• Written by Muhammed Abdulla NC, CEO

When a purchasing manager asks AsscherAi how many orders are pending for a specific supplier, the system returns an answer that seems straightforward, but the manager soon realises that the number is lower than expected, because the purchase order table only includes orders that have been fully approved, excluding those that are still in draft or awaiting review. This discrepancy highlights the importance of understanding the field-level meaning of the data, as the column name "order_status" does not provide sufficient context for the system to accurately answer the question.

Field-level meaning and column names

The issue with relying solely on column names as documentation is that they often lack the nuance required to accurately convey the meaning of the data. For instance, a column named "customer_id" may seem self-explanatory, but it does not indicate whether the ID refers to the customer's account number, their loyalty program ID, or some other identifier. To ensure that AsscherAi provides accurate answers, it is essential to have a clear understanding of the field-level meaning of the data, including the specific definitions and constraints that apply to each column.

A shift supervisor checking the maintenance log, for example, needs to know that the "equipment_id" column refers to the unique identifier assigned to each piece of equipment, and that it is used to track maintenance schedules and repair histories. Without this context, the system may return incomplete or inaccurate information, leading to misunderstandings and potential errors.

Systems that get left out of scope

When implementing AsscherAi, it is easy to overlook certain systems or data sources that may seem peripheral or irrelevant, but these omissions can lead to partial answers and incomplete information. For instance, if the enterprise resource planning system is not integrated with AsscherAi, the system may not have access to critical data on inventory levels, supply chain logistics, or production schedules.

This can result in answers that are incomplete or misleading, as the system is only considering a limited subset of the available data. To avoid this, it is essential to carefully scope the implementation and ensure that all relevant systems and data sources are included, even if they seem minor or insignificant.

Stockouts, backfills, and other gaps

Stockouts, backfills, and other gaps in the data can teach a model the wrong lesson, as it may learn to recognize patterns or relationships that are not actually present in the data. For example, if there is a stockout of a particular product, the system may learn to associate the product with low sales or low demand, when in fact the product is simply out of stock.

This can lead to inaccurate predictions or recommendations, as the system is basing its decisions on incomplete or flawed data. To mitigate this, it is essential to carefully monitor the data for gaps or inconsistencies and take steps to address them, such as implementing data validation or data cleansing processes.

Data quality problems become visible

Once answers are presented in sentences, data quality problems become much more visible, as the system is forced to reconcile inconsistencies and ambiguities in the data. For instance, if the system returns an answer that states "the customer has placed 10 orders, but the order total is only 5", it becomes clear that there is a discrepancy in the data that needs to be addressed.

This can be a valuable opportunity to identify and correct data quality issues, as the system is able to highlight inconsistencies and ambiguities that may have gone unnoticed otherwise. By addressing these issues, the enterprise can improve the overall accuracy and reliability of the system, and ensure that it is providing the most accurate and helpful answers possible.

What this does not do

AsscherAi is not a substitute for careful data management and analysis, and it is not a magic bullet that can solve all data-related problems. It is a tool that can help to provide answers and insights, but it is only as good as the data that it is given, and it requires careful implementation and maintenance to ensure that it is providing accurate and reliable results.

For more information on how to implement AsscherAi, or to discuss your specific needs and requirements, please visit our website at www.asscher.ai or contact us directly at www.asscher.ai/contact-us.

When it is the wrong choice

There are certain situations in which AsscherAi may not be the best choice, such as when the data is highly unstructured or uncertain, or when the questions being asked are highly subjective or open-ended. In these cases, other approaches or tools may be more suitable, such as data mining or machine learning algorithms that are specifically designed to handle uncertain or unstructured data.

Ultimately, the decision to use AsscherAi will depend on the specific needs and requirements of the enterprise, and it is essential to carefully evaluate the potential benefits and limitations of the system before making a decision.