery).

Campaign portfolio optimisation: Each campaign was evaluated on two axes – cost efficiency and contribution. This drove a continuous cycle of measure → optimise → improve.

Exam tip: The 2×2 matrix (cost efficiency × contribution) is a classic resource allocation heuristic. Be able to explain when to pause vs. invest.

Customer Experience (CX) Tracking

Lenovo formalised CX measurement using two primary KPIs:

Net Promoter Score (NPS) = % Promoters – % Detractors
Customer Satisfaction (CSAT) = average rating on a satisfaction scale (e.g., 1–5)

The trend showed NPS fluctuating (e.g., ~40 in April, dropping then rising by August) while CSAT stayed stable around 4 (slight dip in June). 32 metrics were tracked monthly across the customer decision journey, providing actionable insights for regional managers.

Project Zero Friction aimed to measure CX in real time at 23 touchpoints across web, phone, email, brand retail, and service engagements. At each touchpoint, a two‑question survey was triggered:

1. Rate the current experience.
2. How likely are you to recommend? (NPS) and satisfaction (CSAT).

Three‑stage measurement process:

1. Baseline study (qualitative, 3–4 weeks) – in‑depth interviews to identify important touchpoints and parameters.
2. Real‑time KPI tracker (4–5 weeks) – periodic, real‑time measurement at touchpoints; create monthly dashboards.
3. Diagnostic study (as needed) – deep dive on sub‑parameters influencing each touchpoint’s score.

Mapping to the Customer Decision Journey

Touchpoints were mapped across stages (need recognition, assessment, research, purchase, post‑purchase) and channels (non‑digital, digital, direct, retail).

Exam tip: The touchpoint map is a direct application of the customer decision journey model. Expect to be asked how to identify friction points and design surveys for each stage.

Key takeaways

• Lenovo’s Digital COE charter rests on four pillars: social media, discoverability, e‑commerce, and CRM/personalisation.
• Innovation is split into disruptive (transformation, investment) and sustaining (performance, revenue).
• Campaigns are continuously evaluated on cost efficiency vs. contribution to decide pause, review, or invest.
• CX is tracked via NPS and CSAT at 23 real‑time touchpoints using a three‑stage measurement process (baseline, real‑time, diagnostic).
• Touchpoint mapping across the customer decision journey enables granular friction elimination (Project Zero Friction).

Customer Experience Framework (Lenovo Case Study)

Customer experience (CX) is the sum of all interactions a customer has with a brand. Lenovo’s CX framework organises the discipline into five pillars that together create a system to deliver remarkable customer experiences in a hyper‑competitive market.

Five pillars of the CX framework

These pillars are often mapped to an action cycle: Align → Measure → Understand → Communicate → Govern → Embed & Engage.

Hyper‑local targeting with Google My Business (example)

Lenovo drives traffic to retail stores by combining search ads with location data.

• Goal: Target users looking for PCs/laptops near stores → store walk‑ins or call‑ins.
• Mechanism: Use Google My Business (GMB) data feed to list all retail stores; run search ads with location extensions and click‑to‑call extensions.
• Challenges with multi‑brand stores: Partner stores carry competing brands; must ensure Lenovo is chosen at the point of purchase.
• Store‑discovery requirements: Accurate data (addresses, directions), managed ratings & reviews, support for small retailers (adverts, forwarding numbers, training).

Result: 450+ Lenovo exclusive stores and service centres mapped; data from GMB synchronised into a single intuitive listing on buylenovo.com.

Why digital is needed for retail

1. Target the right consumer – Reach a larger, segmented audience; design specific communication (e.g., back‑to‑school campaigns for parents in June/July).
2. Drive to retail – Increase online store discovery, requests for demo, call‑ins, and walk‑ins.
3. Location‑specific communication – 51% of PC queries are on mobile with location‑based targeting (study cited).
4. Measurable insights – Track leads to action rates, answer rates, and conversions.

Tone of Voice (ToV) methodology

Lenovo’s tone of voice defines how the support team (LenCare) communicates, balancing the brand personality (unexpected, brave, confident) with the customer’s context (facing an issue).

Scoring process: Randomly pick 50 responses per period; score each on the four dimensions. Results are plotted as percentages (e.g., 70% of responses scored as formal in the formality dimension). Weekly workshops, training, role‑plays, and real‑time feedback sessions refine the team’s tone.

Voice of Customer (VOC) monitoring – example data

• Keywords tracked: LenCare, Lenovo Care, Lenovo Support, Lenovo Service.
• Time range: March 2019 – April 2021.
• Metrics: Reach (622M mentions during Q1 2021 peak), mentions (30K–29K range), positive/negative/neutral sentiment.
• Trend: Percentage positive increased over time; percentage negative initially high then declined; neutral fluctuated.

Exam tip: The Lencare example shows that CX extends beyond purchase – most customer experience is determined during months and years of product use. Digital marketing strategy must cover the entire customer journey, not just acquisition.

Key takeaways – CX Framework

• Lenovo’s CX framework has five pillars: Strategy, Metrics, Understanding, Delivery, Culture.
• Hyper‑local targeting with GMB data drives foot traffic to retail stores.
• Tone of voice (humour, formality, respectfulness, emotiveness) is systematically scored and trained.
• VOC monitoring with sentiment analysis helps track service quality over time.
• Customer experience is determined throughout the product lifecycle, not just at purchase.

---

Performance Marketing

Performance marketing focuses on measurable, short‑term outcomes – clicks, leads, and sales – driven by digital tools. It is the digital evolution of return on marketing investment (ROMI).

Brand marketing vs. performance marketing

Blending both: The best strategy combines brand and performance. Nike uses storytelling for brand, while Google Shopping ads drive conversions. Croma uses search ads for sales and YouTube tutorials for brand building (to promote its own private brands).

Defining goals for digital presence

Before any digital investment, ask: Why am I online? Possible goals:

Conversion depends on context:

• Domino’s → online order.
• Apollo Hospitals → appointment booking (gynecologist, health checkup).
• NGO → donation or volunteer sign‑up.

All performance marketing must define the target action and measure its success.

Exam tip: Performance marketing is not a replacement for brand marketing – they work together. The key is to align the metric (conversion) with the business model (e.g., e‑commerce → purchase; lead gen → form submission).

Key takeaways – Performance Marketing

• Performance marketing = measurable, short‑term digital marketing (clicks, leads, sales).
• Contrast with brand marketing which builds long‑term equity.
• Examples: search ads, display banners, retargeting, affiliate marketing.
• Goals vary by business: sell, generate leads, awareness, information sharing.
• Conversion is context‑dependent (order, appointment, donation).
• Best practice blends brand and performance marketing for full‑funnel impact.

Performance Marketing Framework

A performance marketing framework structures online marketing investments into a three-stage cycle: attraction → conversion → retention. This mirrors the traditional retail funnel: footfall → purchase → repeat visit. Each stage has distinct goals, metrics, and optimisation levers.

Attracting Traffic

Why traffic matters – without visitors no conversion is possible. Traffic volume directly creates sales opportunities, enables testing (more data points), and improves SEO ranking through activity, relevance, and site quality.

Key question: Do I have enough traffic?
Benchmark using tools:

• Google Keyword Planner – estimates demand and competitor bids
• Google Trends – shows seasonal/interest fluctuations

Types & Sources of Traffic

Paid vs. unpaid: Paid is quick, scalable, costly; unpaid is slower but sustainable. Even Amazon, dominant in organic traffic, still spends heavily on paid ads.

Assessing Traffic Quality

Volume alone is insufficient – must attract the right visitors.

• Bounce rate – % who leave without action. Low bounce rate = good engagement.
• Pages per visit – more pages indicate deeper interest; average 2–4.
• Time spent – engagement depth.
• Returning visitors – loyalty indicator; healthy range 30–35%.
• Geographic origin – must align with target market. A Bangalore cafe with 70% US traffic wastes money.

Key takeaways

• Without traffic, no conversion is possible.
• Traffic sources split into direct, organic, paid, display, referral, social, email.
• Quality metrics (bounce rate, pages/visit, time, returning %) matter more than raw volume.
• Paid traffic = fast but costly; organic = slower but sustainable.

Converting Traffic

Conversion turns visits into measurable actions: purchases, downloads, sign-ups.

Core Metrics

• Conversion rate – % of visitors who complete a desired action.
• Cost per acquisition (CPA)
  $\text{CPA} = \frac{\text{Total spend}}{\text{Customers acquired}}$ Example: ₹10,000 spend, 100 customers → CPA = ₹100.
• Customer lifetime value (CLV)
  $\text{CLV} = \text{Margin per customer} - \text{CPA}$ Example: Margin ₹10,000 over 24 months, CPA ₹1,000 → CLV = ₹9,000.

The Attribution Problem

Attribution determines which channel receives credit for a conversion. Customers interact with multiple touchpoints – a display ad may create awareness, a search ad completes the sale. Common models:

Exam tip: Last-click attribution is the most common in practice but often underweights display and social. Always consider multi-touch attribution to avoid misallocating budget.

Diagnosing Low Conversion

Solutions: A/B testing, benchmark competitors’ offers, invest in site performance (speed, mobile optimisation).

Key takeaways

• Conversion = turning visits into desired actions.
• CPA = spend ÷ customers; CLV = margin – CPA.
• Attribution is the central challenge – no single model is perfect.
• Low conversion stems from traffic quality, UX, value proposition, or market size.

Retaining Customers

Without retention, the business is a leaky bucket: acquiring a new customer costs 5× more than retaining an existing one.

Tactics

• Loyalty programs – e.g., Starbucks Rewards
• CRM-based re-engagement – tools like HubSpot, Zoho
• Content marketing – keep customers coming back
• Email nurturing campaigns – reminders, cross-sells, upsells

Goal: increase repeat purchase frequency, loyalty, and value per customer.

Improving CLV

CLV = Margin – CPA. Levers to pull:

1. Reduce CPA – better targeting, cheaper channels
2. Extend relationship duration – keep customer longer (e.g., 2 → 3 years)
3. Increase purchases per period – encourage higher frequency
4. Increase margin per purchase – larger order size, cross-sell, upsell

Amazon Prime example:

• Annual membership ~₹1,500 India, ~$100 globally
• Prime members buy 4× more than non-Prime
• 200 million members generate ~$20 billion in subscription fees alone

Diagnosing Retention Problems

Solutions: Strengthen CRM (Salesforce, HubSpot, Zoho); conduct cohort analysis to study churn behaviour; personalise communications (birthday emails, abandoned cart reminders, tailored discounts).

Key takeaways

• Retention is 5× cheaper than acquisition.
• CLV improvement levers: lower CPA, longer duration, higher frequency, larger margin.
• Loyalty programs, CRM, and personalisation drive repeat business.
• Cohort analysis reveals why customers churn.

---

Digital Marketing Performance Gap

Despite rising digital spend (57% of marketing budgets per 2022 CMO survey), over 30% of customers report low or no ROI. This is the digital marketing performance gap.

Why the Gap Exists

• Lack of integrated digital marketing across the organisation – silos between retail and digital.
• Steep learning curve in data analytics – failure to turn data into actionable metrics.
• Over-reliance on vanity metrics (likes, follower counts) instead of customer-journey–aligned measures.
• Complex, multi-channel customer journeys – especially when physical channels are involved.
• Third-party data loss due to privacy regulations (GDPR, India’s DPDP, upcoming US rules).
• Outsourced agency dependency – slows agility; CMOs increasingly bring analytics in-house.

Closing the Gap

Exam tip: The digital marketing performance gap is a recurring theme – marketers must move from “spend heavy” to “performance driven.” First-party data + AI is the winning formula.

Key takeaways

• Digital spend is high but ROI often disappoints.
• Causes: silos, data illiteracy, vanity metrics, privacy loss, agency over-reliance.
• Close the gap via experimentation, cross-functional teams, first-party data, and AI.

---

Marketing Performance Measurement (MPM)

Marketing Performance Measurement (MPM) is a framework of metrics used to assess marketing performance within budget. Currently many firms track only short-term, ad-hoc metrics (sales volume, clicks); only 40% link marketing metrics to revenue.

Risks of poor MPM:

• Harder to secure budgets (CFO demands evidence).
• Misses long-term brand value – strong brands command price premiums and higher firm valuations.

Five Essentials for Effective MPM

1. Align metrics with strategy – “What gets measured gets managed.” Prioritise metrics tied to business goals (e.g., landmark group’s CRM aligned marketing to corporate outcomes).

2. Measure desired outcomes, not just financials – Include brand equity, customer experience (e.g., English Premier League tracks fan behaviour and equality initiatives, not just profits).

3. Establish cause and effect – Metrics must be connected, not isolated. Adidas refined NPS into a “brand health NPS” to link service to revenue.

4. Triangulate metrics – Cross-validate with multiple data sources (e.g., combine behavioural and firmographic data for better lead scoring).

5. Reject vanity metrics – Avoid flattering but empty numbers (blog likes, follower counts). Focus on metrics that signal customer equity growth.

Exam tip: Vanity metrics (e.g., page views without conversion) look good in dashboards but mislead. Always ask: “Does this metric link to revenue or customer value?”

Key takeaways

• MPM is a systematic dashboard – not just fragmented click-throughs.
• Effective MPM aligns with strategy, measures outcomes, shows causality, validates data, and rejects vanity.
• Strong MPM proves value to the C-suite, secures budgets, and drives long-term growth.

Marketing Performance Measurement

Marketing performance measurement evaluates how effectively digital and offline data are used to segment audiences, build personas, and drive acquisition and retention. The core challenge: scattered customer data across many touchpoints. A real case from a large life insurance company illustrates the process.

The Problem: Data Silos & Fragmented View

A 20+ year-old insurance company with 30+ platforms and tools had dispersed customer data across backend systems, digital channels (websites, apps), call centers, and agent networks. This prevented a unified understanding of customers, hindering efficient acquisition and personalized communication.

Key challenge: Create a single, unified customer view from offline (CRM, call center) and online (web analytics, social media, ad platforms) data.

The Solution: Audience Segmentation via Machine Learning

The company used a partner’s collect → collate → activate framework with triangulation algorithms to merge data from multiple touchpoints. The goal was to identify valuable customer cohorts beyond basic demographics.

Process flow:

Key analytics steps:

1. Feature selection – 39 distinct parameters captured from user points across devices.
2. Cluster analysis – Used elbow method, silhouette method, and hierarchical clustering to determine optimal number of clusters.
3. Validation – Internal measures: silhouette coefficient, different indexes.
4. Output – For each segment: likelihood of churn (low vs. high loyal), one-time discount seekers, economic frequent buyers, etc.

Data Sources & Parameters (39 parameters)

The Four Personas (Cohorts)

The model produced four distinct audience segments:

Exam tip: The percentages add to 103% – this is acceptable rounding in segmentation output. The key insight: the largest segment (55%) is young and budget-conscious, yet the 5% “digital savvy” segment may be the most responsive to online-only campaigns.

Strategic Applications

These personas enable:

• Product design – Tailor products (e.g., low-premium plans for “Young & Budget Conscious”).
• Media planning – Target lookalike audiences on digital channels with similar behavior.
• Communication – Personalized messaging via call center, email, or agents.
• Retention – Identify churn risk segments and design re-engagement campaigns.

Key takeaway: The company moved from “insurance is sold” (agent-driven) to data-driven acquisition by combining digital analytics with offline data, creating a unified customer view.

Key takeaways

• Marketing performance measurement requires unifying scattered data (online + offline) into a single view.
• Audience segmentation uses ML (clustering) with 39+ parameters to go beyond demographics into psychographic and monetary insights.
• Four distinct personas emerged: Young & Budget Conscious (55%), Midlife Engaged (28%), High Net Worth (15%), Digital Savvy (5%).
• These personas drive product design, media planning, and personalized communication.
• The process: collect → collate → triangulate → activate across channels.

Analytics in Marketing

Why bother with analytics? Traditional marketing relied heavily on gut feel. Digital channels now capture nearly every customer touchpoint, making it possible to take data-driven decisions in three critical areas:

• Optimize marketing spend – allocate limited budgets across display, social, search, and other digital channels.
• Understand customers deeply – move beyond one-size-fits-all campaigns to segmentation and personalization.
• Personalise the experience – tailor messaging, offers, and timing to individual behaviour.

Exam tip: The three pillars – spend optimization, customer understanding, and personalization – are the standard “why” for marketing analytics. Expect to be asked to list or explain them.

Sources of Digital Data

Digital data comes from multiple owned and operated touchpoints:

Even a simple POS receipt yields rich data:

• Product bought (e.g., pizza + toppings)
• Quantity and total spend
• Location and time of purchase
• Order type (takeaway, dine-in, delivery)

When the order is placed via an app, additional behavioural data becomes available:

• Clicks, cart additions/removals, and what was viewed but not clicked (negatives signal irrelevance).

These data points allow computation of customer-level metrics:

• Recency – when did the customer last buy? (long gap = lapsed customer)
• Frequency – how many orders placed in a given period
• Average order value – total revenue per order

Types of Customer Data

Modern marketing distinguishes four data categories based on ownership and consent:

• First-party data is smaller in volume but most accurate and owned outright.
• Third-party data is large but less reliable in freshness and detail.
• Zero- and first-party data are increasingly critical as data privacy regulations tighten (e.g., cookie consent, GDPR).

Customer Segmentation: Behavioural Clusters

From transaction and browsing data, marketers identify distinct customer motivations:

• Habit-driven – orders the same product repeatedly (e.g., same pizza every time).
• Deal-driven – first clicks on offers/promotions; motivated by discounts (BOGO, 50% off).
• Novelty-driven – tries different items, explores new products.

Additional segmentation based on timing:

• Weekend vs. weekday customers
• Lunch vs. dinner customers
• Ritual customers – e.g., buy every Wednesday after a recurring activity (swimming class).

These segments inform message motivation: some respond to crave imagery (close-up food shots), others to promotions or family consumption scenarios.

Worked Application: Fast-Food Chain (KFC / Pizza Hut / McDonald’s)

A company can use order data alone to:

1. Identify recency, frequency, and average order value for each customer (linked via phone number).
2. Classify customer as habit, deal, or novelty driven.
3. Combine with campaign response data to learn which creative style resonates.
4. Target lapsed customers with win-back offers; frequent customers with loyalty rewards.
5. Cross-promote – e.g., a lunch customer targeted for dinner using behaviour-driven personalised messaging.

Overlaying Data for Precision: Example with Tata Sky

A satellite TV provider (Tata Sky) knows its subscribers but does not know their social media interests. By uploading a hashed list of subscriber phone numbers to Facebook/Instagram (privacy-safe matching), the platform identifies those who have not subscribed to sports channels. With Instagram’s data on pages liked (e.g., tennis pages), the provider can serve sport-specific ads to non-subscribers, increasing relevance and conversion.

Exam tip: The Tata Sky example illustrates how first-party data can be enriched with third-party data (via platform matching) to improve ad targeting without sharing raw customer data. This is a high-yield concept for questions on data integration and privacy.

Key takeaways

• Analytics replaces gut feel with data-driven decisions on spend, personalisation, and customer understanding.
• Digital data flows from campaigns, owned properties, e-commerce, and POS – even a simple receipt yields recency, frequency, and order value.
• Zero-party (knowingly given), first-party (observed behaviour owned by company), second-party (partner-sharing), and third-party (aggregated) data differ in ownership, accuracy, and volume.
• Customers segment into habit-driven, deal-driven, and novelty-driven – each responds to different creative and offers.
• Overlaying first-party data on third-party platforms (e.g., Facebook) enables precise targeting without violating privacy, as shown in the Tata Sky example.

Data Analysis Techniques

A marketing analyst’s toolkit spans from simple statistics to machine learning. Each technique answers a different question: What causes sales? Who are my customers? What will they do next? What do they buy together? What are they saying about me? The choice depends on the data available and the business decision at hand.

Overview of Techniques

---

1. Causal Techniques (Regression)

A regression model estimates the relationship between an independent variable (e.g., price, ad spend) and a dependent variable (e.g., sales). It answers “if I change X by one unit, how much does Y change?”. In marketing, it is used to allocate budgets across media combinations by predicting the expected lift in sales.

• Example: “By changing price, I am able to get so much of a response in terms of sales increase.”

---

2. Clustering

Clustering is a machine‑learning technique that partitions a large set of customers into groups (segments) based on similarity across many variables. Unlike manual segmentation using a few columns in Excel, clustering handles dozens of columns and millions of rows automatically.

• How it works (K‑means): The algorithm iteratively assigns each customer to the nearest cluster centre, then recalculates centres until stable.
• Output: A list of clusters; human analysts then interpret and name them (e.g., “Family Stylist”, “One Percenter”).

Exam tip: Clustering is unsupervised – no labelled outcome. It reveals patterns the analyst may not have anticipated. Regression is supervised – it requires a known outcome variable.

---

3. Predictive Modeling

Predictive modeling uses a training set (where the outcome is known) to build a function that predicts the outcome for new, unseen data. For example, from a sample of customers whose product preferences are known, the model assigns a likelihood score to each product for the entire customer base.

• Key idea: From known to unknown. The model learns patterns in the known data and extrapolates.

---

4. Association / Market Basket Analysis

Association analysis (often called market basket analysis) finds combinations of items that appear together in transactions more often than by chance. It generates rules like “if item A is bought, item B is bought with high probability”.

• Real‑world examples: Amazon’s “Frequently bought together”, Netflix’s “Because you watched…”.
• Business use: Bundle products, design cross‑sell campaigns, optimise shelf placement.

---

5. Recommendation Engines

A recommendation engine predicts what a user will like, typically using rating data. Two common approaches:

• User‑based collaborative filtering: Find users similar to the target user; recommend items those similar users rated highly.

• Item‑based collaborative filtering: Find items that are rated similarly across users; recommend items similar to ones the user already liked.

• Distinction from market basket analysis: Recommendations often rely on explicit ratings (e.g., 1–5 stars) rather than just co‑purchase. In practice, both appear together on e‑commerce sites.

---

6. Text Mining (Natural Language Processing)

Text mining extracts insights from unstructured text. With modern Natural Language Processing (NLP) , marketers can:

• Perform sentiment analysis – classify reviews as positive, negative, or neutral.

• Run topic modelling – automatically discover themes (e.g., “quality”, “delivery”, “size”) from thousands of reviews.

• Monitor brand mentions and detect emerging issues.

• Example: A product manager mines a huge set of customer reviews to find “where my product is okay, where my product is not okay”.

---

7. Image Mining

Image mining (or computer vision) analyses visual features – colour, composition, objects – to predict user behaviour. Marketers use it to select images that maximise click‑through rates.

• Example: Netflix tests thousands of thumbnail images for a show. Image mining predicts which thumbnail will get the highest engagement for a given user segment.

---

Use Cases in Practice

Decathlon: Customer Segmentation from Search & Browsing Behaviour

• Data sources: Search keywords (on‑site and in ad campaigns) and filters applied on the website.
• Insight extracted: A “newbie” searches for “how to choose running shoes” and uses beginner‑friendly filters; a “pro hiker” searches specific product names and uses expert filters.
• Segments formed: New hikers, urban recreational (know what they want but easy terrain), pro hikers.
• Action: Tailor communication (content, offers, product recommendations) for each segment – no longer “one size fits all”.

Ralph Lauren (Southeast Asia): Point‑of‑Sale Segmentation with K‑Means

• Data: Two years of transaction data with membership info (gender, purchase history across categories). Rolled up per customer: what they bought (men’s, women’s, children’s), frequency, monetary value.
• Method: K‑means clustering on 20–30 columns and ~1 million customers.
• Segments discovered (examples):
  • Family Stylist – female member buying mostly children’s and men’s clothes, very little for herself. Opportunity: encourage self‑purchase.
  • One Percenter – tiny segment (1% of customers) that shops only once or twice a year but spends 25× the average. Later identified as Chinese tourists on annual shopping trips. Action: time marketing precisely before their expected arrival.
• Outcome: Only two segments were prioritised due to limited budget and growth potential; differentiated messaging and timing were designed for each.

Meal Prep Company (UK): Google Analytics Age‑Group Analysis

• Context: A fine‑dining restaurant pivoted to meal‑prep delivery during COVID. Website traffic was high, but conversions were lower than expected.
• Tool: Google Analytics (free).
• Analysis: Compared all website visitors vs. purchasers by age group.
  • Finding: Purchasers skewed older (45+); visitors skewed younger. Bounce rate (users who leave without any interaction) was similar across age groups.
  • Interpretation: The landing page wasn’t turning away younger users, but something deeper (menu, experience, vibe) was not resonating with them.
• Business decision: Either redesign the menu/experience to appeal to younger visitors, or shift targeting to attract more older (converting) visitors.
• Key principle: Simple behavioural data (age, bounce rate) from a free tool can reveal a misalignment between ad‑spend targeting and actual conversion.

---

Key takeaways

• Marketing analytics techniques range from regression (causal) to clustering (unsupervised segmentation) to predictive modelling, market basket analysis, recommendation engines, text mining, and image mining.
• Clustering is essential when data has many dimensions and millions of customers – it discovers patterns that manual Excel analysis cannot.
• Google Analytics (free) provides instant insights on customer demographics, behaviour, and conversion – even with AI‑generated insight summaries.
• Use cases show how same technique (e.g., clustering) can yield very different segments (Decathlon’s hobby‑based vs. Ralph Lauren’s family‑role‑based) depending on data source and business context.
• The ultimate goal is to move from one‑size‑fits‑all to relevant, segmented communication – and to decide not only how to target but whether a segment is worth targeting at all.

Enhancing Customer Experience with Data

Experience optimization is the strategic use of analytics to deliver a seamless, coherent, and relevant experience across every customer touchpoint—website, app, email, call center. The goal: ensure messaging, product recommendations, and offers match each customer’s context and behavior. This moves beyond static segmentation into real-time, personalized interactions.

Predictive Modelling for Personalization: Huggies Diaper Case

A critical moment in a customer’s lifecycle is the transition from one product variant to the next (e.g., diaper size change at 3–6 months). Huggies needed to break Pampers’ stronghold by winning customers at that transition point.

Problem: Huggies had a 4‑million database of new mothers (with data on first vs. second baby, location, ethnicity, etc.) and three diaper variants – economy (low margin), mid‑level, and premium (high margin). The goal was to send the right variant’s coupon to each mother to maximize revenue and margin.

Method: Use predictive modelling. On a smaller set of customers where past behaviour was known (which variant they used), a model learned patterns linking customer attributes to variant choice. That model was then applied to the remaining 4 million to generate a likelihood score for each person for each variant (Level 4, 5, 6). The variant with the highest score was selected; ties resolved by sending the higher‑margin one.

Exam tip: This is a classic multi‑class classification problem. The feature set (baby order, demographics) drives the probability for each class. The decision rule is argmax with margin‑based tie‑breaking.

A/B Testing to Validate Lift

To confirm the model’s impact, a random holdout group received coupons without personalization. The model‑driven group showed a lift – higher conversion/revenue – versus the random group.

Layered Personalisation: Beyond Product Variant

The same campaign added more layers of personalisation into a single email:

Technology assembles all these elements on‑the‑fly, producing thousands of unique email variants – one per customer – to maximise relevance and conversion.

Market Basket Analysis & Recommendation Engines

Market basket analysis finds associations between products (e.g., “people who bought X also bought Y”). Recommendation engines use these associations plus collaborative filtering (e.g., “other customers like you also liked Z”) to suggest items. They solve the infinite shelf problem – customers in digital stores cannot browse unlimited products, so algorithms surface the most relevant few.

• Amazon attributes 30–50% of sales to click‑through on recommendations.
• Same logic powers OTT platforms (Netflix suggests genre‑based and similar‑user favourites).
• Even physical stores use these insights for merchandising (e.g., placing chocolate near children’s products to increase basket size).

Microsegmentation: Tactical Personalisation with Lego Blocks

Microsegmentation is a more granular, campaign‑level version of segmentation. Instead of broad strategic segments, it tailors each communication to an individual using multiple attributes.

How it works (Lego‑block model):
Decompose a message (email or ad) into components (e.g., header, main image, sub‑headline, call‑to‑action, offer). For each component, create several options. With 5 components × 3 options each → $3^5 = 243$ possible combinations. Software merges these based on customer data.

KFC example:
A customer profile shows:

• Solo buyer (low average order value)
• Deal‑driven
• Prefers takeaway

Then the email assembles:

• Image: small ₹69 burger (not large bucket)
• Offer block: discount deal
• Call‑to‑action: “Order takeaway now”

Another customer – group, high‑spending, delivery – gets a large‑bucket image, premium offer, and delivery CTA.

Exam tip: Microsegmentation requires two components: operationalised data (customer profile) and a communication engine (Adobe, Salesforce, or platform‑level tools like Google/Meta Ads) to assemble and serve the right variant instantly.

Beyond First‑Party Data: Ogilvy Germany Rail Case

When a company lacks direct customer data (not D2C), creative third‑party data and analytics can still power deep personalisation.

Challenge: German Rail wanted to boost domestic tourism, but Germans preferred holidays abroad.

Solution (Ogilvy Germany):

1. Image mining – Identify the most shared images of foreign landmarks.
2. Match – Find visually similar locations within Germany (e.g., a castle that looks like Neuschwanstein, but in a different region).
3. Facebook data – Determine which foreign location a user is interested in.
4. Price comparison – Scrape airfare to the foreign destination vs. rail fare to the matching German location.

Result: Each user sees an ad: “Fly to X costs €Y. Take German Rail to this similar place in Germany for only €Z.” Persuasive, relevant, and using only public/Facebook data – no first‑party profile required.

Key takeaways

• Experience optimization uses analytics to deliver seamless, personalised interactions across channels.
• Predictive modelling (e.g., variant propensity scores) drives product‑ and retailer‑level personalisation, validated by A/B testing.
• Market basket analysis and recommendation engines solve the infinite‑shelf problem, boosting basket size and conversion.
• Microsegmentation treats message components as Lego blocks, assembled per customer attributes (e.g., solo vs. group, deal‑prone, channel preference).
• Creative data sourcing (image mining + social media data) can personalise even without first‑party transaction data.
• The sophistication of personalisation is limited only by available data, technology, and imagination in applying insights.

Budget Allocation for Campaigns

Budgets are always constrained. The central challenge is maximizing return from a fixed marketing budget. This breaks into two distinct sub-problems:

1. In-flight campaign optimization – how to allocate spend across versions/channels within a running campaign to maximise conversion.
2. Strategic budget allocation – how to distribute a limited total budget across multiple campaigns or channels, typically reviewed quarterly or annually.

---

Campaign Optimization (In-Flight)

The goal is to shift money toward ads and channels that deliver the real business outcome, not just easy-to-measure intermediate metrics.

The Vaccine-Ad Example: Vanity vs. True KPI

A paediatric vaccine campaign in Southeast Asia ran five social‑media ads (mother‑child, father‑child, etc.). The social‑media team optimised on click‑through rate (CTR) because it was immediately available on the platform dashboard. They allocated most budget to Ad 2 (CTR ~3.4%), which showed a calm mother‑child image.

However, the true KPI was conversion – users clicking a “find a clinic” or “book an appointment” button. When website data was joined with media data, Ad 4 (child crying) had the highest conversion. The crying child’s emotional realism may have deterred casual clicks but resonated with parents already experiencing the problem, making them more likely to take action.

Exam tip: This illustrates the danger of vanity metrics – metrics that look good but don’t correlate with business outcomes. CTR is a classic vanity metric if conversion is the goal.

The Full-Funnel View

Optimising on CTR alone ignores downstream steps. A full‑funnel view requires connecting media‑server data (impressions, clicks, spend) with website‑analytics data (visits, actions, conversions). This is done via UTM tagging – appending campaign parameters to URLs so that Google Analytics can attribute website actions to the specific ad, channel, and creative.

• Without proper tagging, media and website data remain siloed; optimisation can only be upper‑funnel.
• Tagging is simple but tedious, often neglected in the rush to launch campaigns.
• The next campaign for a different vaccine invested heavily in tracking infrastructure, enabling far richer optimisation.

---

Multi‑Campaign Budget Allocation

When a company runs many campaigns (e.g., IBM’s various business units), the CMO must allocate a limited total budget. A useful visual approach plots campaigns on two axes:

• X‑axis: Efficiency – leads per $1000 spent.
• Y‑axis: Contribution – total leads generated.
• Bubble size: Total spend on that campaign.

Interpretation:

Scaling trade‑off: Adding budget to an efficient campaign often forces it into less relevant keywords or audiences, reducing efficiency. The path is rarely a straight line to the top‑right; it zig‑zags.

Channel‑level drill‑down – within a single campaign, the same bubble chart can be applied to channels (search, social, display, video). Under‑performing channels (e.g., a social channel with low efficiency) can be cut, while high‑efficiency channels receive more budget. This can be repeated for sub‑channels (LinkedIn vs. Instagram), audience segments, and even individual keywords or creatives.

Exam tip: The “zig‑zag” effect – increasing budget often dilutes performance. This is critical for any question about scaling efficient campaigns.

---

Strategic Budget‑Allocation Models

For high‑level budget splits (once a year or once in 6 months), three classes of models are used:

Key difference: Attribution assumes only media touchpoints determine conversion; MMM incorporates external variables (e.g., tax season for insurance). Incrementality uses a control‑vs‑treatment design to isolate the causal effect of advertising.

Incrementality in Detail

• Digital twins: Platforms identify a user (e.g., “Shainesh”) and a similar user with the same profile. One is shown the ad, the other is not (dark). Both are later surveyed or measured for conversion outcomes. The lift = difference between exposed and control groups.
• Market‑level approach: Identify two similar markets; spend in one, keep the other dark. The lift in the active market (net of baseline brand equity) is the incremental effect.
• Especially valuable for brand campaigns (where performance metrics are hard to link to sales).

---

Conversion‑Rate Optimisation (CRO) & Customer‑Journey Alignment

Even the best‑targeted media fails if the landing page experience is poor. CRO tools go beyond standard analytics:

• Scroll maps (e.g., Hotjar) – show how far users scroll; most users never scroll, so all key content must be above the fold.
• Click maps – show where users click; if a product has poor conversion but is at the bottom of the page, the solution is to move it up, not kill the product.
• Video recordings – reveal real behaviour (e.g., a glitching video causing drop‑off).

The IBM data‑center example:

• Ad promised “minimise downtime”.
• Landing page talked about “multi‑vendor support services” – no mention of downtime.
• Conversion was poor because of a dissonance between ad and landing page.
• Organisational silos (agency vs. web‑team) prevented a quick fix; global approval cycles meant the page couldn’t be updated before the budget was exhausted.

Living the customer journey:

• Before launch, a “campaign tear‑down team” (separate from the creators) maps each step of the customer journey and checks consistency (the “thread test”: where the thread breaks, the campaign fails).
• This should be done proactively, not reactively after launch.

A/B testing then formalises hypothesis‑driven optimisation – form a hypothesis from data (e.g., “moving the CTA above the fold will increase conversion”), test, and iterate.

---

Key takeaways

• Optimise on true KPIs (conversion, revenue) not vanity metrics (CTR, impressions).
• Full‑funnel data requires integrating media and website data via UTM tagging.
• Multi‑campaign allocation uses efficiency (leads/$) and contribution (total leads) – scaling often reduces efficiency.
• Three strategic models – attribution (tactical, media‑only), media mix (strategic, includes external factors), incrementality (causal via hold‑out tests).
• CRO tools (scroll maps, recordings) reveal page‑level friction; align ad promise with landing page content to avoid dissonance.
• Organisational details (silos, access, tagging) are as critical as the analytics techniques.