Cost optimisation for production
AI cost optimisation for production reduces waste by analysing entire workflows, considering unit cost versus total cost, and applying caching, batching, and model routing to minimize computational resources, resulting in more consistent and reliable production, and reduced unplanned downtime and stock cover.
Every week, production managers sit through meetings where cost optimisation plans are presented, only to see them fail to survive contact with the production floor. The business consequence is clear: money is wasted on plans that do not account for the complexities of production. The mechanism behind this failure lies in the misunderstanding of unit cost versus total cost of an AI workflow. Unit cost focuses on the individual components, while total cost considers the entire workflow, including the interactions between components.
The difference between unit cost and total cost of an AI workflow
Most people get wrong the idea that reducing unit cost will automatically lead to a reduction in total cost. This is wrong because it neglects the impact of workflow interactions on total cost. For instance, a reduction in unit cost may lead to an increase in the number of components required, resulting in a higher total cost. A shift supervisor checks the maintenance log to identify areas where unit cost reductions can be made without increasing the total cost.
The total cost of an AI workflow includes not only the cost of individual components but also the cost of training, deployment, and maintenance. Therefore, any cost optimisation plan must consider the entire workflow to ensure that reductions in unit cost do not lead to increases in total cost.
Caching, batching and model routing as the three real levers
Caching, batching, and model routing are the three levers that can be used to reduce the total cost of an AI workflow. Caching involves storing frequently accessed data to reduce the need for redundant calculations, while batching involves processing multiple requests together to reduce overhead. Model routing involves directing requests to the most efficient model available, reducing the computational resources required.
These levers can be used to reduce the total cost of an AI workflow by minimizing the computational resources required. For example, a category manager compares two suppliers to determine which one can provide the most efficient caching and batching solutions for their AI workflow.
The production data this depends on
The effectiveness of cost optimisation plans depends on the availability of production data, including output against plan, cycle times, scrap and quality records, and maintenance history. This data is used to identify areas where costs can be reduced without impacting production. A production manager uses this data to determine the most effective caching and batching strategies for their AI workflow.
Output against plan data helps to identify areas where production is not meeting expectations, while cycle times and scrap and quality records help to identify areas where costs can be reduced. Maintenance history data helps to identify areas where equipment is not being used efficiently, allowing for more effective model routing.
What it changes about unplanned downtime and stock cover
Cost optimisation plans can help to reduce unplanned downtime and stock cover against the plan by identifying areas where costs can be reduced without impacting production. By minimizing the computational resources required, caching, batching, and model routing can help to reduce the likelihood of unplanned downtime.
For example, a production manager uses production data to identify areas where caching and batching can be used to reduce the computational resources required, resulting in a reduction in unplanned downtime. This, in turn, helps to reduce stock cover against the plan, as production is more consistent and reliable.
Measured with OEE, unplanned downtime hours, and scrap rate
The effectiveness of cost optimisation plans is measured using key performance indicators such as Overall Equipment Effectiveness (OEE), unplanned downtime hours, and scrap rate. These metrics help to identify areas where costs can be reduced without impacting production.
By tracking these metrics, production managers can determine the effectiveness of their cost optimisation plans and make adjustments as needed. For example, a reduction in OEE may indicate that caching and batching strategies need to be adjusted to minimize computational resources required.
What this does not do
This approach to cost optimisation does not replace the need for human judgement and expertise. While AI can help to identify areas where costs can be reduced, human expertise is still required to interpret the results and make decisions. If you are unsure about how to implement cost optimisation plans for your production workflow, contact us to discuss your options.
Frequently asked questions
How can AI cost optimisation improve production efficiency?
AI cost optimisation improves production efficiency by identifying areas where costs can be reduced without impacting production, and applying strategies like caching and batching.
What is the difference between unit cost and total cost in AI workflows?
Unit cost focuses on individual components, while total cost considers the entire workflow, including interactions between components, to ensure reductions in unit cost do not lead to increases in total cost.
How can production data be used to inform cost optimisation plans?
Production data, including output against plan, cycle times, and maintenance history, is used to identify areas where costs can be reduced without impacting production, and to determine effective caching and batching strategies.
What metrics can be used to measure the effectiveness of cost optimisation plans?
Key performance indicators like Overall Equipment Effectiveness, unplanned downtime hours, and scrap rate are used to measure the effectiveness of cost optimisation plans and identify areas for improvement.