What AI actually costs in quality
AsscherAi's true cost lies in data preparation and integration, where a logistics manager must work with the data team to ensure accurate and reliable responses, by cleaning and structuring data from various sources, such as inventory levels and shipping schedules, to support queries like what stock is available for immediate dispatch.
When a large organisation deploys AsscherAi to answer questions about its own data, the expected outcome is a system that provides accurate and timely responses, but in practice, the actual cost of achieving quality is often underestimated, and it accumulates in unexpected places, particularly in the integration and data work required to get the system up and running.
Integration and data work as the dominant line
The biggest cost in achieving quality in an AI deployment is not the AI model itself, but the integration and data work required to prepare the data for the model to work with, and this is where most of the spend accumulates, as the data needs to be cleaned, transformed, and loaded into the system, and the various data sources need to be integrated, which can be a time-consuming and labour-intensive process, and one that requires significant expertise and resources.
A shift supervisor, for example, may need to work with the data team to ensure that the maintenance log data is accurately reflected in the system, and that the data is properly formatted and structured, so that the AI model can provide accurate and reliable responses to queries, and this requires a significant amount of time and effort, as well as close collaboration between the different teams involved.
The integration and data work required for an AI deployment can be complex and challenging, and it requires a deep understanding of the data, the systems, and the processes involved, and it is not something that can be rushed or taken lightly, as the quality of the data and the integration of the systems has a direct impact on the quality of the responses provided by the AI model.
Run cost against query volume
Another area where costs can accumulate is in the run cost of the system, particularly as the volume of queries increases, as the system needs to be able to handle the increased load, and this can require significant additional resources, such as more powerful hardware, additional storage, and increased bandwidth, and these costs can add up quickly, particularly if the system is not properly optimised for performance.
A category manager, for example, may need to compare the costs of different suppliers, and the AI system may need to provide responses to a large number of queries in order to support this process, and if the system is not properly optimised, the run cost can become prohibitive, and the system may not be able to provide the required level of performance, which can have a direct impact on the quality of the responses provided.
The cost of a wrong answer nobody caught
One of the most significant costs of an AI deployment is the cost of a wrong answer that nobody caught, as this can have serious consequences, particularly if the wrong answer is used to make important decisions, and this cost can be difficult to quantify, as it may not be immediately apparent, but it can have a significant impact on the organisation, and it is something that needs to be carefully considered when deploying an AI system.
This is why it is so important to have proper testing and validation procedures in place, to ensure that the AI system is providing accurate and reliable responses, and to catch any errors or inconsistencies before they can cause harm, and this requires a significant amount of time and effort, as well as close collaboration between the different teams involved, and it is not something that can be rushed or taken lightly.
Attributing spend to a process rather than a model
When it comes to attributing spend to an AI deployment, it is often more useful to attribute it to a process rather than a model, as the model is just one part of the overall process, and the costs of the deployment are often spread across multiple areas, such as data preparation, integration, and testing, and attributing spend to a process rather than a model can provide a more accurate picture of where the costs are accumulating, and can help to identify areas where costs can be reduced or optimised.
For example, a production team may be using the AI system as part of a larger production process, and the costs of the AI deployment may be just one part of the overall cost of the process, and attributing spend to the process rather than the model can help to identify areas where costs can be reduced or optimised, and can provide a more accurate picture of the overall cost of the deployment.
When AsscherAi is the wrong choice
While AsscherAi can be a powerful tool for answering questions about an organisation's own data, there are times when it may not be the right choice, such as when the data is not well-structured or is difficult to access, or when the queries are not well-defined or are highly complex, and in these cases, other solutions may be more suitable, and it is important to carefully consider the requirements of the project before deciding whether to use AsscherAi or not, and to contact us to discuss the options in more detail.
In these cases, it is often better to take a step back and re-evaluate the requirements of the project, and to consider alternative solutions that may be more suitable, rather than trying to force AsscherAi to work in a situation where it is not the best fit, and this requires a deep understanding of the data, the systems, and the processes involved, as well as the requirements of the project, and the capabilities and limitations of AsscherAi.
Frequently asked questions
What is the biggest cost in achieving quality in an AI deployment?
The biggest cost is not the AI model itself, but the integration and data work required to prepare the data for the model to work with.
How can the run cost of an AI system be managed?
The run cost can be managed by optimising the system for performance, such as using more powerful hardware or increasing bandwidth, to handle increased query volumes.
What is the most significant cost of an AI deployment?
The most significant cost is the cost of a wrong answer that nobody caught, which can have serious consequences if used to make important decisions.
How should spend be attributed in an AI deployment?
Spend should be attributed to a process rather than a model, as the costs of the deployment are often spread across multiple areas, such as data preparation and testing.