• Written by Muhammed Abdulla NC, CEO

On Monday morning, a purchasing manager checks the purchase order table in AsscherAi to see which orders are pending approval, only to find that the numbers do not match the enterprise resource planning system. This discrepancy is a common problem when integrating AI systems with existing systems of record, and it stems from a lack of care in designing the write paths to these systems.

Read paths, write paths and why the second needs more care

When integrating AsscherAi with other systems, there are two types of paths to consider: read paths and write paths. Read paths are used to retrieve data from the systems of record, while write paths are used to update these systems. While read paths are relatively straightforward, write paths require more care because they can potentially introduce inconsistencies into the system. For example, if the write path to the purchase order table is not properly synchronized with the enterprise resource planning system, it can lead to discrepancies like the one the purchasing manager encountered.

A thorough understanding of the write paths is necessary to ensure that the data in AsscherAi remains consistent with the systems of record. This requires careful consideration of the update mechanisms and the potential for conflicts or errors. By taking the time to design and test the write paths, organizations can avoid the pitfalls of inconsistent data and ensure that their AI systems provide accurate and reliable information.

Reconciling definitions that differ between systems

Different systems often have different definitions for the same data elements, which can cause problems when integrating these systems. For instance, the definition of a "pending" order may vary between the purchase order table and the enterprise resource planning system. To reconcile these differences, it is necessary to establish a common understanding of the data elements and their definitions. This can be achieved through a process of data mapping and transformation, which ensures that the data is consistent across all systems.

Reconciling definitions is a critical step in integrating AI systems with existing systems of record. It requires a thorough understanding of the data elements and their relationships, as well as the ability to establish a common language and set of definitions. By taking the time to reconcile these differences, organizations can ensure that their AI systems provide accurate and reliable information, and that the data is consistent across all systems.

Latency budgets and what can be precomputed

Latency is a critical consideration when integrating AI systems with existing systems of record. The latency budget refers to the amount of time that is available to retrieve and process the data before it is needed. In some cases, it may be possible to precompute certain data elements or queries to reduce the latency and improve the performance of the system. For example, the purchasing manager may be able to precompute the list of pending orders to reduce the time it takes to retrieve this information.

Understanding the latency budget and what can be precomputed is essential to ensuring that the AI system provides fast and reliable information. By precomputing certain data elements or queries, organizations can reduce the latency and improve the performance of the system, making it more responsive and useful to the users. To learn more about optimizing the performance of AsscherAi, visit our documentation page.

What happens when a source system is down

Despite the best efforts to design and test the integration, there will be times when a source system is down or unavailable. In these cases, the AI system must be able to handle the situation gracefully and provide a useful response to the user. For example, if the enterprise resource planning system is down, the purchasing manager may still be able to retrieve some information from the purchase order table, even if it is not up to date.

Handling downtime and unavailable systems is a critical aspect of integrating AI systems with existing systems of record. It requires careful consideration of the potential failure modes and the development of strategies to mitigate these failures. By taking the time to design and test the integration, organizations can ensure that their AI systems provide accurate and reliable information, even in the face of system downtime.

Common pitfalls and misconceptions

Many people believe that integrating AI systems with existing systems of record is a straightforward process that can be completed quickly and easily. However, this is not the case. Integrating AI systems requires careful consideration of the read and write paths, the reconciliation of definitions, and the handling of latency and downtime. It is a complex process that requires significant time and effort to get right.

Another common misconception is that the AI system can simply override the existing systems of record. However, this is not the case. The AI system must be designed to work in conjunction with the existing systems, providing a complementary view of the data rather than replacing it. By understanding these common pitfalls and misconceptions, organizations can avoid the mistakes of the past and ensure that their AI systems provide accurate and reliable information.

When integration is not the right choice

While integration is often the best approach, there are cases where it is not the right choice. For example, if the existing systems of record are highly customized or proprietary, it may be more difficult to integrate the AI system. In these cases, it may be necessary to consider alternative approaches, such as building a new system from scratch or using a different technology platform.

To determine whether integration is the right choice, organizations should carefully consider their goals and requirements. They should also consult with experts and conduct thorough research to determine the best approach. For more information on how to get started with AsscherAi, visit our contact us page.