AI is not the route to great products – and here’s why

These are exciting times for the research industry. Advances in tech can make a tangible difference to the quality and depth of insight. But it can also generate misguided and misleading outputs – something that no research professional would want to be associated with.

Recently, our concerns about the reliability of facial coding as a means of predicting emotions were vindicated by a new report issued by the UK Government’s Information Commissioner’s Office (ICO). It stated that these ‘emotional analysis’ techniques are based on a ‘false premise’ and should not be relied on in circumstances where the truth really matters.

With this in mind, we turn our attention to so called ‘universal predictive models’. These draw upon learnings from previous tests with products across different markets and categories, spanning different consumer targets, and claim to be able to predict how much a new product will be liked by consumers based on the sensory experience reported by a limited number of trained/semi-trained consumers.

Unsurprisingly, predictive models are gaining a lot of interest because they could – in theory - reduce the time and investment required in consumer research. They claim to improve the chances of success by identifying, via ‘AI modelling’, innovation and optimization opportunities beyond the products being tested. A mighty bold claim, if ever we saw one!

But, as much as we love tech, we are unable to endorse such rhetoric because such models are excessively reductive. The reality is that we live in an age of great products, so if you’re seeking to make improvements, we’re often talking about very small tweaks. For entirely new products, simply creating something that is liked is now an outmoded ambition. Increasingly, brand owners are shooting for distinctiveness, heightened sensory theatre - bigger, bolder, and brighter product signatures that build memory structures with the power to displace existing repertoires. Add to this the complexity associated with renovation and cost reduction with no loss of appeal, and you’ll quickly realize that your testing approach demands detail, nuance, and granularity!

Whilst predictive modelling often appears to deliver results that look sensible and actionable, it simply cannot deal with the nuance and subtlety needed to make ‘successful products’ in the modern world.

To create a winning product, you must first understand people’s expectations of your brand and category, then understand in detail the sensory space of your particular category and the preferences of your target consumers. We cannot endorse universal models to get you there. They invariably ‘chase averages’ and will lead to a continuation of mediocre, in a world that craves meaningfully distinctive experiences.

May we offer these further observations to help you plan your next move:

• Your target population will exhibit a range of sensory preferences that are rarely dictated (or predicted by) who they are or where they live. Failure to identify or appreciate the existence of these preference segments risks development of products that will satisfy many, but delight few.
• Consumer preferences are driven by the entire product experience – its appearance, sound, aroma, taste and texture. Predicting performance necessitates analysis of all these modalities and their interactions, and how these interactions are interpreted according to the nuanced expectations people will have across different categories.
• Predicting success involves much more than liking. Brands need to motivate consumers with a combination of immediate pleasure (liking) and emotional reward, which ultimately dictates longer-term adoption.
• Making improvements can be about reengineering the last 5% of your products delivery. To achieve this requires a detailed understanding of how products within the category differ, or could differ, in the smallest of ways.
• Finally, consider this: if all brands pursue the same ideal, then the likelihood is that competing manufacturers will end up in the same sensory space - making it hard to differentiate and build meaningful distinctiveness to hook people in.

In spite of our concerns, we can see a way forward for AI in identifying white space opportunities - such as which flavor combinations might work well in a new market. Even in these situations however, relying too heavily on AI risks that such suggestions are based on hypothetical combinations which will be untested.

We will continue with our quest to offer solutions that support faster NPD and speed to market, and whilst we understand the appeal of AI driven predictive models, as things stand today we strongly believe that well designed research, delivered by genuine product, sensory and branding experts, is still the best way to fast-track your NPD whilst ensuring the product really hits the spot for consumers.

This is where our co-creation approaches and sensory drivers toolkit excel, allowing us to identify the optimal profile for a target consumer or optimization opportunities, which might seem small or counter intuitive, but can make a big in-market difference. Get in touch to learn about what we believe it takes to truly understand sensory drivers in order to develop products that delight.

You can also access our resource on Sensory AI drivers below:

Be first to read our assessments on the latest tech by signing up to The NOVA Partnership, by MMR below.