• Written by Muhammed Abdulla NC, CEO

This week, the category manager at a large apparel retailer must decide which styles to include in the upcoming seasonal range, and how to balance the need for fresh designs with the risk of markdowns on unsold stock. Currently, this decision is based on a combination of historical sales data, market trends, and intuition, but the complexity of the supply chain and the uncertainty of consumer demand make it difficult to optimise costs effectively.

Textiles and apparel complexities

The textiles and apparel industry presents unique challenges for cost optimisation, particularly in range planning under uncertainty. The long lead times required for production and shipping mean that decisions must be made months in advance, without knowing exactly what consumers will want to buy. Supplier fragmentation is another issue, as many different suppliers are often involved in the production of a single garment, making it hard to negotiate prices and manage inventory. Markdown exposure is also a major concern, as unsold stock must be cleared at a discount, eating into profit margins.

These complexities make it difficult to predict where money will actually go, and which reductions will survive contact with production. Many companies rely on traditional cost accounting methods, which can be misleading and fail to account for the nuances of the industry. For example, a shift supervisor may check the maintenance log to identify areas for cost savings, but this may not take into account the impact of supplier fragmentation on overall costs.

Unit cost vs total cost

A common mistake in cost optimisation is to focus solely on unit cost, without considering the total cost of an AI workflow. While it may be possible to reduce the cost of a single garment by negotiating a better price with a supplier, this may not necessarily lead to overall cost savings if the supplier is unreliable or the garment requires expensive rework. The total cost of an AI workflow includes not just the cost of the AI itself, but also the cost of data preparation, model training, and deployment.

In contrast, considering the total cost of an AI workflow allows companies to make more informed decisions about where to invest their resources. By taking into account the entire cost structure, companies can identify areas where reductions will have the greatest impact, and make targeted investments in AI and other technologies to drive cost savings.

Real levers for cost optimisation

There are three real levers for cost optimisation in AI workflows: caching, batching, and model routing. Caching involves storing frequently-used data in memory, to reduce the time and cost of retrieving it from storage. Batching involves processing multiple requests together, to reduce the overhead of individual requests. Model routing involves directing requests to the most appropriate model, based on the specific requirements of the request.

These levers can have a significant impact on costs, particularly in industries with high volumes of data and complex supply chains. By implementing caching, batching, and model routing, companies can reduce the cost of their AI workflows and improve overall efficiency. For example, a company using AsscherAi to optimise its supply chain may be able to reduce its costs by implementing caching and batching, and routing requests to the most appropriate model.

Token cost is rarely dominant

Token cost is often cited as a major factor in AI cost optimisation, but in reality it is rarely the dominant line. While token cost can be significant, particularly for large language models, it is often dwarfed by other costs such as data preparation, model training, and deployment. Furthermore, token cost is often a fixed cost, which can be amortised over a large number of requests.

In contrast, other costs such as data preparation and model training can be highly variable, and may depend on factors such as the complexity of the data and the requirements of the model. By focusing on these costs, companies can make more targeted investments in AI and other technologies, and drive greater cost savings. Companies can contact us to learn more about how to optimise their AI workflows and reduce costs.

Limitations of AI cost optimisation

While AI cost optimisation can be a powerful tool for reducing costs and improving efficiency, it is not a panacea. There are many situations in which AI cost optimisation may not be the best choice, such as when the costs are highly variable or uncertain, or when the supply chain is highly complex. In these situations, other approaches such as traditional cost accounting or manual optimisation may be more effective.

Furthermore, AI cost optimisation requires significant expertise and resources, particularly in industries with complex supply chains and high volumes of data. Companies must carefully consider their needs and capabilities before investing in AI cost optimisation, and must be prepared to make significant changes to their workflows and processes in order to realise the benefits. By understanding the limitations of AI cost optimisation, companies can make more informed decisions about how to use this technology to drive cost savings and improve efficiency.