Skip to main content

Driving AI innovation in product design

Everything starts with delivering customer value

We are a month into 2024 and there are no signs that interest in generative AI is relenting. Much of the focus in 2023 was on generative AI models, but in 2024, this has shifted to realising tangible commercial value from generative AI.

One of the most important drivers of commercial value is adoption, and designing, building and integrating effective generative AI products that make the underlying models accessible and useful for employees.

To achieve this, we not only need to build effective products for users, we also need to design our products to collect rich feedback from users. This helps improve product design, improve the underlying generative AI models, and identify new areas for deployment within businesses.

This approach has been critical to our generative AI development both internally and for our clients, and is key for any organisation looking to achieve a return on investment (ROI).


It’s all about product

The translation of new technologies into commercial value is nothing new, and ultimately boils down to one thing: product. Through effective product design and build, we can transform new technologies such as generative AI into accessible, usable and useful tools, and this in turn enables individuals and organisations to complete tasks, solve problems, and drive value.

Of course, products are not new to generative AI, and have been at the centre of its rapid growth since late 2022. Some of the most well-known names in generative AI are products – not models – and their design has undoubtedly been a key part of the success of the underlying transformer and diffusion technologies.


As a society, we’re pretty good at building products

Effective product design will be key to the successful adoption of generative AI-based tools, and over the last few decades, we’ve got much better as a society at building high quality technical products, including those that incorporate AI.

This improvement has been driven by a variety of factors, including improvements to design and build via approaches such as agile, lean, scrum, Development Operations (DevOps) and Machine Learning Operations (MLOps). This has been mirrored by improvements to product development tools, including cloud infrastructure, collaboration software, version control, and new technical languages. These advances are all relevant to generative AI, and will help unlock value via robust, intuitive and scalable generative AI products.


It’s a three-way street

This is only part of the story. Products will not only help bridge the gap between existing generative AI models and commercial value, but also help improve the underlying models, and enhance the product development process.

Starting with the latter, generative AI tools are being used across the entire product lifecycle, from idea generation to technical design, build, and user testing. This has created a virtuous cycle where generative AI products enhance technical product development, which in turn helps create better generative AI products.

What’s more, generative AI products are also improving the underlying AI models. User interactions with frontend applications create rich datasets that not only contain feedback on frontend design, but also on the performance of backend models. For example, if a user repeatedly asks the same question with slightly different phrasing, this may suggest that a model is failing to provide a satisfactory answer. Conversely, if a user copies text out of an application, this may suggest that the generative AI output is useful.

These implicit behaviours can be combined with more explicit feedback, such as via “thumbs up” and “thumbs down” buttons, product ratings and reviews, and more traditional user surveys. Collectively, this feedback creates labelled datasets that can further train and finetune the underlying models, and signal to model and product developers potential new areas that can be augmented and automated using generative AI.


Garbage in, garbage out

With interactions recycled back into model training and product design, designers and developers need to ensure that these interactions produce reliable, useful, interpretable and contextually relevant insights.

One way that we are doing this in Deloitte is through our own in-house generative AI platform called PairD. PairD has allowed us both to get generative AI quickly and safely into the hands of our colleagues, and to learn more about how generative AI is used within commercial organisations. By building PairD in-house, we are able to collect and analyse user data via custom user interfaces, and then feed these insights back into product design, customisation of underlying models, and greater understanding of potential new areas within the firm that can be augmented with generative AI.


It's all about learning

This all points to the importance of on-going learning, and building an approach to generative AI that enables rapid and effective feedback.

Products are undoubtedly at the heart of delivering commercial value using generative AI, and effective learning is the key to unlocking this value. By focusing relentlessly on delivering customer value and designing products that gather feedback and measure success, we can continuously improve our products to drive commercial outcomes and ultimately achieve a positive ROI.

If you would like to learn more about effective product design, including ways in which Deloitte can help your organisation design, build and integrate effective generative AI products, please get in touch.