• Written by Muhammed Abdulla NC, CEO

A category manager at a large FMCG company is deciding this week how to allocate production resources to meet the expected surge in demand for a new product launch, and they are currently basing it on historical sales data and intuition, but this approach often leads to costly mistakes, such as overproduction and waste, or underproduction and lost sales, which can be mitigated by using AI cost optimisation tools to analyse the company's own data and provide more accurate forecasts.

FMCG complexities

The FMCG industry poses unique challenges for cost optimisation, particularly due to promotional distortion in historical sales data, which can make it difficult to accurately forecast demand, and shelf life write-offs, which can result in significant losses if not managed properly, and distributor visibility, which can be limited, making it hard to track inventory levels and shipments, all of which can lead to inefficiencies and waste in the supply chain.

For instance, a shift supervisor at a manufacturing plant may struggle to optimize production schedules due to the lack of visibility into distributor inventory levels, leading to overproduction or underproduction, and the category manager may find it challenging to allocate resources effectively due to the distorted historical sales data, which can be addressed by using AI cost optimisation tools to analyse the company's own data and provide more accurate forecasts, as can be seen on our website.

Unit cost vs total cost

A common mistake in cost optimisation is focusing solely on unit cost, without considering the total cost of an AI workflow, which includes not only the cost of the AI model itself, but also the cost of data preparation, model deployment, and maintenance, and the cost of integrating the AI model with existing systems and processes, and the potential costs of errors or inaccuracies in the forecasts, which can be significant if not properly managed.

The total cost of an AI workflow can be substantially higher than the unit cost, and companies that fail to consider this may end up with a cost optimisation strategy that is not effective in the long run, and may even lead to increased costs and decreased efficiency, which is why it is essential to consider the total cost of an AI workflow when developing a cost optimisation strategy, and to consult with experts, such as those on our team, to ensure that the strategy is effective and sustainable.

Real levers for cost optimisation

There are three real levers for cost optimisation in AI workflows: caching, batching, and model routing, which can be used to reduce the total cost of an AI workflow, and improve its efficiency and effectiveness, and these levers can be used in combination to achieve even greater cost savings, and improved performance, and can be applied to various aspects of the supply chain, from production to distribution.

Caching involves storing the results of expensive computations so that they can be reused instead of recomputed, batching involves grouping multiple requests together to reduce the overhead of processing individual requests, and model routing involves directing requests to the most appropriate model or system to handle them, and these techniques can be used to improve the efficiency and effectiveness of AI workflows, and reduce their total cost, and can be implemented using various tools and technologies, such as those provided by AsscherAi.

Token cost is rarely dominant

Token cost, which refers to the cost of processing individual units of data, is rarely the dominant line in the total cost of an AI workflow, and companies that focus solely on reducing token cost may miss opportunities to reduce the total cost of the workflow, and improve its efficiency and effectiveness, and may even end up increasing the total cost, and decreasing the efficiency and effectiveness of the workflow, which is why it is essential to consider the total cost of an AI workflow, and not just the token cost, when developing a cost optimisation strategy.

Other costs, such as the cost of data preparation, model deployment, and maintenance, and the cost of integrating the AI model with existing systems and processes, can be substantially higher than the token cost, and companies that fail to consider these costs may end up with a cost optimisation strategy that is not effective in the long run, and may even lead to increased costs and decreased efficiency, which is why it is essential to consider the total cost of an AI workflow, and not just the token cost, when developing a cost optimisation strategy.

Limitations of AI cost optimisation

AI cost optimisation is not a silver bullet, and it is not suitable for all companies or all situations, and there are limitations to its use, and potential risks and challenges that must be considered, and companies that are considering using AI cost optimisation must carefully evaluate their needs and goals, and consider the potential benefits and risks, and develop a strategy that is tailored to their specific situation, and that takes into account the unique challenges and complexities of their industry and operations.

For instance, companies that have limited data or limited resources may find it challenging to implement AI cost optimisation effectively, and companies that have complex supply chains or multiple stakeholders may need to develop customized solutions that take into account their specific needs and challenges, and companies that are considering using AI cost optimisation must be aware of these limitations, and potential risks and challenges, and must develop a strategy that is tailored to their specific situation, and that takes into account the unique challenges and complexities of their industry and operations.