• Written by Muhammed Abdulla NC, CEO

Many people object to using AI for cost optimisation in metals and mining because they believe it cannot handle the complexity of the industry, but this objection is misguided because it misunderstands where the real challenges lie. The actual difficulty in optimising costs in metals and mining stems from the unique characteristics of the industry, such as grade variability, maintenance backlog, and energy intensity, which can make it hard to predict and control costs.

Industry challenges

The metals and mining industry is particularly challenging for cost optimisation due to the inherent variability of the grade of the ore being mined, which can significantly impact production costs. Additionally, the maintenance backlog of heavy assets can lead to unexpected downtime and increased maintenance costs. The energy intensity of the industry also means that small changes in energy prices can have a large impact on overall costs.

These challenges are further complicated by the long cycles of production in the industry, which can make it difficult to respond quickly to changes in the market or in production conditions. As a result, any cost optimisation effort must be able to take into account these unique characteristics of the industry in order to be effective.

Unit cost vs total cost

A common mistake in cost optimisation efforts is to focus solely on reducing the unit cost of production, without considering the total cost of the AI workflow itself. This can lead to situations where the cost of implementing and maintaining the AI system outweighs any potential savings from reduced unit costs. In contrast, a more effective approach is to consider the total cost of the AI workflow, including the cost of data preparation, model training, and deployment.

By taking a holistic view of the costs involved, organisations can make more informed decisions about where to focus their cost optimisation efforts, and avoid implementing solutions that may not provide a net benefit. For example, a company may find that the cost of collecting and preparing the data needed to train an AI model outweighs any potential savings from using that model to optimise production.

Real levers for cost optimisation

There are three key levers that organisations can use to optimise the cost of AI workflows in metals and mining: caching, batching, and model routing. Caching involves storing the results of expensive computations so that they can be reused instead of recalculated, which can significantly reduce the computational cost of the AI workflow. Batching involves processing multiple inputs together as a single batch, which can reduce the overhead cost of processing each input individually.

Model routing involves using multiple models in tandem to process different types of inputs, which can help to reduce the cost of processing complex or variable inputs. By using these levers, organisations can reduce the cost of their AI workflows and make them more efficient and effective. For more information on how to implement these levers, visit our website.

Token cost is not the dominant line

Many people assume that the token cost of an AI model is the dominant line item in the cost of the AI workflow, but this is rarely the case. In reality, the cost of data preparation, model training, and deployment often outweighs the token cost of the model itself. Additionally, the cost of integrating the AI model with existing systems and processes can also be significant.

As a result, organisations should focus on optimising the overall cost of the AI workflow, rather than just the token cost of the model. This may involve investing in automation tools to reduce the cost of data preparation, or working with experts to optimise the deployment and integration of the AI model.

Limitations of AI cost optimisation

While AI can be a powerful tool for cost optimisation in metals and mining, it is not a silver bullet and there are limitations to its use. For example, AI may not be effective in situations where the production process is highly variable or unpredictable, or where the cost structure is highly complex. In these situations, other approaches to cost optimisation may be more effective.

Additionally, AI cost optimisation may not be suitable for organisations that lack the necessary data or infrastructure to support the use of AI. In these cases, more traditional approaches to cost optimisation may be more effective, at least until the necessary infrastructure and data are in place. By understanding the limitations of AI cost optimisation, organisations can make more informed decisions about how to use this technology to achieve their goals.