• Written by Azhar Aliyar, Director

A category manager comparing two suppliers using AsscherAi may wonder whether to use vector search or keyword search to find the relevant information in the purchase order table. This decision is not as straightforward as it seems, and the answer is usually both, depending on the specific situation.

Commitments of each option

Vector search commits you to a specific way of indexing the data in the purchase order table, which can be time-consuming to set up and maintain. It requires a deep understanding of the data and the relationships between different fields, as well as the ability to fine-tune the search algorithm to get the best results. On the other hand, keyword search is a more straightforward approach that simply looks for exact matches of the search terms in the data. However, this can lead to a larger number of false positives and false negatives, especially if the search terms are not well-chosen.

The choice between vector search and keyword search also depends on the type of data being searched. For example, if the data is mostly unstructured text, such as product descriptions or customer reviews, vector search may be a better choice. But if the data is mostly structured, such as dates or numbers, keyword search may be sufficient.

The case for the other option

There are cases where keyword search is the better choice, even if vector search is available. For example, if the category manager is looking for a specific supplier or product, a keyword search may be faster and more efficient. This is because keyword search can take advantage of indexes on the data, which can speed up the search process. Vector search, on the other hand, often requires a full scan of the data, which can be slower.

Another case where keyword search may be preferred is when the search terms are very specific and well-defined. In this case, the category manager may want to use a keyword search to find exact matches, rather than relying on the more nuanced results of a vector search. For example, if the category manager is looking for all purchase orders from a specific supplier, a keyword search may be the better choice.

Hidden costs

One of the hidden costs of choosing between vector search and keyword search is the cost of maintaining the search indexes. Vector search requires a significant amount of computational resources to maintain the indexes, especially if the data is large and complex. Keyword search, on the other hand, requires less computational resources, but may require more manual maintenance to ensure that the indexes are up-to-date.

Another hidden cost is the cost of training and supporting users. If the category manager is not familiar with the search interface, they may require training and support to use it effectively. This can be especially true for vector search, which can be more complex and nuanced than keyword search. The AsscherAi documentation can provide some guidance on how to use the search interface, but additional support may be necessary.

Deciding without a six month evaluation

One way to decide between vector search and keyword search without a six month evaluation is to start with a small pilot project. This can involve setting up a small test dataset and trying out both vector search and keyword search to see which one works better. The category manager can then use the results of the pilot project to inform their decision about which search approach to use.

Another approach is to use a hybrid approach that combines elements of both vector search and keyword search. For example, the category manager could use keyword search to narrow down the results to a specific subset of data, and then use vector search to refine the results further. This approach can provide the benefits of both search approaches, while minimizing the drawbacks.

Limitations of this approach

This approach to deciding between vector search and keyword search is not suitable for all situations. For example, if the data is extremely large and complex, a more comprehensive evaluation may be necessary to determine which search approach is best. In this case, the category manager may need to consult with a search expert or conduct a more thorough analysis of the data and search requirements.

In some cases, the category manager may need to contact AsscherAi support for additional guidance on how to use the search interface and determine which search approach is best for their specific use case. This can be especially true if the category manager is not familiar with the search interface or has specific requirements that are not met by the standard search functionality.