Could AI have role in testing content? Hannah Johnson explains how it’s now possible to generate results as valid as those from focus groups and surveys using synthetic personas.
The most effective marketing pulls on people’s emotions – whether that’s desire, hope or something else entirely – and drives them to take action. But while on the face of it, great, impactful marketing may not look like the result of a scientific process, a data-led approach can help marketers refine messaging, giving the end result far bigger impact and a lot more bang for the buck.
Testing content is nothing new. Relying on ‘gut feel’ isn’t a credible method for most brands and their agencies – they want to know, particularly when a significant amount of money is being invested in a campaign, that the work is going to resonate. They want to be able to predict, with a reasonable amount of certainty, how their messages will be received and interpreted.
However, testing content isn’t without challenges. People are prone to bias; we know that in focus groups, for example, there is often a difference between what people say and what they actually do. And from the brand’s side, content being tested isn’t yet refined and could be in danger of being taken out of context by potential customers. A brand may also have a requirement to test sensitive or highly confidential messaging, or gather feedback on a new positioning without revealing too much.
We wanted to know if there was a way to understand what content would resonate with people without physically putting it in front of them, and without the use of paid media. Artificial intelligence (AI) and machine learning had the answer.
Research showed us that we could use AI or machine learning-based synthetic customer personas to essentially create a proxy representation of customer interests, with the technology helping us to model existing known customer behaviours and preferences. To develop these simulated customer personas, we worked with one provider called Tanjo, a technology platform that offers advanced customer modelling and automation that sits at the intersection of machine learning and behavioural science.
The customer personas are created using GDPR-compliant data, from publicly available sources like economic purchase data, national statistical and demographic data or even an organisation’s own customer data. This is then used to create representations of real audience segments, but without being tied to any one person’s data, and therefore not invading privacy. We saw it had potential for brands to explore opportunities within audience insight, by essentially acting as a proxy for human focus groups, offering message testing and uncovering new content opportunities without requiring paid media spend.
Synthetic customer personas are based on the hypothesis that people’s personalities and unique interests can predict how they will react to content. By simulating these interests and conducting testing with them, you can generate results that can sometimes be as valid as, or very close to, those from focus groups and surveys. When budget or time is a constraint, this can be performed for far less cost, staff time and effort than qualitative research with real people.
Having said this, using synthetic personas doesn’t mean marketers can just sit back and put their feet up. As with most things in life, the more you put in, the more you’ll get out. More information means better, more representative customer personas, especially when using open-source data where the health or age of the set isn’t always known. Of course, organisations with a wealth of first-party customer data will see the most benefits and the most representative, useful outputs.
When maintained, the customer personas learn and evolve continually over time. We’ve found that implementing a quarterly programme of data ingestion, including adding more open-source data as well as our own findings from tools like social listening, helps the personas and their responses to develop and become more robust. Our synthetic customer personas become more and more representative of the customer segments we are trying to model, and are responding to content in ways that replicate our own human customer segments.
It’s also important to note that audiences evolve quickly. In the real world, people’s habits, interests and motivations change over time. Trends rise and fall away again. This is another reason why regular data ingestion is essential; there’s little point testing content on outdated customer models. And like audiences, the platforms and technologies also develop quickly. Customer behavioural modelling is one small facet of the huge spectrum of use cases of AI and machine learning for marketers.
We also see huge promise in the areas of image and video recognition, sentiment analysis and personalised customer care. This space moves and evolves rapidly and Tanjo is now one of many providers, joined by tech companies such as Quid, Concured and Codec across a range of different applications including trend prediction, audience segmentation and content production.
As with any new technology, many brands will no doubt be wary about making significant business decisions based on the outputs of AI or machine learning applications. Despite being low cost and low risk compared to traditional methods and providing a live, dynamic picture, it’s not the easiest tool to explain to a CEO when justifying its slice of the marketing budget. Particularly as it needs a long-term commitment, if you want to see truly useful results.
But I could see brands of all shapes and sizes adding the technology to their toolkit – perhaps when the perception of AI and machine learning starts to become less futuristic and more commonplace within organisations. Based on the current rate of progress, this may not be too far away.