AI Glossary / Data engineering and architecture / Beginner

Information integration

Definition

Information integration brings information from different sources into a coherent form or view so that people and applications can use it together. It involves more than connecting systems: structures, meanings, identities and context must be reconciled. Integration may create consolidated copies or provide access across sources that remain separate. The chosen approach determines freshness, performance and operational dependencies.

What it means

A business question often spans several systems. Sales records may identify an account, support tickets a contact and contracts a legal entity. Simply displaying these records together does not show how they relate. Information integration establishes the mappings and interpretations needed for the intended use. The term often includes structured databases and less structured content such as documents. Its boundaries overlap with data integration, and particular vendors may use the terminology more narrowly.

How it works

Teams identify relevant sources and examine their formats, identifiers, metadata and meaning. They define mappings, transformation rules and handling for missing or conflicting values. A pipeline may copy information into a shared store, or a federation layer may query distributed sources through a common interface. Validation checks expected relationships and totals. Lineage records where results came from. Ongoing monitoring addresses source changes, access failures and stale information; a successful initial connection is only one step in maintaining an integrated service.

Three practical examples

• A service dashboard connects contract status with support activity using a documented account mapping.
• A research portal combines publication metadata with links to underlying datasets while preserving the source identifiers.
• An AI assistant retrieves approved documents and structured product facts through a common application layer, retaining source references and access rules.

Think of it this way

Think of assembling a meeting briefing from several departments. A useful briefing resolves whether their terms, dates and units agree; stapling their reports together is not enough. The analogy captures semantic reconciliation. Software integration also needs repeatable transformations, machine-readable identifiers and handling for partial failures, which a person preparing one briefing might resolve informally.

Worked example

Suppose a sales table contains two orders for account A: 40 and 60 units. A support table contains three tickets for A. Joining both tables directly on account and then summing order units creates six joined rows: each order appears once per ticket. The resulting sum is 300 instead of 100. A correct design can aggregate orders and tickets separately at account level before joining, yielding 100 units and three tickets. The example shows why grain and relationship multiplicity matter even when identifiers and file formats match perfectly.

Illustrative case study

In a fictional company, Alder Equipment builds an account overview from sales and support systems. The prototype produces inflated totals because it combines records at different levels of detail. The team documents each source's grain, corrects the aggregation plan and tests accounts with several orders and several tickets. It also labels the refresh time for each component. The resulting design is more interpretable, but the case claims no measured commercial gain and does not imply that a single join strategy suits every question.

Common misconceptions and mistakes

Integration is not synonymous with moving all information into one database. Federation can leave sources in place, although availability and query performance then depend on those sources. Matching column names does not establish matching meaning. Combining records can duplicate or omit information when relationship rules are wrong. A unified view does not make every source equally authoritative or equally fresh. It also does not authorize wider access: permissions and purpose restrictions need to survive the integration process.

Business use

Start from a concrete question and define the required entities, time period and level of detail. Identify the authoritative source for each field and specify conflict rules. Test difficult examples, including duplicate identifiers, late updates and missing matches. For AI applications, preserve provenance so reviewers can distinguish retrieved evidence from generated interpretation. Monitor mapping changes as carefully as connection failures. A system can continue running while silently producing misleading answers after a source changes its definition of active customer or revenue.

People also ask

Is information integration the same as application integration?

They overlap, but application integration often emphasizes coordinating actions and workflows, while information integration emphasizes a usable view of information.

Must integrated information be real time?

No. The required freshness depends on the use case and architecture.

Can an AI model do the mapping automatically?

It can suggest mappings, but meanings, exceptions and consequences still require validation against source documentation and business rules.

Check your understanding

Two source tables share an account ID, so an analyst joins them and sums a monetary column. Is that enough to ensure a correct total?

Answer: No. The analyst must examine the grain and relationship multiplicity. A many-to-many join can repeat values and inflate the result even when the join key is correct.

Related terms

• Semantic mapping
• ETL
• Data integration
• data federation
• entity resolution
• data lineage
• canonical model

Learn next

Write down the grain of two small datasets and predict the number of rows their join will produce. Then compare consolidation with federation for the same question. Include freshness, source availability and permission behavior in the comparison, rather than judging only whether a connector exists.

More in Data engineering and architecture

• Indexed access - Indexed access retrieves data through an index that organizes searchable values and points to, or otherwise identifies, relevant records.
• Industry standard data model - An industry standard data model is a shared description of important data concepts, attributes and relationships for a particular sector, established...
• Information governance - Information governance is the coordinated system of responsibilities, policies and oversight through which an organization manages information throughout...
• Information model - An information model describes the concepts in a chosen domain, their properties and relationships, and the rules that give them meaning.
• Information repository - An information repository is an organized environment for storing, describing, managing and retrieving information.
• Infrastructure as a service - Infrastructure as a service (IaaS) is a cloud service model that provides fundamental computing resources such as processing, storage and networking.

Sources

Sources: Roth and colleagues, “Information integration: A new generation of information technology,” IBM Systems Journal, 2002, https://research.ibm.com/publications/information-integration-a-new-generation-of-information-technology; IBM, “InfoSphere Federation Server,” https://www.ibm.com/docs/en/iis/11.5.0?topic=components-infosphere-federation-server; IBM, “What is Data Integration?” https://www.ibm.com/think/topics/data-integration. Primary research and product documentation search excerpts consulted; full-document and editorial review remain pending. Historical product sources illustrate architectural approaches, not a recommendation to deploy those versions.

Previous: Information governance Next: Information matrix test