How to start with cost optimisation
Getting started with AI cost optimisation involves choosing a simple problem with a clear answer, like a shift supervisor checking the maintenance log to identify frequent repairs, then using tools like AsscherAi to analyse and compare costs, and finally applying real levers like caching and batching to reduce costs.
The first thirty days of a cost optimisation project can be make or break, and I have seen many organisations struggle to get started, often because they try to tackle a problem that is too complex, with too many unknowns, and no clear way to measure success. A project that starts with a clear goal, a known answer to compare against, and a simple way to check progress, is much more likely to succeed. This is why choosing a first case that already has a known answer to compare against is crucial.
Choosing a first case
When getting started with cost optimisation, it is tempting to try to tackle the biggest, most complex problems first, but this is often a mistake. Instead, choose a simple problem, with a clear answer, that can be used as a baseline to compare against. For example, a shift supervisor checks the maintenance log, and notices that a particular machine is always being repaired at the same time every week. This could be a good first case, because the current cost of repairs is known, and any changes can be easily measured against this baseline.
A good first case should also be one where the data is already available, and can be easily accessed and analysed, using tools like AsscherAi, which can answer questions about an organisation's own data in real time.
Unit cost and total cost
One of the biggest mistakes people make when trying to optimise costs, is to focus solely on unit cost, without considering the total cost of an AI workflow. Unit cost is the cost of a single unit of something, like the cost of a single repair, but total cost is the cost of the entire workflow, including all the repairs, and all the other associated costs. For example, a category manager compares two suppliers, and chooses the one with the lowest unit cost, but fails to consider the total cost of shipping, storage, and handling, which may be much higher for the cheaper supplier.
This mistake can lead to decisions that actually increase costs, rather than reducing them, which is why it is so important to consider both unit cost and total cost, when evaluating different options.
Real levers for cost optimisation
When it comes to cost optimisation, there are only a few real levers that can be used to reduce costs, and these are caching, batching, and model routing. Caching involves storing frequently used data in a way that makes it quickly accessible, which can reduce the time and cost of retrieving the data. Batching involves grouping similar tasks together, and performing them in batches, which can reduce the overhead of performing each task individually. Model routing involves choosing the most efficient model to use for a particular task, based on the specific requirements of the task.
For example, a data scientist uses AsscherAi to analyse customer purchase history, and notices that the same data is being retrieved multiple times, which is slowing down the analysis. By caching the data, the data scientist can speed up the analysis, and reduce the cost of retrieving the data.
What to stop doing
If the comparison between the baseline and the optimised workflow fails to show any cost savings, then it is time to stop and re-evaluate the approach. This may involve stopping some activities altogether, if they are not providing any value, or changing the way that certain tasks are performed, to make them more efficient.
It is also important to be willing to challenge assumptions, and to be open to new ideas and approaches, rather than just continuing to do things the same way, because that is the way they have always been done. If you need help with this, you can contact us for more information.
When cost optimisation is not the answer
Cost optimisation is not always the answer, and there are some cases where it is the wrong choice. For example, if the goal is to improve customer satisfaction, or to increase revenue, then cost optimisation may not be the best approach. In these cases, it may be more important to focus on other factors, such as quality, or speed, rather than just cost.
Additionally, cost optimisation may not be the best approach if the costs are already very low, or if the potential savings are not significant enough to justify the effort. In these cases, it may be better to focus on other areas, where the potential savings are greater. You can learn more about how to get started with cost optimisation, and how to determine whether it is the right approach for your organisation.
Choose a simple problem with a clear answer and known baseline to compare against, like a frequent repair issue. Focusing solely on unit cost can lead to decisions that increase overall costs, like choosing a cheaper supplier with higher shipping costs. Caching, batching, and model routing are the key levers to reduce costs, by storing frequently used data, grouping similar tasks, and choosing efficient models. If the comparison between the baseline and optimised workflow fails to show any cost savings, it's time to stop and re-evaluate, and be willing to challenge assumptions and try new approaches.Frequently asked questions
What is the best way to get started with AI cost optimisation?
Why is it important to consider total cost, not just unit cost?
What are the real levers for cost optimisation in AI workflows?
When should I stop trying to optimise costs and re-evaluate my approach?