By Mark Ward
Technology Correspondent, BBC News website
Good gambling strategies can pay off for websites
Running an online shop can be a brutal business because the net is the ultimate level playing field.
The sheer number of competitors selling any and every product or service means shop sites can struggle to make themselves stand out in the crowd.
In a bid to keep up with the pace an increasing number of websites are turning to insights that have emerged from studies of how to gamble on fruit machines.
Running an online store and gambling on a one-armed bandit may seem to have little in common - beyond the near certainty that you are going to lose money - but there is an affinity that might not at first be apparent.
Research into the so called "bandit problem" has produced techniques that ensure theoretical gamblers get the best possible return.
These techniques funnel most cash into the machine giving the current best payout. But they hedge this bet by pumping some coins into other machines to see if they produce a higher return.
In the case of websites, visitors take the place of coins and the fruit machines are the different webpages that can be presented to them as they spend time on a site.
"Five to 10 years ago websites were seen as online catalogues more or less," said Peter Ahl, managing director of Serenata Flowers. "Now it's something living and breathing, an entity that responds to visitors."
Serenata is one of a growing band of online businesses that changes what visitors see as they navigate through the site. It means the site can be adapted to those different expectations and experiences.
Selling flowers via the net presents particular problems
"We look at the time they are visiting," said Mr Ahl. "We know there's a difference when people surf from office or at home over the weekend."
"There are other forces that drive differences in behaviour such as where they come from," he added. "If they came from a search page what did they search for? And if they know us they might be willing to spend more than if they are a first time visitor."
Paul Phillips, technology boss at Omniture which has built website-tuning software out of solutions to the bandit problem, said reacting almost instantly was a big change for many web shops.
Before now, he said, many firms analysed data gathered over weeks and months to get insights into how to target marketing campaigns to improve responses from customers.
"In the online world that's just not fast enough," he said. "A model based on last week or last month is no good."
The insights from work on the bandit problem help make the most of what customers want now, he said, but also explore other ways to entice them to spend more.
"It's not just based on history but it continuously appraises all they are doing and makes judgements about the risk of being wrong and the consequences of being wrong," he said.
Other behavioural monitoring goes beyond using insights into the bandit problem. Swedish firm Avail Intelligence aims to do a similar job by consulting the wisdom of the crowds that visit a site.
The web-tuning system created by Avail plunders information about what people are doing and how they make their buying decisions as they use a site.
It notices when someone has embarked on a particular path through a site and remakes a site to feed them relevant information, be it adverts or prompts to other webpages, to get that transaction finished.
The real-time systems could help machines become more intelligent
"It's not offline or post facto," said Rolf Elmer, head of Avail Intelligence. "It's about transferring directly the power of the purchase process. What is most relevant for that consumer."
Companies such as Barclays, BT, HSBC, Capital One, Lloyds TSB, MSN and many others have turned to behavioural monitoring systems, said Mr Phillips, because of the harsh realities of doing business online.
"Online is a very low friction, highly interactive environment," he said. "And it's natural for people to be in comparative shopping mode. All competitors are challenged because it's so easy to buy elsewhere."
The importance of this is underscored by the old business maxim that it costs far less to get an extra sale from an existing customer than it does to attract a new one. Because of this getting, keeping and getting more out of customers become hugely important.
This meant that any website that consistently gave people a marginally better experience than their rivals was likely to see a big boost to their business.
"As with any type of service small marginal gains are a very big deal and cause a large flip of business in your direction," said Mr Phillips.
Critics have pointed out before now that the web management systems of many websites have been so poor that almost any system would produce tangible improvements to sales.
But, said Mr Phillips, the real-time decision systems may one day find their place in a very different arena.
The push to develop such systems came from the AI community, he said, and eventually robots could use this approach to become more intelligent.
"This type of real-time learning technology is going to be very important for other domains. It's about the trade off between what you believe to be true while exploring what might be your risk," he said. "This is where machines start becoming generally intelligent."