Web, Mobile and Product Analytics

Introduction

Digital products generate a continuous record of interactions: a visitor arrives from a search engine, views pages, searches for information, adds an item to a cart, installs an application, completes onboarding, or abandons a task. Web, mobile, and product analytics convert these interaction records into evidence for improving acquisition, usability, activation, retention, and business performance.

The central analytical challenge is not simply counting clicks. It is connecting who did something, what they did, when they did it, where they came from, and whether the interaction contributed to a meaningful outcome. A high-traffic website may perform poorly if visitors cannot find relevant information. An application may have many downloads but low activation. A product may record thousands of events while failing to answer important managerial questions because its instrumentation is incomplete or inconsistent.

This chapter develops a practical framework for analysing traffic sources, users, sessions, events, paths, funnels, search behaviour, usability, activation, and feature adoption. It also examines data quality, privacy, attribution, experimentation, and the responsible use of AI.

Learning Objectives

By the end of this chapter, students should be able to:

1. Distinguish users, sessions, events, page views, screens, and product accounts.
2. Analyse traffic sources and assess the limits of attribution.
3. Construct and interpret funnels, paths, search reports, and retention measures.
4. Diagnose abandonment using quantitative and qualitative evidence.
5. Design a reliable event-tracking and measurement plan.
6. Evaluate platform reports rather than accepting their default definitions uncritically.
7. Recommend improvements to activation, usability, and feature adoption.
8. Discuss privacy, governance, accessibility, and ethical issues in digital analytics.

1. The Digital Analytics Measurement Model

A useful measurement model links four layers:

The same interaction can be described at several levels. A user is an identifiable person or device profile under a platform’s rules. A session is a bounded period of activity, although the boundary differs by platform. An event is a recorded action, such as product_view, search_submitted, or checkout_completed. A page view records a page load or virtual page view; a mobile application commonly records a screen view instead. An account, organisation, or household may contain several users, which is important in business-to-business products.

These objects must not be treated as interchangeable. Ten sessions do not necessarily represent ten people, and one person may use multiple browsers or devices. Event counts are especially easy to misinterpret: one user can generate hundreds of events through repeated scrolling, automatic video events, or background activity.

A measurement plan should begin with decisions rather than available reports. For each business question, specify:

• the decision to be supported;
• the population and time period;
• the metric and its denominator;
• the dimensions needed for comparison;
• the required events and properties;
• data-quality tests and ownership;
• privacy and retention constraints.

For example, “Is onboarding effective?” is not answered by an onboarding-page view. A stronger design defines activation, such as completing a core task within seven days, and identifies the events required to observe that outcome.

2. Traffic Sources and Attribution

Traffic-source analysis explains how users reach a digital property. Common source categories include:

• Organic search: unpaid search-engine results.
• Paid search: sponsored search placements.
• Display and social advertising: paid placements on websites or social platforms.
• Referral: links from another domain or application.
• Email and other owned campaigns: tagged links in controlled communications.
• Direct or unknown: no reliable referrer or campaign information.

A typical campaign link includes source, medium, campaign, and sometimes content or term parameters. Naming conventions should be standardised; otherwise, Email, email, and newsletter may appear as separate categories.

Attribution assigns credit for an outcome to one or more touchpoints. A last-touch model credits the final recorded source before conversion. A first-touch model credits the source that initiated the relationship. A linear model distributes credit equally across recorded touchpoints. These are conventions, not observations of causal influence.

The conversion rate for a defined population is:

The denominator must be stated. Session conversion rate and user conversion rate answer different questions. If 2,000 eligible sessions produce 80 purchases, session conversion rate is 4%. If those sessions came from 1,500 users, the corresponding user-based rate is 80 ÷ 1,500 = 5.33%, assuming each user is counted once in the numerator and denominator.

Source comparisons should control for context. Paid traffic may be intentionally targeted at awareness and therefore have a lower immediate conversion rate than branded search. A source with fewer conversions may produce higher-value customers. Analysts should therefore consider revenue, margin, lead quality, downstream retention, and acquisition cost rather than optimising a single short-term rate.

Attribution is weakened by privacy restrictions, cookie deletion, ad blockers, cross-device behaviour, dark social sharing, consent choices, and tracking loss. “Direct” traffic often includes untagged campaigns, copied links, mobile-app referrals, or missing referrer data. It should not automatically be interpreted as deliberate typing of a URL.

3. Users, Sessions, Events, and Identity

A session helps analyse sequences of activity, but its definition is operational. A platform may end a session after inactivity, at a fixed time, at midnight, or when campaign information changes. Mobile applications may pause and resume in ways that do not resemble browser sessions. Consequently, session counts from two platforms may differ even when both are functioning correctly.

Identity resolution joins activity across visits or devices. Common identifiers include anonymous device IDs, authenticated user IDs, account IDs, and consented advertising identifiers. A robust design distinguishes:

• anonymous identity before login;
• authenticated identity after login;
• account or organisation identity;
• device and application instance identity.

Identity stitching should be deterministic where possible and should not create false certainty. Shared devices, recycled identifiers, multiple users of one account, and delayed login events can cause incorrect joins. Analysts should document whether metrics are person-based, device-based, account-based, or event-based.

Event data normally contains an event name, timestamp, actor or identifier, platform, and optional properties. For example:

Properties should describe the event at the time it occurred. Sensitive free text should not be collected unless there is a documented and lawful purpose. A good event taxonomy uses stable names, clear ownership, versioning, and a data dictionary.

4. Paths, Funnels, and Abandonment

A path analysis examines sequences of actions. It can reveal common journeys, loops, dead ends, and unexpected entry points. However, path reports become unreadable when every URL, parameter, or minor action is treated as a separate node. Analysts should group equivalent pages, define meaningful events, and filter bot or internal traffic.

A funnel represents an ordered set of required steps. For example:

1. product page viewed;
2. item added to cart;
3. checkout started;
4. payment submitted;
5. purchase completed.

For step i, the step conversion rate is:

The cumulative funnel conversion rate from the first to the final step is:

These formulas assume that the population, time window, identity rule, and event definitions are consistent. If users can repeat a step, the analyst must specify whether the funnel counts users, sessions, or occurrences. A strict funnel requires the order and time conditions to be defined; an open funnel may count completion even when users enter at different steps.

Worked Example

An online retailer observes the following user counts in a weekly funnel:

Product-view to cart conversion is 2,500 ÷ 10,000 = 25%. Cart to checkout conversion is 1,500 ÷ 2,500 = 60%. Checkout to purchase conversion is 900 ÷ 1,500 = 60%. Overall conversion is 900 ÷ 10,000 = 9%.

The largest relative loss occurs before cart addition: 75% of product viewers do not add an item. This does not prove that the product page is defective. Users may be browsing, comparing prices, or arriving through poorly targeted advertising. Diagnosis requires segmentation and supporting evidence, such as device type, page speed, stock status, recordings obtained with consent, customer feedback, and usability testing.

Abandonment analysis should ask:

• Is the drop-off concentrated in a browser, operating system, country, or acquisition source?
• Is an error, validation failure, or slow response recorded?
• Did the user return later and complete the task?
• Is the event missing because instrumentation failed?
• Does qualitative research explain the observed behaviour?

A funnel is descriptive. It does not by itself establish why abandonment occurred.

5. Search Behaviour and Information-Seeking

Internal search is both a behavioural signal and a usability diagnostic. Important measures include search usage rate, query volume, zero-result rate, reformulation rate, result-click rate, and downstream task completion. A useful search-result click-through rate is:

The unit is a search, not necessarily a user. Query normalisation should account for case, spelling, punctuation, synonyms, and personally identifiable information. Zero-result searches may indicate missing content, poor indexing, vocabulary mismatch, or an overly restrictive filter. A high click rate is not always positive if users must click repeatedly before finding the answer.

Search quality should be assessed against outcomes. For a support site, successful search may mean resolving an issue without contacting support. For an online store, it may mean product discovery and purchase. Ranking changes should be evaluated with experiments or carefully designed quasi-experimental comparisons, while protecting users from discriminatory or unsafe results.

6. Usability, Activation, and Feature Adoption

Usability concerns whether intended users can achieve goals effectively, efficiently, and satisfactorily in a specified context. Digital analytics supplies behavioural evidence, but usability cannot be inferred from clicks alone. A low number of clicks may mean a simple interface—or that users cannot find the control.

Useful measures include task completion rate, time on task, error rate, rage-click indicators, accessibility failures, and support contacts. Time-on-task is particularly difficult to interpret because idle tabs, interruptions, and background activity distort it.

Activation is the point at which a new user experiences the product’s core value. It should be defined through behaviour, not merely registration or installation. A collaboration tool might define activation as creating a workspace, inviting a colleague, and completing a shared task within seven days.

An activation rate can be expressed as:

The eligibility rule and observation window are essential. A seven-day activation rate cannot be fairly compared with a one-day rate. Cohort analysis is preferable to mixing users acquired at different times or under different product versions.

Feature adoption measures meaningful use of a feature by an appropriate population. A basic adoption rate is:

“Meaningful use” might require successful completion rather than a mere click. Analysts should also examine repeat use, retention, feature discoverability, and whether the feature improves the user’s outcome. High adoption of a compulsory feature does not necessarily indicate preference or value.

7. Instrumentation and Platform Reports

Instrumentation is the technical and organisational process of collecting behavioural data. A reliable implementation includes:

1. a measurement specification;
2. event names and property definitions;
3. ownership and implementation tickets;
4. test cases for web, iOS, Android, and server-side flows;
5. validation in development and production;
6. monitoring for volume, duplication, latency, and missing fields;
7. version control and change documentation.

Client-side tracking is convenient but vulnerable to browsers, blockers, network failures, and JavaScript errors. Server-side tracking may better capture transactions but can lose interface context and may create duplicates when retries occur. Hybrid designs require event IDs and deduplication rules.

A platform report is not a neutral window onto reality. Its results depend on identity settings, sessionisation, attribution, filters, sampling, consent, time zone, bot exclusion, data retention, and processing delays. Before interpreting a report, ask:

• What is the unit of analysis?
• What population is included or excluded?
• Are counts users, sessions, events, or accounts?
• Are data sampled or modelled?
• How are late, deleted, or consent-restricted events treated?
• Can the metric be reconciled with a trusted source?

Data-quality checks include comparing completed orders with the order system, checking that revenue is not duplicated, testing event sequences, monitoring sudden changes, and validating that releases did not alter definitions. A sudden 40% fall in conversions may reflect a broken payment flow—or a missing purchase_completed event.

8. Analysis and Optimisation Workflow

A disciplined optimisation cycle is:

1. Frame the decision. Define the user problem and business outcome.
2. Specify the metric. State numerator, denominator, population, window, and unit.
3. Validate the data. Check implementation, identity, consent, and exclusions.
4. Locate the problem. Segment by device, source, geography, cohort, product, and experience.
5. Generate explanations. Combine quantitative patterns with research, surveys, logs, and usability testing.
6. Prioritise interventions. Consider expected impact, confidence, effort, risk, and accessibility.
7. Test the change. Use controlled experiments where feasible; define primary and guardrail metrics in advance.
8. Monitor and learn. Examine short- and long-term effects, unintended consequences, and durable adoption.

A/B testing can estimate the effect of a change when assignment is sufficiently random, exposure is correctly recorded, interference is limited, and the analysis follows the pre-specified outcome. A higher click-through rate is not sufficient if purchases, customer satisfaction, performance, or cancellation worsen. Guardrail metrics are therefore important.

9. Applied Case: Illustrative Mobile Banking Onboarding

Background

A fictional mobile bank has strong advertising performance and 100,000 monthly app installations. Only 31% of new users complete identity verification, and fewer than 18% make a first deposit within seven days.

Problem

The product team initially blames the verification form. However, the analytics team discovers that the installation event is recorded on the device while onboarding completion is recorded against an account, making user-level rates difficult to interpret.

Analysis

The team first creates a reliable identity bridge after account creation and audits events across Android and iOS. It finds three issues:

• camera permission denial is not distinguished from document-recognition failure;
• a slow upload screen lacks a visible progress indicator;
• users who leave the flow receive no clear return path.

A funnel segmented by operating system shows substantially higher abandonment on older Android devices. Qualitative testing confirms that users interpret the upload screen as frozen. Search within help content also shows repeated queries about acceptable identity documents.

Approach and Solution

The bank introduces explicit permission explanations, a progress indicator, clearer document guidance, resumable uploads, and error-specific messages. It defines activation as “verification completed and first deposit made within seven days.” The primary experiment metric is activation; guardrails include fraud-review rate, support contacts, crash rate, and verification time.

Outcome

In this illustrative scenario, the redesigned flow increases verification completion and first-deposit activation without reducing fraud controls. The improvement is attributed cautiously: the experiment compares randomly assigned versions, tracks exposure correctly, and confirms that event collection remained stable.

Learning Points

The case demonstrates that funnel loss may arise from instrumentation defects, technical performance, unclear communication, or user motivation. It also shows why installation counts are not equivalent to active users and why optimisation should protect security, accessibility, and long-term outcomes.

10. Ethics, Privacy, and Governance

Digital analytics can expose sensitive behaviour, including health interests, financial activity, location, identity, and inferred characteristics. Responsible practice requires purpose limitation, data minimisation, transparency, lawful processing, access controls, retention limits, and mechanisms for correction or deletion where applicable. Consent should be meaningful rather than hidden in confusing interfaces.

Teams should avoid collecting passwords, payment secrets, unnecessary free text, or precise location when a less intrusive signal is sufficient. Session recordings and heat maps can capture personal information accidentally; masking and strict access controls are essential. Re-identification risk increases when multiple datasets are joined.

Ethical analysis also includes fairness and accessibility. Personalisation can improve relevance but may exclude groups, produce unequal prices, or reinforce biased historical behaviour. Automated recommendations and search ranking should be monitored for disparate impacts, unsafe content, and feedback loops. Analysts must distinguish correlation from causation and avoid using behavioural data to manipulate vulnerable users through deceptive design.

11. Contemporary Developments and Best Practices

Modern digital analytics increasingly combines event streams, product data warehouses, experimentation platforms, customer feedback, and machine-learning models. AI can classify search queries, summarise feedback, detect anomalies, and predict churn. These uses require human review, representative training data, explainable decisions where stakes are high, and controls against personal-data leakage.

Privacy-preserving measurement is becoming more important as third-party cookies decline and platform identifiers become less available. First-party data, consent-aware server-side collection, aggregated reporting, and clean-room approaches can support measurement while reducing unnecessary tracking. No technical approach eliminates the need for lawful purpose, transparency, and governance.

Best practices include:

• define outcomes before selecting tools;
• maintain a governed event taxonomy;
• use stable identifiers and document identity rules;
• report denominators and time windows;
• segment rather than relying only on averages;
• reconcile analytics with operational systems;
• combine behavioural, qualitative, and performance evidence;
• predefine experiment metrics and guardrails;
• design for accessibility and privacy from the start;
• preserve a clear audit trail for changes and assumptions.

Chapter Summary

Web, mobile, and product analytics connect digital interactions with decisions about acquisition, usability, activation, retention, and feature value. Users, sessions, events, accounts, and devices are different analytical units and must not be casually substituted. Traffic-source and attribution reports are useful but depend on tagging, identity, consent, and model assumptions.

Funnels quantify ordered progression and abandonment; paths reveal sequences and loops; internal search exposes information needs; and cohort analysis supports fair comparison over time. Conversion, activation, and adoption rates are meaningful only when their numerator, denominator, population, and observation window are explicit. Instrumentation quality is a prerequisite for interpretation, while platform defaults should be examined critically.

Effective optimisation combines data validation, segmentation, qualitative research, experimentation, and guardrail monitoring. Ethical practice requires privacy, security, accessibility, fairness, transparency, and restraint in the collection and use of behavioural data.

Key Terms

User: An identifiable person or device profile under a platform’s identity rules.

Session: A bounded period of activity defined by a platform’s sessionisation rules.

Event: A recorded user or system interaction with a timestamp and properties.

Page view: A recorded web-page load or virtual page display.

Screen view: A recorded view of a mobile-application screen.

Traffic source: The origin or channel through which a visitor arrives.

Attribution: Assignment of credit for an outcome to one or more touchpoints.

Conversion rate: The proportion of an eligible population completing a defined outcome.

Funnel: An ordered sequence of steps leading to an outcome.

Abandonment: Failure to continue or complete a defined journey.

Path analysis: Examination of sequences and transitions among interactions.

Internal search: Search conducted within a website or application.

Usability: The effectiveness, efficiency, and satisfaction of use in context.

Activation: The point at which a new user experiences core product value.

Feature adoption: Meaningful use of a feature by an eligible population.

Cohort: A group sharing a defined characteristic or starting period.

Instrumentation: Technical implementation used to collect behavioural data.

Guardrail metric: A measure used to detect harmful or unintended effects of an intervention.

Identity resolution: Joining activity across visits, devices, or accounts under defined rules.

Review Questions

Short Answer Questions

1. Why are users, sessions, and events different units of analysis?
2. What is the difference between first-touch and last-touch attribution?
3. Define activation and explain why installation is usually an insufficient activation measure.
4. What information should be included in an event specification?
5. Why should a funnel report state its denominator and identity rule?

Long Answer Questions

1. Explain how an analyst should evaluate traffic sources while accounting for attribution limitations and downstream customer value.
2. Describe a complete workflow for diagnosing abandonment in a digital product.
3. Compare path analysis and funnel analysis, including the strengths and limitations of each.
4. Explain how internal search data can identify usability and content problems.
5. Discuss the ethical, privacy, and governance responsibilities involved in collecting behavioural analytics.

Analytical / Application Questions

1. A funnel has 12,000 product viewers, 3,000 cart additions, 1,800 checkouts, and 1,080 purchases. Calculate each step conversion rate and the cumulative conversion rate. Identify the stage requiring further investigation, and state why the data alone cannot establish causation.
2. A mobile application reports 50,000 installations but only 14,000 users complete the defined activation event within seven days. Design an analysis plan that checks identity resolution, platform differences, event quality, usability, and acquisition-source effects.
3. A feature has a 35% click rate but only 8% successful task completion and a rising support-contact rate. Recommend metrics, qualitative research, and an experiment design for deciding whether to redesign, simplify, or remove the feature.

Multiple Choice Questions with Answers

Question 1

Which metric uses searches as its unit of analysis?

A. Search result click-through rate

B. User retention rate

C. Account activation rate

D. Revenue per customer

Correct Answer: A. Search result click-through rate

Question 2

A funnel begins with 10,000 users and ends with 900 purchases. What is the cumulative conversion rate?

A. 0.9%

B. 9%

C. 10.9%

D. 90%

Correct Answer: B. 9%

Question 3

Which statement about attribution is correct?

A. Last-touch attribution proves the final channel caused the conversion.

B. First-touch attribution assigns credit to every touchpoint equally.

C. Attribution models are conventions for assigning credit, not direct proof of causation.

D. Direct traffic always represents users who typed the URL manually.

Correct Answer: C. Attribution models are conventions for assigning credit, not direct proof of causation.

Question 4

Which event is most appropriate for measuring successful product adoption?

A. Feature button displayed

B. Feature button clicked once

C. Feature task completed successfully

D. Application installed

Correct Answer: C. Feature task completed successfully

Question 5

What is a major risk of poorly governed event instrumentation?

A. It guarantees unbiased attribution.

B. It can make a tracking failure appear to be a behavioural change.

C. It eliminates the need for qualitative research.

D. It ensures that users are counted consistently across platforms.

Correct Answer: B. It can make a tracking failure appear to be a behavioural change.

Question 6

Which definition best represents activation?

A. The first advertisement impression

B. The moment a user downloads an application

C. A behaviour indicating that a new user has experienced core product value

D. Any event generated after account deletion

Correct Answer: C. A behaviour indicating that a new user has experienced core product value

Question 7

Why should funnel abandonment be segmented by device or operating system?

A. To replace all user-level analysis with device counts

B. To identify whether the problem is concentrated in a particular experience or platform

C. To ensure every segment has the same conversion rate

D. To avoid checking instrumentation

Correct Answer: B. To identify whether the problem is concentrated in a particular experience or platform

Question 8

Which practice best protects privacy in analytics?

A. Collect all available free-text fields indefinitely

B. Store passwords so journeys can be reconstructed

C. Apply data minimisation, purpose limitation, access controls, and retention limits

D. Publish user-level event logs for transparency

Correct Answer: C. Apply data minimisation, purpose limitation, access controls, and retention limits

Question 9

A feature is meaningfully used by 600 of 2,000 eligible users. What is the adoption rate?

A. 3%

B. 12%

C. 30%

D. 33.3%

Correct Answer: C. 30%

Question 10

What is the primary purpose of a guardrail metric in an experiment?

A. To replace the primary outcome

B. To detect harmful or unintended effects of a change

C. To increase the number of tracked events regardless of relevance

D. To guarantee statistical significance

Correct Answer: B. To detect harmful or unintended effects of a change