How to integrate systems that do not talk to each other

In this article
When systems do not talk to each other, the most visible symptom is usually copying and pasting information. But integration is not only about moving data from one screen to another. You need to define which information is trustworthy, when it should move, what happens if it fails and who is responsible for correcting it.
That is why it is better to start with the workflow that needs continuity, not with the tool that promises to connect everything. An ERP, a custom application, a machine, a dashboard and an external service may participate in the same operation, but they do not all need to connect in the same way.
Signs that disconnection is creating a cost
- A value is entered in one system and entered again in another.
- Operations teams keep side spreadsheets to work around a software limitation.
- Teams argue about the correct number because each one checks a different source.
- A delay in one system creates no notice in the others.
- A status change requires manual messages before the next team can act.
- There is no clear record of what happened when a transfer failed.
The question is not only how many integrations are missing. It is which decision or task is blocked by the lack of continuity.
Five questions to answer before connecting anything
1. Which system is the source of truth?
For every important data point, define which system creates it and which system is allowed to change it. If an order, batch or work order can change in three places without a clear rule, integration will spread the confusion faster.
2. Which event triggers the exchange?
Some data needs to move immediately — a critical alert, for example — while other information can be sent in batches. Defining frequency avoids overbuilding the architecture and makes acceptable delay explicit.
3. How are the same objects identified?
A customer, product, machine or order may have different codes in each system. Before transferring records, define how those identifiers relate and what happens when no match exists.
4. What happens when something fails?
Every integration needs to detect errors, keep the attempt, prevent duplicates and allow controlled recovery. If someone discovers the problem only when a report no longer reconciles, the connection is hiding risk instead of reducing it.
5. Who owns the data and the operation?
Technology does not replace responsibility. Someone must decide which correction is valid, who reviews an error queue and which team maintains each end of the integration.
Three common paths
Direct API integration
This fits when systems have stable interfaces, the workflow is clear and exchange frequency is defined. Authentication, permissions, limits, data formats and error handling still need to be agreed.
An intermediate integration layer
This can help when several applications participate, data needs transformation or traffic should be observed in one place. It adds a component to maintain, but can organize dependencies and avoid hard-to-govern connections.
Controlled file or batch exchange
Not every process needs real-time data. A validated file, with an agreed schedule and a receipt record, can be reasonable while a deeper integration is prepared. It needs an owner, a format and error controls.
An assisted manual upload can also be a valid stage if it is documented and helps the team learn about the process. What should be avoided is a patch with no owner and no review date.
What if there are ten or twenty systems?
Volume changes the architecture. With two systems, a direct integration is enough; with ten or twenty, connecting everything to everything becomes impossible to maintain. That is when an intermediate layer, an event bus or a controlled batch exchange with a common format makes sense.
| Scenario | Reasonable approach |
|---|---|
| Two systems with a stable API | Direct integration |
| Several systems with shared rules | Intermediate layer or iPaaS |
| Systems that cannot expose an API | Controlled file or batch |
| Processes with real-time events | Event bus or messaging |
| Third-party apps across regions | Data contract, monitoring and retries |
A common case is contact synchronization: the CRM creates a contact, the ERP needs the customer for invoicing and the support system needs the identifier to open tickets. If each integration uses a different identifier, the problem is not connectivity but the data model. First define the source of truth and the event that triggers the update; then choose the technology.
How to structure an integration project
- Choose one end-to-end workflow and define its expected outcome.
- Inventory systems, equipment, files, owners and access constraints.
- Map each data path, including transformations and exceptions.
- Agree on identifiers, statuses, frequency, retries and audit records.
- Test normal cases and simulated failures before expanding scope.
- Check whether the workflow is clearer, more traceable and easier to operate.
The test should not be “the systems are connected.” It should be “the team can complete this workflow with less duplication, less uncertainty and a visible way to detect deviations.”
At Gist Point, our business management platform can be part of the workflow when states and data need to be centralized. When equipment or industrial processes are involved, industrial automation consulting helps evaluate the full operation. To compare technical paths, you can also review technical advisory.
If the project starts with spreadsheets and isolated tools, complement this guide with how to move from spreadsheets to a management system to decide what should be organized first.
A common scenario
An industrial company may have a production order in an administrative system, parameters in a machine and quality checks recorded by the team in a spreadsheet. Connecting all three does not guarantee an improvement. First define which order status enables each action, which data comes from the machine and who validates an exception.
Once those rules are clear, the integration can be staged: start with a record and an alert, add traceability and then introduce automations that make sense for the operation.
Integration is responsibility design
A good integration makes the path of information and the boundaries of each system visible. It does not try to connect everything to everything; it makes sure each data point reaches the right place at the right time, with a recovery path when something goes wrong.
Tell us which systems or equipment you need to connect and we will start by mapping the workflow, constraints and level of integration worth implementing.