Brand Data and Digital Ecosystems

Introduction

Brand management increasingly depends on the ability to understand, connect and govern data across a customer’s journey. A customer may discover a brand through a search engine, interact with a social platform, visit a website, use a mobile application, contact customer service, purchase through a retailer and later receive an email or loyalty offer. Each interaction can produce data, but the data is usually distributed across different organisations, technologies and legal jurisdictions.

Brand data management is therefore not simply a matter of collecting more information. It involves identifying which data the brand can legitimately use, understanding its origin and quality, linking records that refer to the same person or household, and activating insights in ways that create customer value without violating trust. The relevant environment is a digital ecosystem: a network of customers, brands, platforms, intermediaries, devices, data providers, software systems and regulatory institutions.

This chapter examines first-, second- and third-party data; customer journeys; customer relationship management systems; customer data platforms; data quality; identity resolution; application programming interfaces; cloud systems; cookies; and data integration. It also considers artificial intelligence, privacy, governance and the strategic implications for brand managers.

Learning Objectives

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

1. Distinguish first-, second- and third-party data and evaluate their branding uses.
2. Map data creation and activation across customer journeys.
3. Explain the roles of CRM systems, customer data platforms and adjacent technologies.
4. Assess data quality, identity resolution and integration choices.
5. Explain the changing role of cookies, consent and privacy regulation.
6. Evaluate APIs, cloud architecture and AI-enabled activation in brand management.
7. Recommend governance practices that balance personalisation, commercial value and customer rights.

1. The Digital Ecosystem as a Brand Infrastructure

A traditional brand model often treated communication as a sequence: the firm sent a message and the customer received it. Digital ecosystems are more interactive. Customers generate behavioural, transactional and preference data while interacting with brand-owned and third-party services. Platforms also influence what can be observed, how it can be measured and which audiences can be reached.

A simplified ecosystem contains five layers:

• People and entities: customers, households, prospects, employees, retailers and business partners.
• Touchpoints: websites, applications, stores, call centres, email, social media, search, connected products and advertising platforms.
• Data assets: identity, transactions, events, preferences, content interactions, consent records and service histories.
• Technology infrastructure: CRM systems, data warehouses, data lakes, customer data platforms, APIs, identity services and analytics tools.
• Governance: privacy law, contracts, security controls, ethical standards, retention rules and accountability processes.

A brand’s competitive advantage does not arise automatically from possessing data. It arises when the organisation can convert reliable data into useful decisions and experiences. This conversion may be expressed as:

The arrows are not guaranteed. Poor data quality can prevent information from becoming insight; weak governance can prevent a technically possible activation; and an inappropriate decision can damage trust even when the data is accurate.

Figure: A simplified brand data ecosystem in which customer interactions generate data, data foundations support decisions, and activation produces further feedback.

2. Data Ownership and Provenance

First-party data

First-party data is collected directly by an organisation from its own interactions with customers or prospects. Examples include purchase records, account details, loyalty activity, customer-service conversations, declared preferences, website events and consent choices. It is generally valuable because the brand knows the context in which it was collected and can design the collection relationship.

First-party data is not automatically accurate, freely usable or ethically unproblematic. A customer may provide an outdated address, misunderstand a preference centre or consent to one purpose but not another. The brand must record purpose, collection date, source, permissions and retention requirements.

Second-party data

Second-party data is another organisation’s first-party data shared with a brand, usually through a direct partnership or contractual arrangement. For example, an airline and a hotel group may cooperate on a joint offer. A retailer may share aggregated audience information with a manufacturer. The recipient should assess whether the provider collected the data lawfully, whether the proposed use matches customer expectations, and whether the contract defines security, permitted purposes and deletion.

Second-party data can improve reach or context, but it creates dependency and governance risk. A brand may not be able to inspect the original collection process fully. Data-sharing arrangements therefore require due diligence rather than a simple assumption that a partner’s data is safe.

Third-party data

Third-party data is obtained from an entity that did not collect it directly from the brand’s customer relationship. It may be supplied by data aggregators, advertising intermediaries, research companies or public sources. Historically, third-party data supported audience targeting and enrichment. Its usefulness has been constrained by questions about provenance, accuracy, consent, opacity and the decline of unrestricted third-party tracking.

The categories describe the brand’s relationship to collection, not necessarily the data’s moral or legal status. Publicly accessible information can still contain personal data. Conversely, first-party data can still be unlawful if collected without a valid basis or used beyond customer expectations.

Zero-party data

A related concept is zero-party data, information a customer intentionally and proactively provides, such as a stated product preference, size, communication choice or future need. It can be highly useful because it expresses declared intent, but incentives and question design influence the response. It should not be treated as perfectly truthful or permanent.

3. Mapping Data Across the Customer Journey

A customer journey is not always linear, but a staged model helps identify data opportunities and risks.

Journey mapping should answer five questions:

1. What is the customer trying to accomplish?
2. Which touchpoint captures the event?
3. What data is collected, and is it necessary?
4. Where is the data stored and who may use it?
5. What outcome indicates value or harm?

For example, a customer who abandons an online basket might be sent a reminder. However, the reminder should depend on consent, frequency limits, product sensitivity, stock availability and whether the customer has already purchased through another channel. Journey data must be interpreted as a sequence, not as isolated clicks.

4. CRM, CDP and the Data Technology Stack

Customer relationship management systems

A customer relationship management (CRM) system stores and manages information needed to conduct customer-facing processes. It commonly supports sales activities, service cases, account management, campaign execution and interaction histories. CRM data is often structured around identifiable accounts, contacts, opportunities, cases or transactions.

CRM systems are strong at operational coordination. A service agent can view a customer’s previous cases; a sales team can manage a pipeline; and a marketing team can define an audience using known records. Their limitations may include fragmented identifiers, limited event-scale processing and dependence on manual entry or connected systems.

Customer data platforms

A customer data platform (CDP) is a technology category designed to ingest data from multiple sources, unify customer profiles, manage audiences and activate those audiences in connected channels. CDP capabilities differ by vendor. Some are especially strong in real-time events; others emphasise marketing activation, analytics or data governance.

A CDP is not a magical “single source of truth.” It is a layer whose value depends on source systems, identity rules, consent signals, data models and operating discipline. A CDP should not be purchased merely to solve a problem that is actually caused by unclear ownership or poor data definitions.

Adjacent systems

• Data warehouse: a structured analytical repository, usually designed for governed reporting and analysis.
• Data lake: a repository that can store large volumes of raw or semi-structured data; governance and discoverability are essential.
• Data lakehouse: an architecture seeking to combine flexible storage with warehouse-like management and analytical reliability.
• Customer data infrastructure: the broader set of pipelines, storage, identity, consent, analytics and activation services.
• Marketing technology stack: tools used for content, advertising, email, experimentation, measurement and automation.

A practical architecture often separates systems by purpose while connecting them through controlled interfaces. Operational CRM records should not necessarily be replaced by a CDP, and an analytical warehouse should not necessarily become the system that sends every customer message.

5. Data Quality and the Principle of Fitness for Purpose

Data quality is multidimensional. Common dimensions include:

• Accuracy: whether a value represents reality.
• Completeness: whether required values are present.
• Consistency: whether the same concept has compatible values across systems.
• Timeliness: whether the data is current enough for its use.
• Validity: whether it conforms to permitted formats and meanings.
• Uniqueness: whether duplicate records are controlled.
• Accessibility: whether authorised users can obtain it when needed.
• Traceability: whether source, transformations and changes can be identified.

Quality is relative to purpose. An approximate city may be sufficient for regional planning but not for delivery. A purchase event may be timely enough for a weekly report but not for fraud prevention. Consequently, data quality should be measured against business and customer consequences rather than an abstract ideal.

A simple completeness rate for a field can be calculated as:

The formula assumes that records are correctly counted and that “missing” is defined consistently. It measures presence, not accuracy. If 950 of 1,000 records contain an email address, completeness is 95%, but the formula says nothing about whether those addresses are valid or belong to the correct people.

A data-quality dashboard may therefore combine rule-based checks, exception queues, sample audits and downstream outcome measures. Brand managers should ask whether poor quality causes failed deliveries, irrelevant offers, duplicated communications, exclusion or unfair treatment.

6. Identity Resolution and the Customer View

Identity resolution is the process of determining whether records from different sources refer to the same entity. An identity may be a person, household, organisation, device or account. The task is difficult because identifiers change, are shared, are missing or are deliberately protected.

Deterministic and probabilistic approaches

Deterministic matching uses an agreed identifier or rule, such as a verified account ID or a combination of exact fields. It is relatively explainable but can miss matches when data is incomplete.

Probabilistic matching estimates the likelihood that two records belong to the same entity using several signals, such as name similarity, address similarity, device relationship and behavioural patterns. It can increase coverage but creates false positives and false negatives. It requires thresholds, validation samples, review procedures and a way to represent uncertainty.

A false positive incorrectly merges two people; a false negative fails to merge records belonging to the same person. In branding, a false positive can be especially damaging: one person may receive another person’s service history or a wrongly personalised message. Identity resolution should therefore be conservative in sensitive contexts and transparent about confidence.

Identity graphs may connect identifiers over time, but a graph is not proof of identity. A shared device, household address or browser can be used by several people. The design must distinguish a person profile from a household, account, device and anonymous browser relationship.

Practical controls

• Establish a canonical identifier and document its meaning.
• Maintain source-system identifiers for traceability.
• Record match method, confidence and match date.
• Separate verified facts from inferred attributes.
• Provide correction and deletion processes.
• Test outcomes for demographic or geographic bias.
• Avoid using weak matches for high-impact decisions.

7. APIs, Integration and Cloud Systems

An application programming interface (API) is a defined way for software systems to request or exchange data and functions. APIs can support real-time lookup, event submission, audience synchronisation and service orchestration. For example, a website may send a purchase event to a customer platform, which then updates an authorised profile and triggers a fulfilment workflow.

Common integration patterns include:

• Batch integration: data is transferred on a schedule. It is efficient for routine reporting but may be stale.
• Streaming or event integration: events are transmitted as they occur. It supports timely experiences but requires resilience and monitoring.
• Request-response API: one system requests information and another responds. Latency, authentication and availability matter.
• Webhooks: a system sends a notification when an event occurs.
• Extract, transform, load (ETL): data is extracted, transformed before loading and then stored.
• Extract, load, transform (ELT): raw data is loaded first and transformed within the analytical environment.

Cloud computing provides scalable storage and processing through network-accessible services. Its benefits include elasticity, managed infrastructure and access to specialised analytics. Risks include vendor concentration, misconfiguration, cross-border transfer issues, uncontrolled costs, service outages and unclear responsibility. Cloud adoption does not remove the brand’s accountability for data governance.

Sound integration design defines data contracts: field meanings, formats, permitted values, ownership, update frequency, error handling, authentication and versioning. Systems should be able to reject invalid events, retry temporary failures and preserve logs without storing unnecessary personal information.

8. Cookies, Identifiers and the Changing Addressability Environment

A cookie is a small data object stored by a browser and associated with a website or service. A first-party cookie is set in the context of the site the user is visiting; a third-party cookie is associated with a different domain and has historically supported cross-site advertising and measurement. Browser restrictions, platform policies and privacy regulation have reduced the reliability of third-party cookies.

Other identifiers include mobile advertising IDs, login IDs, hashed contact information, device signals and contextual information. Hashing is a transformation, not automatic anonymisation: if a value can be linked or matched, it may remain personal data under applicable law.

The strategic consequences are significant:

• Brands should strengthen consent-based first-party relationships.
• Measurement should not depend on one identifier or one platform.
• Contextual advertising can use the content environment rather than a person’s cross-site history.
• Clean rooms may allow controlled collaboration without exposing raw customer-level data, although they require careful governance and do not eliminate legal duties.
• Incrementality testing and aggregated measurement can reduce dependence on individual-level tracking.

The disappearance or restriction of a tracking method does not make measurement impossible. It changes the evidence available and increases the importance of experimentation, modelling, declared information and privacy-preserving analytics.

9. Consent, Privacy and Ethical Governance

Privacy governance is a brand issue because customers judge the organisation, not merely its software supplier. Applicable requirements vary by jurisdiction, but responsible practice commonly includes purpose limitation, data minimisation, transparency, security, retention control, access and correction rights, deletion where applicable, and lawful handling of sensitive information.

Consent should be specific, informed, freely given where required, and as easy to withdraw as to provide. A preselected box or confusing interface may produce a technical signal without producing meaningful customer choice. Brands should maintain a consent record linked to purposes, channels, time, version of notice and withdrawal status.

Ethical concerns extend beyond formal compliance:

• Surveillance: extensive observation may create a chilling effect.
• Manipulation: personalisation can exploit vulnerability or urgency.
• Discrimination: models may reproduce unequal access or biased targeting.
• Opacity: customers may not understand why a message or decision occurred.
• Data asymmetry: firms may know much more about customers than customers know about the firm.
• Function creep: data collected for service may later be used for unrelated advertising.

A data protection impact assessment or equivalent risk review is appropriate when processing is large-scale, sensitive, systematic or likely to create significant harm. Governance should assign accountable owners, not merely circulate policies.

10. AI in Brand Data Ecosystems

AI can support classification, prediction, recommendation, anomaly detection, natural-language analysis and next-best-action systems. For example, a model may classify service conversations by topic, predict likely churn or recommend content based on permitted behavioural signals.

AI does not remove the need for sound data foundations. It can amplify missing data, historical bias, incorrect identity links and proxy discrimination. A model trained on past campaign responses may learn to prioritise audiences who were previously over-targeted, not audiences with objectively greater need.

Responsible use involves:

1. Defining the decision and its customer impact.
2. Confirming lawful purpose, data relevance and consent conditions.
3. Establishing data lineage and documentation.
4. Testing performance across relevant groups.
5. Monitoring drift as behaviour and platforms change.
6. Providing human review for consequential or ambiguous decisions.
7. Explaining outcomes in language appropriate to the audience.
8. Retaining only the data and model outputs necessary for the purpose.

Generative AI can create brand content and summarise feedback, but confidential customer data should not be entered into tools without approved contractual, security and retention controls. Human editorial review remains important for accuracy, tone, intellectual property and harmful stereotypes.

11. An Applied Case: Northstar Outdoor Equipment

The following is a fictional illustrative case.

Background

Northstar Outdoor Equipment sells through its own website, branded stores and independent retailers. It operates a loyalty programme and a customer-service centre. Its marketing team uses an email platform, its sales team uses a CRM, and its analysts use a cloud warehouse. These systems were implemented at different times.

Problem

Customers received duplicate emails after purchasing in stores and online. Service agents sometimes saw incomplete purchase histories. Marketing reported a large loyalty database, but many profiles had outdated email addresses or multiple accounts. The company considered buying a third-party audience file to improve prospecting.

Analysis

A journey and data audit identified four problems:

• Store transactions used a loyalty number, while online purchases used email address as the main key.
• The CRM and email platform used different consent fields.
• The warehouse received nightly files, so suppression after a purchase could be delayed.
• The proposed audience file had unclear provenance and could not be reliably connected to Northstar’s consent records.

A quality review of 20,000 loyalty records found 18,000 with a non-missing email address. The completeness rate was therefore:

This result did not establish that 90% of addresses were accurate, deliverable or permissioned.

Approach and solution

Northstar adopted a staged programme:

1. It defined a persistent customer ID and preserved source IDs from stores, web and service.
2. It created an identity-resolution policy using verified account IDs first and conservative secondary rules only for low-risk uses.
3. It established one consent vocabulary with separate purposes for service, loyalty and marketing.
4. It connected purchase and suppression events through an authenticated API, while retaining the warehouse for analysis.
5. It introduced validation rules for email format, country, duplicate accounts and impossible dates.
6. It replaced the immediate third-party purchase with contextual prospecting and a small, contractually reviewed second-party pilot.
7. It created dashboards for duplicate rate, event latency, consent conflicts, correction requests and campaign complaints.

Outcome

Within one quarter, duplicate promotional messages declined, service agents gained a more complete view and campaign suppression became faster. Northstar did not claim that all profiles represented unique people; instead, it reported match confidence and excluded uncertain matches from sensitive personalisation. The organisation also learned that improving first-party data capture and governance produced more durable value than simply adding an external audience source.

Learning points

The case demonstrates that brand data maturity is organisational as well as technical. A CDP or API could not solve unclear definitions by itself. Identity resolution must be risk-based; data quality must be measured by purpose; and customer trust is an operational outcome, not merely a legal checkbox.

12. Professional Implementation Framework

A brand manager assessing a data ecosystem can use the following sequence:

1. Define the customer and brand objective. Do not begin with a technology purchase.
2. Map the journey and touchpoints. Record where events occur and what the customer expects.
3. Catalogue data sources. Document owner, origin, fields, sensitivity, purpose and retention.
4. Assess quality and identity. Test completeness, duplication, accuracy, latency and match confidence.
5. Design the minimum viable architecture. Select CRM, CDP, warehouse, APIs and activation tools according to need.
6. Build consent and access controls. Link permissions to purposes and channels.
7. Pilot a measurable use case. Examples include service continuity, suppression or onboarding.
8. Evaluate customer and business outcomes. Include complaints, opt-outs, fairness and service quality, not only conversion.
9. Monitor and improve. Track data drift, vendor changes, model performance and regulatory developments.

Useful metrics may include delivery success, profile duplication, identity-match precision, event latency, opt-out rate, complaint rate, incremental conversion and customer satisfaction. Metrics should not encourage excessive collection or reward intrusive tactics.

Challenges and Limitations

Digital ecosystem programmes face several persistent limitations. Data silos can be political as well as technical: departments may disagree about ownership or fear loss of control. Platform identifiers are often unavailable or unstable. Data may be biased toward digitally active customers, creating an incomplete view of the market. Integration projects can become expensive when every system is treated as critical and every historical field is retained.

Attribution is also uncertain. A recorded touchpoint does not prove that it caused a purchase. Personalisation can increase short-term response while weakening long-term trust. Probabilistic identity can improve reach while increasing mistaken merges. AI can increase efficiency while making decisions less explainable. These trade-offs require testing, restraint and governance rather than a claim that data is inherently objective.

Best-Practice Principles

• Collect data for clear, communicated purposes.
• Prefer quality and relevance to volume.
• Treat first-party relationships as responsibilities, not only assets.
• Make identity decisions proportionate to risk.
• Preserve lineage from collection to activation.
• Separate observed facts, declared preferences and model inferences.
• Use privacy-preserving measurement where possible.
• Give customers meaningful control and correction routes.
• Design systems for deletion, retention and access from the beginning.
• Evaluate customer welfare and brand trust alongside commercial performance.

Chapter Summary

Brand data ecosystems connect customers, touchpoints, data assets, technology and governance. First-party data comes directly from a brand’s interactions; second-party data is shared by a partner; and third-party data comes through an external intermediary. Zero-party data is intentionally declared by customers. Each category has value and risk, and none is automatically accurate or permissible.

Customer-journey mapping links interactions to data creation, storage and activation. CRM systems support operational relationship management, while CDPs unify data and activate audiences; warehouses, lakes, APIs and cloud services provide complementary infrastructure. Data quality includes accuracy, completeness, consistency, timeliness, validity, uniqueness, accessibility and traceability. Identity resolution may be deterministic or probabilistic, and uncertain matches should not be used carelessly.

Cookies and other identifiers are changing as browsers, platforms and privacy expectations restrict cross-context tracking. Brands therefore need consent-based first-party relationships, contextual approaches, aggregated measurement and privacy-preserving collaboration. AI can improve analysis and activation, but it magnifies poor data and bias unless governed through documentation, testing, monitoring and human accountability.

The central managerial lesson is that data capability is not merely a technology project. It is a coordinated brand practice involving customer value, architecture, quality, identity, privacy, ethics and organisational responsibility.

Key Terms

First-party data: Data collected directly by an organisation through its own customer or prospect interactions.

Second-party data: Another organisation’s first-party data shared through a direct relationship.

Third-party data: Data obtained from an entity that did not collect it directly through the brand’s relationship.

Zero-party data: Information a customer intentionally and proactively provides.

Digital ecosystem: Interconnected people, organisations, platforms, technologies and rules involved in digital value creation.

Customer journey: The sequence of customer needs, interactions and experiences related to a brand.

CRM: A system supporting customer accounts, sales, service, interactions and relationship processes.

Customer data platform: A platform that ingests, unifies and activates customer data from multiple sources.

Data warehouse: A structured analytical repository designed for governed reporting and analysis.

API: A defined interface through which software systems exchange data or functions.

Data quality: The fitness of data for a stated purpose, including accuracy and completeness.

Identity resolution: The process of determining whether records refer to the same entity.

Deterministic matching: Identity matching based on exact or verified identifiers and rules.

Probabilistic matching: Identity matching based on estimated likelihood from multiple signals.

Cookie: A browser-stored data object associated with a website or service.

Consent management: The processes used to record, apply and respect customer permissions.

Data lineage: The documented history of data origin, movement, transformation and use.

Clean room: A controlled environment for analysing or matching data without broadly exposing raw records.

Data minimisation: Collecting and retaining no more personal data than is necessary for the purpose.

Incrementality: The additional outcome caused by an intervention compared with an appropriate counterfactual.

Review Questions

Short Answer Questions

1. What distinguishes first-party data from second-party data?
2. Why is a CDP not automatically a single source of truth?
3. Name four dimensions of data quality.
4. What is the difference between deterministic and probabilistic identity matching?
5. Why has third-party cookie dependence become less reliable?

Long Answer Questions

1. Explain how data can be mapped across the stages of a customer journey and discuss the governance questions that should accompany each stage.
2. Compare CRM systems, CDPs and data warehouses in terms of purpose, strengths and limitations.
3. Discuss how data quality and identity resolution affect customer experience and brand trust.
4. Evaluate the strategic shift from third-party tracking toward first-party, contextual and privacy-preserving approaches.
5. Explain the ethical and managerial risks of using AI for customer segmentation and next-best-action decisions.

Analytical / Application Questions

1. A retailer has separate store, website and loyalty records with duplicate customers. Design an identity-resolution and data-quality plan, including controls for false matches.
2. A brand wants to share customer audiences with a partner. Recommend the due-diligence, consent, contractual and measurement steps it should complete before activation.
3. Using the Northstar case, identify which problems are technical, which are governance problems and which require organisational change. Justify your classification.

Multiple Choice Questions with Answers

Question 1

Which statement best defines first-party data?

A. Data purchased from an unrelated data broker

B. Data collected directly through a brand’s own interactions

C. Data shared by a partner that originally collected it

D. Data generated only by anonymous advertising impressions

Correct Answer: B. Data collected directly through a brand’s own interactions

Question 2

What is the primary operational role of a CRM system?

A. Managing customer-facing sales, service and relationship processes

B. Replacing all advertising platforms

C. Guaranteeing that every customer record is unique

D. Providing legal consent automatically

Correct Answer: A. Managing customer-facing sales, service and relationship processes

Question 3

A dataset contains 18,000 non-missing email addresses among 20,000 records. What is its completeness rate?

A. 80%

B. 85%

C. 90%

D. 95%

Correct Answer: C. 90%

Question 4

Which identity-resolution method relies primarily on verified account identifiers?

A. Deterministic matching

B. Probabilistic matching

C. Contextual targeting

D. Sentiment analysis

Correct Answer: A. Deterministic matching

Question 5

Why can a false-positive identity match be harmful?

A. It prevents all data from being stored

B. It may merge two different people and cause inappropriate personalisation

C. It always increases data completeness

D. It removes the need for consent

Correct Answer: B. It may merge two different people and cause inappropriate personalisation

Question 6

What is a key function of an API in a brand data ecosystem?

A. It physically replaces a cloud data centre

B. It defines how software systems exchange data or functions

C. It guarantees that data is accurate

D. It converts all third-party data into first-party data

Correct Answer: B. It defines how software systems exchange data or functions

Question 7

Which practice best reflects data minimisation?

A. Retaining every historical field indefinitely

B. Collecting and retaining only data necessary for a stated purpose

C. Buying the largest available audience file

D. Using inferred sensitive attributes whenever possible

Correct Answer: B. Collecting and retaining only data necessary for a stated purpose

Question 8

What is a major limitation of probabilistic identity matching?

A. It cannot use more than one signal

B. It can create false positives and false negatives

C. It requires no validation

D. It works only with paper records

Correct Answer: B. It can create false positives and false negatives

Question 9

Which approach reduces dependence on individual-level cross-site tracking?

A. Contextual advertising and aggregated measurement

B. Collecting unlimited browser histories

C. Ignoring consent preferences

D. Merging all household devices into one person

Correct Answer: A. Contextual advertising and aggregated measurement

Question 10

Why does AI require strong data governance in brand applications?

A. AI eliminates historical bias automatically

B. AI can amplify poor-quality data, biased patterns and incorrect identity links

C. AI cannot process customer data

D. AI makes privacy obligations irrelevant

Correct Answer: B. AI can amplify poor-quality data, biased patterns and incorrect identity links