• Written by Muhammed Abdulla NC, CEO

A common objection to cost optimisation for research and development is that it will stifle innovation, but this is a misguided concern, as the real issue is not the amount of money being spent, but where it is actually going, and which reductions will survive contact with production. In reality, cost optimisation can help researchers and developers focus on the most critical aspects of their work, leading to more efficient use of resources and better outcomes.

The difference between unit cost and total cost of an AI workflow

When it comes to AI cost optimisation for research and development, it's essential to understand the difference between unit cost and total cost of an AI workflow. Unit cost refers to the cost of a single iteration or run of the workflow, while total cost takes into account the cumulative cost of all iterations, including re-runs, failures, and maintenance. Most people get wrong that unit cost is the only factor to consider, but this is wrong, as total cost is often the more significant factor in the long run.

A shift supervisor checking the maintenance log can see that while the unit cost of a single AI model run may be low, the total cost of re-running the model multiple times due to errors or changes in requirements can add up quickly. This is where cost optimisation comes in, helping to identify areas where costs can be reduced without compromising the quality of the research or development.

Caching, batching and model routing as the three real levers

Caching, batching, and model routing are the three real levers that can be used to optimise AI workflows for research and development. Caching involves storing frequently-used data or results to avoid redundant calculations, while batching involves grouping similar tasks together to reduce overhead. Model routing, on the other hand, involves directing tasks to the most suitable model or resource, taking into account factors such as computational power and memory usage.

By applying these levers, researchers and developers can significantly reduce the total cost of their AI workflows, without compromising on quality or accuracy. For example, a category manager comparing two suppliers can use caching to store the results of previous comparisons, and batching to group similar comparisons together, reducing the overall cost of the process.

The research and development data this depends on

The data required for AI cost optimisation for research and development includes test results and failure analyses, design review records, timesheets and allocation, and supplier assessments. This data helps to identify areas where costs can be reduced, and provides a basis for making informed decisions about resource allocation. By analysing this data, researchers and developers can identify patterns and trends that may not be immediately apparent, and make adjustments to their workflows accordingly.

For instance, a team leader reviewing the design review records can identify areas where the design process can be improved, reducing the likelihood of errors and re-work, and thereby reducing costs. By linking this data to the research and development process, teams can create a more efficient and effective workflow.

What it changes about failures rediscovered across programmes and effort invisible against project stage

AI cost optimisation for research and development can help to reduce the incidence of failures rediscovered across programmes, by identifying and addressing common issues and areas of inefficiency. It can also help to make effort more visible against project stage, by providing a clear picture of where resources are being used, and where they can be optimised.

This is particularly important in research and development, where the cost of failure can be high, and the benefits of success can be significant. By optimising AI workflows, teams can reduce the risk of failure, and increase the likelihood of success, leading to better outcomes and more efficient use of resources.

Measured with failures found before validation, effort against stage, estimate revision count

The effectiveness of AI cost optimisation for research and development can be measured using a range of metrics, including failures found before validation, effort against stage, and estimate revision count. These metrics provide a clear picture of where costs are being reduced, and where efficiencies are being gained.

By tracking these metrics, teams can see the tangible benefits of AI cost optimisation, and make adjustments to their workflows accordingly. For example, a project manager reviewing the estimate revision count can see where changes to the workflow have led to reductions in costs, and where further optimisation is needed. To discuss how to implement AI cost optimisation for your research and development team, contact us today.

What this does not do

AI cost optimisation for research and development is not a silver bullet, and it's not a replacement for good management practices. It's a tool that can be used to identify areas for improvement, and to optimise workflows, but it requires careful implementation and ongoing monitoring to be effective.

It's also not a one-size-fits-all solution, and what works for one team or organisation may not work for another. Each team must carefully consider their specific needs and requirements, and tailor their approach to AI cost optimisation accordingly. By doing so, they can unlock the full potential of AI cost optimisation, and achieve better outcomes for their research and development efforts.