Walk before you run: How to get Gen AI working for your business

Brendan Willkinolls – Head of Machine Learning at Mantel Group

The power of Generative AI is real. I know, you’re tired of hearing about it.

But the fact is, the excitement around Gen AI is justified. As the technology matures, it will have a profound effect on the way products and services are delivered.

However, before setting out on your Gen AI journey, companies need to ensure the fundamentals for implementing AI are in place. This means your AI tech stack needs to establish foundations across people, process and governance in order to be successful.

To help cut through the noise, here are some useful steps that will help build the solid Gen AI foundations required to set off on the right foot. For more information on Gen AI, download our white paper Walk before you run: Getting smart with Gen AI

Identify if Gen AI is useful for your business

From the start, a business needs to carefully consider what unique Gen AI use cases are relevant and useful within your particular company setting, while also being critical about how (or, indeed, if) it can be put into production. This is not a time to be seduced by hundreds of pre-defined hypothetical use cases – the path to value comes from your unique requirements and how to leverage the strengths of Gen AI.

A business will also need to wrap its head around terms that are far from sexy, such as Machine Learning Operations. In fact, we believe that the key to successful Gen AI models come through successful machine learning operations.

Building effective operational value in machine learning will generate a deep level of trust with your executives, your customers and/or your team, to ultimately deliver a successful Gen AI program.

Gen AI is not an end-to-end system. Nor is it a case of simply utilising a publically available Gen AI API service, such as GPT 4, and pushing it onto existing processes. Rather, Gen AI requires time and effort from varied groups across your business to ensure that the planned environment is going to work in the real world—for your technical teams, business stakeholders and customers—without incurring extra effort or costs.

Get Gen AI fit – hire a personal trainer

The AI world is accelerating at a faster pace than many have anticipated. Unlike most other technological advancements where one part of the process lags another, the technology, research, people and platforms underpinning Gen AI are all rapidly developing at the same time. This means that by the time you are up to date with the lingo, there might be a whole set of new things to learn.

Approach Gen AI like a regular gym workout. If you only go once every few months, it’s going to hurt, but if you go a few times a week, it becomes second nature and part of your routine. Ideally, what will really get you to the next level is hiring a personal trainer. Most organisations will need the help of a dedicated delivery partner to balance risk and reward and keep you up to date and motivated.

The key component to success in Gen AI, just as it is for machine learning models, is confidence. A business needs to have confidence that the outputs being generated are the correct outputs, and that your use of AI is doing the right thing. There also needs to be a level of confidence with your executive team that your AI implementation won’t pose any risk to the business, and only offers value.

MLOps is the catalyst for delivering Gen AI

Machine Learning Operations (MLOps) is how you deliver end-to-end machine learning in a productive way. That might sound technical (and it is!), but if you’re thinking about starting a Gen AI program, you need to start with an MLOps program that is absolutely humming.

An MLOps program means you’ve done the work to make your ML projects functionable, reliable, repeatable and with a value-based mindset. A structured MLOps program is one that takes your organisation’s idea through the ML lifecycle starting with data engineering, to feature engineering, to ML experimentation, training & model build, and finally to deployment.

If you’ve done this work, there’s a far greater chance that you’ll be able to avoid the pitfalls of scaling up to a Gen AI program. You will also have a much clearer view of the benefits that your data can deliver. Again, if this still feels overwhelming, having a technology partner to ‘train’ you and your tech teams is essential.

Get your governance in place

Once you’ve got your ML foundations in place, how do you make the step from MLOps to Gen AI? It all starts with something that’s hardly sexy but is absolutely necessary: governance.

Governance is possibly the most important part of a Gen AI program, yet it is regularly overlooked. When you turn on your Gen AI model and it starts generating answers, how do you know that those answers are what you want or need? Do the answers have repeatability, explainability, and reliability? These are all questions that good governance oversight allows you to answer.

Baking in responsible AI at the beginning of your journey not only gives you internal confidence in the model’s performance, it will lead to better outcomes and value realisation. These outcomes will be enabled via a range of factors, from prompt validation, to prompt guardrails, monitoring, access control and data governance.

But governance shouldn’t be set up as a series of insurmountable roadblocks. Businesses need to iterate quickly and innovate as they learn new information about this new technology, so ensure your governance processes allow for some sensible flexibility.

We don’t need to tell you that there’s a lot of hype around Gen AI. As the technology continues to improve, you need to carefully sift through the options and understand exactly how its current maturity could apply to your use cases. Don’t be swept along by the new, shiny toys, or the need to immediately fix the things that aren’t broken. Starting with strong technological foundations and supportive partners is the best place to start.