Integrations & data

ERP system integration with e-commerce, WMS, CRM and BI

Maciej Zaleski · Published · Updated · 7 min read

ERP system integration involves an agreed exchange of data and events between a business management system and applications supporting sales, warehousing, customer relationships, and analytics. The choice of API, middleware, or ETL should follow from the business process, required information freshness, and capabilities of specific software versions. Simply establishing a connection is not enough if the company has not defined data owners and error-handling rules.

In practice, a project often begins with a request to send store orders to ERP. Further questions quickly arise, however. Which system sets the price? How should inventory reservations be handled? What happens if the customer changes their address after picking has begun? The answers define the solution's scope far more strongly than the choice of integration library.

ERP system integration starts with the process

First describe the workflow from a business event to its settlement. An order, shipment, and payment are not the same event, even though they may concern one transaction. Each has its own status, creation time, and consequences for other systems. Integration should preserve these distinctions rather than reduce everything to a single field.

For each process, define the source system, recipients, and expected update time. Product availability may need more frequent synchronization than a management report. Not all data must flow in real time. Overambitious requirements increase dependencies and complicate maintenance without an obvious business benefit.

The process owner should also approve exception scenarios. A missing product, incorrect customer number, or interrupted communication is part of normal operations, not an exotic case. Decide in advance which errors the integration will retry automatically and which should go to a decision-maker.

Business Central, Comarch and SAP — check the specific installation

A product name is not enough to design a connection. Version, deployment model, available modules, extensions, and licensing rights matter. Business Central, Comarch solutions, and SAP products may offer different data-exchange mechanisms depending on their particular configuration. Interface availability must be confirmed in documentation and the customer's environment.

Before estimating costs, collect a description of existing modifications and information about the partner maintaining ERP. Check whether updates will change the integration contract and who approves new fields and extensions. Direct database writes that bypass application logic require particular caution and confirmation of a supported path. The apparent simplicity of this solution may conceal inconsistency risks.

Go2ops carries out work involving Dynamics NAV and Business Central and workflow automation in pharmaceuticals. You can discuss this industry's specifics through our pharma solutions area. This does not mean a design from one implementation can be transferred to another company or ERP product without analysis.

What to exchange with e-commerce, WMS, CRM and BI

Each application represents a different part of the business. Instead of one general synchronization requirement, prepare a map of objects, directions, and responsibilities. Indicate whether the recipient can change received data or only use it. This helps prevent several systems from endlessly overwriting information.

System Example data Key question
E-commerce Products, prices, orders, availability When is an order accepted for fulfillment?
WMS Reservations, receipts, issues, stock levels Who confirms that an operation was actually performed?
CRM Customers, account managers, relationship statuses Which system manages customer master data?
BI Documents, events, reference lists, measures How should data history and corrections be interpreted?

Distinguishing physical, available, and reserved stock is particularly important. Likewise, a customer in CRM does not always correspond to one buyer or payer in ERP. These differences should be visible in the data model. Hiding them in ad hoc mapping rules makes later reporting and discrepancy investigation harder.

API, middleware and ETL need not be mutually exclusive

An API is a system's communication interface, middleware organizes exchanges between applications, and ETL describes extracting, transforming, and loading data. They are not three separate products solving the same problem. An ETL process can use an API, and an intermediary layer can handle both events and scheduled jobs.

When a direct API is enough

A direct connection can be reasonable with a limited scope and few dependencies. It still requires authentication, limit controls, error logging, and retries. Check whether the recipient supports a request identifier that prevents the same order from being created multiple times after a connection interruption.

When an intermediary layer helps

Middleware helps decouple systems, centralize mappings, and handle queues and transformations. It becomes more important as applications, partners, or process variants multiply. However, it becomes a service requiring maintenance itself. Plan monitoring, administrative access, recovery, and responsibility for configuration changes.

When to use ETL

ETL works well for feeding data warehouses and processing data where delay is acceptable. Include incremental extraction, historical corrections, and the ability to recalculate a range. If a report compares data from different points in time, it should clearly show its freshness so the user does not interpret incomparable values as a sales error.

Data quality and clear responsibility

Integration quickly exposes inconsistencies previously hidden in spreadsheets and manual work. The same products may have different identifiers, and customers may appear multiple times. Before starting the exchange, agree on identifiers, units of measure, currencies, date formats, and permitted status values. The rules must also be understandable to business staff.

Prepare identifier mappings between systems rather than assuming they match. Appoint an owner for every important reference list and define how changes are approved. Data cleansing should not mean silently fixing everything in the integration layer. Otherwise, the source error remains, and subsequent processes replicate different interpretations.

If a process involves personal data or documentation subject to specific requirements, the scope of transmitted information and retention period must be agreed separately. Current rules should be verified with a data protection or compliance specialist. A technical connection between systems does not itself confirm fulfillment of legal obligations.

Common pitfalls and acceptance

A common pitfall is building only the happy path. In the demonstration, an order reaches ERP, but after a failure nobody knows which messages were saved. State reconciliation and safe retry mechanisms are needed. A correlation identifier linking logs to business documents without exposing excessive data helps.

Include duplicates, out-of-order messages, corrections, and the unavailability of one system in testing. Also check behavior after data-schema changes. Acceptance should include an operating manual and named people responsible for the error queue. A green technical-job status does not yet prove that documents match on both sides.

Discuss architecture and maintenance scope as part of systems and data integration. If an automation tool will handle part of the flow, the comparison of n8n, Make and Zapier will help. For processes beginning with files, AI document-processing automation is an important complement, with a separate validation stage before writing to ERP.

FAQ

Does integration require replacing ERP?

No. First assess the interfaces, extensions, and limitations of the current installation. Replacement can be a separate business project but should not result solely from a failure to analyze ways to connect existing tools.

Should all data synchronize immediately?

No. Frequency should match the consequences of delay. Accepting an order may require a different approach from a periodic report. Sending everything continuously increases complexity and does not always improve the process.

How can duplicate orders be avoided?

Persistent identifiers, checks on processed messages, and agreed retry rules are needed. A particularly important situation is when ERP saved a document but the response never arrived. A repeated request should not automatically create another document.

Who should be responsible for incorrect data?

The owner of the relevant business object or process, supported by IT. Integration should identify the cause and enable correction but cannot independently resolve every commercial ambiguity. Responsibility must be agreed before launch.

Summary

Successful integration combines technical architecture with clear business rules. Start with processes, data owners, and exceptions, and only then select the communication mechanism. If you plan to connect ERP to further systems, Go2ops can help organize the analysis scope and discuss possible implementation options.

MZ

Maciej Zaleski

Managing Director, Go2ops

Related articles