BBC News

Market research and the primitive mind of the consumer

By Sharif Sakr
Technology of business reporter, BBC News

media captionMarket research enters the subconscious

"The trouble with market research is that people don't think how they feel, they don't say what they think and they don't do what they say."

So said the late advertising don David Ogilvy, and his words get to the heart of what is still one of the biggest challenges in business: How can you tell if people really like what you are trying to sell them?

Technology offers an answer to the question that Ogilvy, who died in 1999, probably never envisaged.

image captionProduct testers are great at telling you exactly what you want to hear

Market researchers are already experimenting with desktop and smartphone applications that promise to reveal the subconscious layers of a consumer's brain.

What's more, because this technology can work automatically and in real-time, it could potentially be used to evaluate the emotional responses of millions of people before any product is released.

And, with one small caveat, this power could transform market research and indeed the whole world of business forever.

(Oh yes, the "small caveat": to go along with this, you must be prepared to believe that computers can decipher the intricacies of human emotion.)

Market failure

Old-fashioned market research is straightforward and unemotional: you get a sample of product testers and ask them what they think of a particular concept, product or brand.

But surveys and focus groups assume that people know what is going on inside their own heads - and that is a risky assumption.

"80% of new products brought to market fail, largely due to failures in traditional techniques", says Rob Stevens, co-founder of UK market research company Bunnyfoot.

"I can't think of any other area of business where such a failure rate would be considered acceptable, yet somehow, in market research, it is."

Lie to me

Mr Stevens likes to describe his company as a real-life version of The Lightman Group - the fictional agency in the TV series Lie to me starring Tim Roth.

Like Roth's character, Bunnyfoot's staff are trained to spot clues in the facial expressions of product testers, which betray their inner feelings - a process known as "facial coding".

They also use eye-tracking technology to monitor exactly where a person looks during a product test.

Though relatively old and low-tech (facial coding goes back to Charles Darwin), these techniques are capable of yielding business insights.

image captionA product tester's facial expressions are filmed while they surf the Boden website

"When you ask a product tester if they spotted a particular feature on a webpage, they often tell you they did," says Mark Batty of online clothing retailer Boden.

"But when you look at the eye-tracking, you discover that they never saw it all."

Boden is in the middle of a usability study of its website, and as its e-commerce manager, Mark Batty has learned not to put much faith in testers' explicit responses.

"Often their verdict of the whole site depends on whatever task they did at the end of the test," he adds.

"If they enjoyed the final task they would be full of praise about the site, even if their facial expressions revealed that they had struggled with it at the beginning."

Look into my phone

The problem with manual techniques like facial coding is that they require a researcher to sit through hours of slow motion video, logging every mind-numbing frown and every humdrum movement of the pupils.

This means that studies are necessarily limited to a small sample of testers.

Boden's usability study, for example, had a sample of just 30 people across three countries.

image captionA product tester's responses to a TV advert are automatically decoded by ThirdSight's software

But this obstacle could soon be removed, by allowing computers to do most of the donkey work.

"Facial expressions can be read by a computer - it's just the movement of pixels in a piece of video," says Dr Roberto Valenti of the University of Amsterdam.

Dr Valenti and his colleague Dr Theo Gevers are so convinced of computerised emotion recognition that they set up a spin-off company called ThirdSight to cash in on it.

"A researcher gets tired, they need to be paid, they need to be trained."

image captionA smartphone monitors and decodes a person's facial expressions as either "positive" or "negative"

"But our software never gets tired and it can analyse thousands and thousands of faces at the same time."

ThirdSight's latest achievement is to get automatic facial coding software running on a smartphone, using the phone's inbuilt camera to record a product tester's expressions.

Machine learning

Of course, all of this hinges on the accuracy of the software.

And so far, Thirdsight's claims of accuracy are relatively humble.

They acknowledge that you still need a human researcher to oversee the software, because it is oblivious to context or hidden meanings - it will treat both a happy smile and a bewildered smile as "positive".

But other scientists are less conservative.

"We can normally tell between different emotions with pretty high accuracy," says Professor Peter Robinson of the Computer Laboratory at Cambridge University.

"Our computers can get an accuracy of two-thirds or better - which is about as well as most people can do it."

Prof Robinson's team is trying to free up emotion recognition software from simple rules, such as smile = happy.

Instead, they're programming it to digest many types of human expression - facial and eye movements, hand and body gestures, the tone of the voice.

image captionProfessor Peter Robinson says his emotion recognition software can be just as accurate as a human

Through a technique called "statistical machine learning", the software then trains itself to recognise which indicators are important and which are not.

If this kind of power and accuracy could be incorporated into a ThirdSight-style smartphone or internet app, then potentially millions of people's emotions could be accurately decoded.

"If you do an experiment on Facebook, you've got half a billion people in your sample," says Prof Robinson.

"That means the statistics become rather bizarre, actually, because you are sampling almost an entire population to get a result."

So, theoretically at least, the day may come when no product is doomed to flop - because businesses will have access to almost complete certainty about their market.