Introducing our new neural network platform – EmNet
As I have a background in neural networks (I completed my PhD at the Neural Systems Engineering Lab at Imperical College London), I have for some time known that neural networks could be used to make sense of consumer data and give rise to predictions and insights that have been so far unavailable.
Artificial neural networks and implicit reaction time tests are based on the very same assumptions about how the brain processes information – that many of these processes are associative. For consumer scientists this means that perception of a product or service is a composite of every- thing that we have come to associate with it, from adverts to word-of-mouth, to direct experience with consuming the product or using the service, and that such memories are activated in the brain, though most often do not reach conscious registration. It is not a coincidence that the same principles that underlie deep neural networks are the same that underlie implicit response testing.
At Split Second Research we have created EmNet a neural network platform that enables us to dive deeper into market research data. The platform enables us to understand purchase drivers more deeply by mapping brand values (in the form of reaction times or explicit choices) to behavioural ‘outcomes’ towards the brand, (such as purchase intent, willingness to recommend, and so on) as well as explore the relationships between brand values in order to understand different kinds of purchasing decisions.
We simply upload the responses from a survey into the platform, make a few selections based on the kind of network we want to produce, and within just a minute or so it produces outputs that show relations between the concepts we are testing in the survey and salient features in the data. We are currently offering this service free with every survey we are commissioned to undertake. We also offer this service to other agencies who have data they want to analyse using a machine learning approach.
We already have some interesting case studies which we have put together with a description of EmNet (with a brief introduction to neural networks). This six-page report can be obtained on request via the contact form.
Split Second Research Scoops Up Two David Ogilvy Awards
Split Second Research, with Edelman Intelligence and Hewlett Packard, took two awards at the ARF David Ogilvy Awards, September 2019.
We won GOLD in the category TECH & COMMUNICATIONS (beating Twitter and Samsung), and SILVER in the category DATA INNOVATION.
The aim of the marketing campaign was to reinforce HP’s leadership in diversity and inclusion. The goal of the research was to understand what Americans from all walks of life associate at conscious and unconscious levels with “The All American Family” . HP wanted to raise awareness that printed photographs have a powerful impact and are still relevant in today’s digital world.
With Edelman Intelligence, Split Second Research designed an implicit test with images of families that varied by their ethnicity, sexual orientation, presence of children, and number of parents. It assessed the extent to which an unconscious bias existed in their associations with what represents and “All American Family” and what does not.
The key finding was that “75% of respondents picture an all-American family as a white mom, dad and kids. However, census data tells us that in reality, only 25% of families match that portrait.”
The research inspired a marketing campaign that has helped HP achieve double digit growth in the quarter immediately following the campaign.
Read the case study here.
How does an implicit reaction time test work?
The commercial test is itself based on the evaluative priming paradigm in academic research (e.g., Fazio, et al., 1986)1. The first phase of the test is to detect target emotion words as belonging to either one category (e.g., Happy) or another (e.g., Sad). On each trial in this first phase, the word appears and the respondent has to press one key for happy words and another key for sad words. This is a very easy task, and respondents can do this very quickly with few errors.
In the second phase, the task is the same but the target emotion words are preceded very briefly by ‘primes’. These primes are either congruent with the target word (the prime is Joy when the target is Happy, or the prime is Gloomy when the target is Sad) or incongruent (the prime is Gloomy when the target is Happy, or the prime is Joy when the target is Sad). So on some trials the respondent might see the word Joy flash on the screen followed by Happy. They respond to the second word by pressing a key for happy words. The task can be performed quicker and with fewer errors in this case because the prime and the target are congruent. On other trials, the word Gloomy is flashed on the screen followed by Happy. This time, because Gloomy and Happy are incongruent, respondents are a little slower to detect Happy in their key press. The response may be slower by only a split second, yet when repeated over several trials and over many respondents, it would become ‘statistically significant’.
So let’s take this test into the commercial arena. The primes might be the same (Joy and Gloomy) but this time the targets could be Adidas and Nike. Respondents who have a strong preference for Adidas over Nike are likely to be quicker to respond to Joy > Adidas than Joy > Nike, and quicker to respond to Joy > Adidas than Gloomy > Adidas.
In implicit response testing, we broaden the primes to refelct attributes related to brand equity, such as Trusted, Modern, Friendly, Cool, and so on. The targets can be more than just brands, and can be pack designs, celebrities, claims, and so on.
In fact, there is not only one type of implicit test but several, with each being designed to measure a specific aspect of consumer evaluation, such as ad testing, pack testing, claims testing, NPD testing, celebrity endorsement testing, as well as brand positioning.
1Fazio, R. H., Sanbonmatsu, D. M., Powell, M. C., & Kardes, F. R. (1986). On the automatic activation of attitudes. Journal of Personality and Social Psychology, 50, 229–238. doi:10.1037/0022-3518.104.22.168
Price promotions: More than you bargained for?
Dr Eamon Fulcher, CEO & Co-founder, Split Second Research.
Everyone, so it seems, loves a bargain. But all bargains may not be equal. A price discount on a single item may be perceived differently from a unit offer, such as buy one get one free, and differently from a free gift with every purchase kind of offer. Just which kind of offer is best for a particular brand or market category is the subject of ongoing research from Split Second Research. Although offers can clearly push sales, the fear is that they will adversely affect perception of the brand, as the consumer comes to question aspects such as its quality, reliability, trustworthiness, premiumness, and other attributes.
One way to avoid this is to give a well-accepted motivation for the offer, such as the end of the season, a few days after a festive period, when signing up for a store card, giving discount for purchasing online, and so on. One interesting question is whether promotional offers on luxury goods has any negative impact on the brands. The idea of say, a price discount seems incompatible with the notion of premium or luxury. So for example, the common promotional mechanisms of ‘half price’ and ‘buy-one-get-one-free’ would seem unsuited to brands such as Rolex, Calvin Klein, and so on. It would make common sense that widening the availability of luxury goods through such offers could devalue its brand equity.
However, hard research on this issue is difficult to find, and the little research that has addressed this issue has used traditional survey methods, such as asking consumers whether they would buy certain products under specific promotions. Inviting respondents to offer their own verbalised opinions may not provide us with the full picture because of the several biases inherent in the subjective approach in traditional methods. This may therefore not help us fully understand how they behave in-store. Market research is moving in the direction of using objective methods as an addition to the traditional methods in measuring emotional and non-conscious responses to brand propositions to increase the odds of predicting purchasing intention and behaviour.
Split Second Research recently carried out a study to examine the relationship between a price promotion and premiumness. The focus of interest was in whether the impact of a promotion on brand perception is the same for a premium product as it is for a basic or standard product; also, whether such promotions affect loyal consumers differently to considerers of the brand (those who have not bought but would consider buying) and lapsed customers of the brand (those who have bought but would no longer intend to buy).
Luxury signals at least one of these: (1) social status – a luxury product is one to be seen with, (2) high quality – a product made with expensive materials or designed by a renowned expert, (3) rarity – a product is made exclusively or is naturally in short supply, (4) emotional engagement – the product gives pleasure and satisfies some aspect of hedonism, (5) aspiration – by acquiring something premium we get a taste for luxury and social status, and aspire to getting more, and of course (6) price – luxury items can be several multiples of the price of a basic or standard item in the category.
Putting a luxury item on offer may affect some or all of these, such as reducing its desirability, raising concerns about its quality (a ‘bad batch’), and reducing its sense of exclusivity. For example, research by Zoellner and Schaefers (2015) shows that luxury goods can be negatively impacted by a price promotion (though they can be resilient to promotional mechanisms involving free gifts and loyalty program benefits).
Some brand categories are exclusively premium but many brands offer their products in both premium and basic forms. In this study we chose toilet tissue as it is one such product category. We also chose it because the luxury range is within the reach of most consumers and therefore allows us to draw comparisons between the two. A quick search for “Toilet tissue” on Tesco online makes one realise what a fiercely competitive category it is.
A search on 9th June 2019 returned six brands (Cushelle, Andrex, Sofcell, Renova, Tesco Spring Force, and Tesco Luxury Soft), seven colours (white, cream, pink, peach, light green, and even red or blue), six sizes (4, 6, 9, 12, 16, or 24 rolls in a pack), rolls of different sizes (160, 180, or 200 sheets per roll, different sheets sizes (120mm x 103mm versus 124mm x 103mm versus 118mm x 104.5mm, etc), and with different formulations (2 ply tissue, quilted, special cleaning extracts, such as, aloe vera, chamomile, and shea butter). Some were even 100% recycled (Husted, Russo & Meza, 2014 found that shoppers may buy more for environmentally friendly products, including toilet tissue). Out-of-pocket ranged from £1.00 to £9.45 and cost per 100 sheets ranged from 12p to 43p.
All in all, the consumer is offered a complex mix of options. Add a number of promotional mechanisms to this mix and the range of choices becomes even more diverse. Of course, the larger quantities are implicit promotions – Tesco Luxury Soft White in 4 rolls is 20p per 100 sheets or 47.5p per roll (for a spend of £1.90), and in 24 rolls is 15p per sheet or 33.3p per roll (for a spend of £8.00). So, if the consumer can cover the out-of-pocket of the larger quantity then the saving is 14.2p per roll or £3.41.
Clearly in-store or online, shoppers do not spend the time making such calculations and will use general rules-of-thumb (mental algorithms) and biases. Example rules-of-thumb are:
- The more quantity one buys the lower the unit price (though not always)
- The lower the price the lower the quality (though not always) The use of the registered trademark symbol implies greater quality and higher cost
- Quilted is more expensive than standard 2 ply
- Tesco home brand will be cheaper than the market leader, and so on.
Given the complex problem the consumer has in determining the best buy for their needs, they may come to rely heavily on rules-of-thumb, intuitive responses, as well as buying habit.
The Split Second Research team recruited 500 respondents through CINT an online consumer survey panel in the UK. Respondents were put through stringent screening tests before qualifying for the main study. As the main study used implicit response priming, the qualifying test included testing their reaction times to a specific criterion that ensured responses would be reliable in the main test. Of the 500 who entered the survey, 390 qualified for the survey proper.
In the first phase of the main study, the brand logos of Tesco Luxury Soft, Tesco Basic, Andrex (premium), and Sofcell (basic) toilet tissue were presented in a rapid implicit reaction time test with five attributes. This enabled us to obtain brand profiles on the attributes, trusted, value for money, favourite, popular, and reliable. These attributes were used as these were the purchase drivers in this category discovered from previous research carried out by Split Second Research. A traditional survey questions was applied to determine respondent purchasing behaviour. This has also enabled us to classify respondents as either Tesco customers/loyals (n = 250), Tesco considerers (n = 50), Tesco lapsed (n = 30), or Tesco rejecters (n = 160).
Profiles of each product and for each of the four consumer groups are shown in Figure 1.
Some key observations from this analysis are: Tesco Luxury Soft has a very strong profile among Tesco customers who feel more favourable to this product than the other three on all attributes, except they feel just as strong towards Andrex on value for money and favourite. Andrex has a stronger profile for Tesco lapsed and Tesco rejecters, especially on trusted, favourite, and reliable. Remarkably, Tesco considerers feel more positive towards Tesco Luxury Soft than they do towards the other three products, but this is the least favoured product among Tesco rejecters. Tesco rejecters prefer Andrex, especially on value for money. An important learning here is that a product does not necessarily need to be less expensive in order to be perceived as value for money. Similarly, a basic product is not necessarily perceived as less reliable than a premium one. Respondents were then exposed to the same products under a price promotion equivalent for all four. The promotion was a price-off and was calculated as a percentage but shown as an amount discounted. We then measured implicit reactions to the products on offer using an identical implicit reaction time test as in the first phase. The aim was to understand how feelings to the brands would change now that they are on offer. Respondents were primed to thinking about the offers in this second phase, in the sense that they were invited to study the offers and to state explicitly which offers were most appealing before completing the implicit tests. The results are shown in Figure 2.
Tesco Luxury Soft under a price-off appear to be detrimental for Tesco loyals with association scores dropping from an average of 70 to an average of 50 (a fall of 28%). Yet, would appear very attractive for Tesco considerers and Tesco rejecters, with increased association averages of +40% and +77%, respectively. So the potential of using this promotion as a tool for penetration is clear. Although the Tesco basic product can drive penetration to Tesco rejecters (perception up by +14%), it would do so at the expense of reduced perception among Tesco loyals (-16%), Tesco lapsed (-21%), and Tesco considerers (-19%). Perhaps the greatest potential for penetration would be the discount offer on Andrex, which yielded increases in positive feelings for Tesco lapsed (+16%), Tesco considerers (+51%), and Tesco rejecters (+18%).
To summarise, for the current set of brands and in consideration of a single type of offer, there is little evidence that volume or frequency of purchase can be positively affected but very good evidence of potential for penetration for Tesco Luxury Soft, dependent upon effective communication of the promotion which needs to reach those considering a visit to store. Merging the luxury products and the basic products, we find that the basic products gain in perceived value for money but lose on trust, popularity, and reliability, while the luxury products gain in perceived value for money and as a favoured product but lose a little on reliability (see Figure 3).
Take-aways and further research
This research revealed that the use of indirect, implicit measures can be used to capture a range of feelings towards brands. It found that Tesco Luxury Soft generated the strongest positive feelings especially as a trusted product among the respondents tested by Split Second Research. It found that those who do not shop at Tesco (Tesco lapsed and Tesco rejecters) feel stronger towards Andrex than any of the other three products.
The research also showed that implicit response methods can measure feelings towards promotional offers. For example, for Tesco’s loyal customers, there was no evidence that a price discount on any of these four products would yield increases in purchase frequency or volume. However, for those considering shopping at Tesco, the Tesco Luxury Soft product has the potential to penetrate this group as their feelings towards this offer were especially strong. Similarly the offer on Andrex has substantial potential to drive penetration.
The overall pattern was that for basic/standard products, the discounted price decreased perception of the product (especially in terms of trust and reliability), whilst for luxury products the discounted price increased the perception of the product (especially in terms of perceived value for money).
This research demonstrates that understanding how shoppers will respond to a promotion is not immediately obvious. Split Second’s previous research has also shown that different types of offers can be perceived in completely different ways (Fulcher, 2018). There is therefore considerable scope for examining other types of offers (percentage discount versus £x price off, volume discounts, pick and mix promos, value added promos, loyalty offers, and so on), different types of products based on shelf life, or food versus health versus house and home, seasonal products versus everyday essentials, and store location, such as end of isle versus on-shelf versus near the check-out versus near the entrance. Split Second Research’s insight is that the type of promotion that will yield the best results in terms of frequency and volume of sales will be very different depending upon various combinations of the above.
Fulcher, E (2018). Neuro-offers: The effect of in-store offers on brand values. Published on LinkedIn August 6. https://www.linkedin.com/pulse/neuropricing-effect-in-store-offers-brand-values-dr-eamon-fulcher/
Husted, BW, Russo, MV, and Meza, CEB (2014). An exploratory study of environmental attitudes and the willingness to pay. Journal of Business, 67, 891-899.
Zoellner, F and Schaefers, T (2015). Do price promotions help or hurt premium-product brands? Journal of Advertising Research, 55, 270-283.
in-cosmetics formulation summit
In its fourth year, the in-cosmetics Formulation Summit is an education-focused convention, that captures the strongest trends driving the industry.
This year it included a number of leading international experts for the benefit of R&D professionals, senior formulators, and decision makers from cosmetics brands, manufacturers and ingredient suppliers.
Dr Eamon Fulcher talked about how the brain rapidly evaluates all that we see, hear, smell, touch, and taste. Faces are a special case, because a significant proportion of the brain is dedicated to face recognition, hence we can recognise faces with incredible accuracy, distinguish between subtle variations in facial expression, and evaluate them very quickly on an array of important attributes. One of these is attractiveness. Evolutionary psychologists have argued that our biological imperative is to protect and propagate our genes. We can do this through reproduction and by searching for a healthy mate. This is important because our genes are more likely to survive if we partner with a healthy mate. Beauty becomes a proxy for health. The use of make-up to try to make oneself look more beautiful has been going on for at least seven thousand years, and can be traced back to the Egyptians and the ancient Greeks.
More recently, psychologists have established that the more beautiful a person is (or appears to be) the more likely they are to be perceived more positively on an array of desirable attributes, such as emotionally secure, sociable, interesting, confident, organised, popular, and intelligent (to name just a few). This is termed the halo effect, and works even when we see the same person with and without make-up. Even though we know they are the same person (and we know that personal attributes are pretty much stable), we can’t help but attribute these positive features to them when they appear more attractive. Even babies have been shown to prefer more attractive faces. It’s not surprising then that most of us would like to be perceived as attractive by most or at least some people! These ideas can help explain why the cosmetics industry is so huge.
One problem for cosmetic scientists is that sooner or later they will need to do some consumer testing on their beauty products. Here they meet the truth gap – people don’t always tell you how they truly feel (they may want just to please you or they may want to hide how they feel), people can’t always say how they feel (they can’t easily put into words how they feel or their feelings may be in their subconscious), and they don’t always do what they say they are going to do (like signing up for the gym on the 1st of January and never going again!). If one’s market research is based only on what consumers verbally tell us (subjective reports), then we are not going to get a true picture of the likely success of our products.
One way to circumvent this problem is to use a more objective method, and one that does not rely on asking questions to obtain explicit responses. Implicit reaction time tests offer a valuable tool to accompany traditional quantitative surveys. Implicit reaction time tests can detect the associations that consumers have with brand names, product endorsers, cosmetics packaging, new formulations, fragrances, and more.
During the talk, Eamon demonstrated how implicit testing, coupled with neural network technology (a form of artificial intelligence or deep learning), can help model consumer perception of the brand or product (or any formula or proposition). Hence one can build a neural network model of the brand or product. A model that represents how consumers perceive and feel. This can yield insights that really help identify the strengths, weaknesses, opportunities, and threats for the stimulus being tested, and in relation to other formulations and those of competitors.
Split Second Team in Brazil
Geraldine, Thaigo, and Eamon taking in the views at the Shopper Brain Conference, Rio de Janeiro, Brazil, July 2018.
Eamon gave a talk on the use of implicit reaction time testing and pricing, especially consumers’ perception of offers and promotions.
Implicit reaction time tests can be used to understand the effect a promotional mechanism can have on both brand equity and intention to make a purchase. Split Second Research has now cnducted many of these kinds of studies and results are nearly always surprising. Although promotions and offers can increase short-term sales, they can have negative long-term effects on the brand, particularly by impacting brand equity. Our studies reveal that brand managers need to know which kind of promotion would work best for its brand because there are no general rules that apply within and between categories.
The talk focussed on pricing promotions and a case study on basic and luxury brands of toilet tissue. We identified Tesco’s best stratgey for pricing its two products in this category.
For more case studies visit Eamon’s post on Linkedin.
Taking time out for some fun 🙂
All American Family – Research results published
Split Second Research recently teamed up with Edelman Intelligence to conduct a study on what the All American Family means for different sectors of American society. The findings are published here:
Dr Eamon Fulcher
Gemma Calvert presented at the World Neuromarketing Forum 2018
Prof Gemma Calvert, co-founder of Split Second Research, gave an insightful and entertaining keynote talk at the World Neuromarketing Forum in Singapore, 2018.
Hosted by the NMSBA, the conference focussed on current thinking and new developments in neuromarketing, bringing together providers and users in a highly interactive forum.
Split Second Research demonstrated it’s integrated IMPRESS platform for the automation of implicit testing. This makes development and analysis happen in a split second! We also gave away 200 Split Second Research aprons with the slogan “KEEP CALM and dinner with be ready in a Split Second”, here modelled by NMSBA chair Caral Nagal: