What can we learn from our past mistakes with AI?
Artificial Intelligence (AI) has evolved from a buzzword into an essential tool for businesses seeking dramatic productivity improvements. At Monolith, we've guided diverse engineering firms, from the automotive to the battery industry, on their AI journeys. This blog covers the best practices and improvements we have seen our clients implement in various Machine Learning projects.
Over the past six years, we have developed software ideally suited to these applications. After working with hundreds of engineering teams over the past six years, we’ve found that data strategy is the key element to maximising the impact AI can have on your engineering process.
A robust data strategy is crucial, yet many engineering teams struggle with what it really means and how to get their data together for AI. Poor data management can delay your AI projects by months and keep you from gaining a competitive edge. Companies thriving with AI today are those that addressed these data challenges early on.
Today, we aim to help you sidestep these common pitfalls. Doing so will save you time and money on your projects and ensure you are ready to benefit from AI. We have encountered these issues in various settings, from small labs with 50 engineers to large companies with multiple departments and thousands of engineers. Let’s explore how you can tackle these challenges and ensure the success of your AI adoption from the start.
Your data strategy
An inadequate data strategy can drastically slow down an AI implementation and impact your potential success and efficiency.
To maximise AI's value, consider these three data pillars: availability, governance, and management.
3 pillars to guide your AI data strategy planning
Availability
Most engineering companies have access to a substantial amount of data, but key questions about availability include:
- Do you have the right data – the inputs and operating conditions needed to characterise the performance of your system?
- Do you have enough data to train an accurate model?
These questions are often the most crucial and have the greatest impact on your AI implementation. Many times, our services team will consult directly with the test team to learn more about the product design and collaborate with them on additional inputs or ranges required for accurate modelling.
Governance
People commonly link the term governance to data security, privacy, and access management. While these are vital, another critical, yet often neglected aspect, is the standards by which your data is maintained. Setting high standards for data management can significantly influence your AI projects.
Consider the following:
- Are all your files named clearly and logically?
- Are your file formats consistent and compatible with common machine learning platforms, and coding languages?
- How do you ensure uniformity in your data across different experiments and use cases?
- Are you defining data labels, units, and categorical values consistently across teams, tests, and products?
- Are your time-series data synchronised in terms of sampling rates and aligned start times?
Although these may seem like small issues, over the scale of an organisation these details can hinder the speed at which you can implement new technologies. Addressing these questions up front avoids key bottlenecks and data rework that slow down AI adoption. Even if you are not planning to implement AI immediately, improving these areas now will prepare you to effectively leverage AI in the future.
Management
Knowing what data you have and what you need to acquire can accelerate your AI adoption. If you do not store your data centrally or maintain good records, you will face issues such as:
- Having to rerun tests because you cannot find old data,
- Spending time accumulating data files from different equipment and devices,
- Overlooking additional data that could enhance your machine learning model.
This is why it is important to ask yourself the following questions:
- Are you structuring and storing your data centrally?
- Do you have tools and policies for access?
- Do you know/have a structured approach for where to access your data?
Building your team’s AI expertise
The gradual adoption of AI through an organisation, across which you should be asking yourself the different questions presented in this blog.
Becoming AI-ready is a gradual process built from experience. However, there are often concerns about acquiring AI expertise. Some key objections we have heard include:
- Integrating AI and data science into our engineering processes could be costly and disruptive, with uncertain benefits.
- Our engineers do not have sufficient familiarity with the required coding/Python skillset to apply AI.
- How will we ensure the security of our sensitive data from exposure to the market, or competitors?
The first step to your data strategy: Building trust while realising ROI
Viewing the introduction of machine learning into your workflows as a risk is understandable. We have worked with clients with various risk tolerances and have discovered multiple ways to initiate machine learning projects that suit everyone’s risk level.
A practical first step we often recommend is developing a Minimum Viable Product (MVP) using historical, non-sensitive data. This allows engineers to explore the benefits of machine learning with minimal risk. By using data that does not impact current operations, teams can experiment and learn without the pressure of immediate stakes nor looming schedule deadlines.
As confidence builds, we advise gradually incorporating a smaller subset of current data. This method allows teams to assess how the system performs under more typical conditions, helping to bridge the gap between theoretical benefits and real-world application.
Structuring your data integration
Data integration is complex. Even a "simple" experiment can utilise an array of equipment and generate tens of thousands of data points. The equipment may include thermal chambers, temperature sensors, battery cyclers, and more. Scaling that to a larger test lab where there are many cells, modules, and packs being tested – the sheer volume of information clearly shows the need for strong data analysis and integration.
The main things when looking for a good integration strategy are:
- Repeatable solutions
- Integration with existing tools and processes
- Trusting that the process will not be broken/slowed down
Accelerating your AI adoption
In this blog, we have highlighted several critical questions to consider when planning your machine learning implementation. Finding answers can sometimes seem like hitting at a moving target. Engaging with industry professionals can help you avoid common pitfalls, learn from others' experiences, and successfully implement your projects.
At Monolith, we bring six years of expertise in AI for engineering, developing machine learning models and solutions with your specific needs in mind.
We are working with some of the best engineering teams in the world, seeing firsthand how leaders apply AI across their organisations and processes. We can help you understand where your team is and how to help you move them forward in AI maturity and impact. AI is here to stay – and it’s important to get started now, before you fall behind the competition.
For a deeper dive, watch our comprehensive webinar featuring automotive experts discussing optimal data strategy approaches.
